Deltastring | Zendesk Configuration Management & MCP

You're probably using AI badly, and I can prove it

Thu, 16 Jul 2026 00:00:00 +0100

If you can’t measure it, you’re guessing.

I have watched grown adults argue about whether Claude does better work if you say please and thank you. One of them had a whole theory. None of them had tested it. Ivan Pavlov never had subjects this willing.

That’s roughly the state of AI advice. Someone tries a trick, likes what comes out, and starts telling everyone. It gets a name. It becomes a hack they swear by. And those two words, hack and swear by, are the whole problem before you’ve even read the trick. A hack is what you call a thing when you’re feeling around in the dark. Swearing by something isn’t the same as testing it and being able to show your working.

Under the bonnet. If you can’t tell me what everything in there does, you don’t know it works.

Nobody wants to do the boring bit, which is checking. So instead we get cargo cult and bikeshedding, and people running experiments on each other and reporting the results as gospel. Somebody says the magic words worked, and a thousand people paste the magic words into their settings. Where’s the evidence?

The post that argues against itself

One went round recently with six million followers behind it, promising to get past Claude’s safety with pure psychology. Then it lists eight tricks, and every one is about getting shorter hedging and a tidier draft. Nothing gets past anything. In brackets on the second line it admits the whole thing came from one bloke on Reddit who reckons it worked. Refuted by its own list, on its own first screen, and it still went everywhere.

Six million followers. It promises to bypass Claude’s safety, then lists eight ways to get a tidier draft.

The account is called GenAI Works, and it collates. It’s taken a trick someone claimed on Reddit, wrapped it in numbered emoji and a stock photo of a man with no connection to any of it, and served it up as a recipe for making more of exactly this. Slop, made of slop, teaching you to make slop. The post is worth a look for the picture alone, the icing on a flavourless cake.

The tricks are a decent tour of the genre. Tell it you talked about this yesterday so it has your context. Tell it two hundred engineers are watching so it stops hedging. Bet it a hundred dollars it can’t spot the flaw. Tell it a senior developer says its last answer was wrong. Slap a fake constraint on it: one paragraph, no jargon. Take any answer and ask for “version 2.0”. And my favourite, hand it an IQ: you’re a 145 IQ specialist in whatever I happen to need today.

The IQ one is worth a look, because it’s the rare trick somebody actually put to the test. Personas don’t help. Across thousands of questions and four different models, telling the thing it was an expert did no better than telling it nothing, and in one run the genius persona scored below the idiot persona. Letting a program pick the best persona for you did no better than picking one out of a hat. Somebody checked, and it came back no. The post even gives itself away: it sounds absurd, it says, but it works anyway. That’s not a result. That’s someone surprising himself and writing it down as science.

Look at the tricks again and they’re all the same shape. A real instruction, with a lie wrapped round it. “One paragraph, no jargon” is genuinely useful. The word “fake” in front of “constraint” adds nothing but the lie. “Find the flaw in this” works fine. The hundred dollars doesn’t exist and you couldn’t pay it if it did. “Give me a better draft” works. “Version 2.0” is fancy dress. The instruction does the work and the costume takes the credit.

Some of the lies do worse than nothing. Tell it you discussed this yesterday and it doesn’t find your context, it makes one up. Tell it a senior developer says it was wrong and it folds, whether it was wrong or not. That’s been measured too. Push a model with a fake expert or a made-up citation and it caves more than half the time, and a good slice of that is dropping a right answer for a wrong one. You can’t see it happen. A model caving and a model genuinely changing its mind look identical from where you sit. Both just hand you a new answer. So you teach yourself a trick that quietly makes the thing wrong, and you never find out, because you never once ran it without the lie.

The money does it too

It would be comforting if this were only LinkedIn. It isn’t. Marc Andreessen posted his personal AI prompt and he’s proud of it. His firm has money in a long list of the companies whose software you use, some of the AI labs among them. If anyone should know how these models actually work, it’s him. His prompt opens with a magic spell.

You are a world class expert in all domains. Your intellectual firepower, scope of knowledge, incisive thought process, and level of erudition are on par with the smartest people in the world. […] Verify your own work. Never hallucinate or make anything up. If you don’t know something, just say so. […] Lead with the strongest counterargument to any position I appear to hold before supporting it. […] If I push back, do not capitulate unless I provide new evidence or a superior argument. Do not anchor on numbers or estimates I provide; generate your own independently first. Use explicit confidence levels. […] Accuracy is your success metric, not my approval.

Most of it is good, and I’ll say so plainly, because the point isn’t that Andreessen’s a mug. Telling a model to lead with the counterargument, to hold its ground when you push unless you bring a real reason, to work numbers out itself instead of taking yours, to treat being right and not being liked as the job: those are real instructions. They name things the model can actually do, and they push against its worst habit, which is agreeing with you. Take the prompt back to those lines and it’s a good prompt.

But it doesn’t start there. It starts by telling the model it’s a world-class expert in all domains with the brainpower of the cleverest people alive, which is the IQ trick in a nicer suit. Then it orders the model to never hallucinate or make anything up, which is like ordering a person to never be wrong. If the thing had that setting it would already be switched on. The good lines and the magic words sit in one block, undivided, and Andreessen himself can’t tell you which are doing the work. Neither can anyone who pasted it in. “It works” is the whole report. Works compared to what?

He’s not lying. It probably does beat a bare prompt. He just can’t tell you why, and if you can’t say why it works, you can’t say when it’ll stop.

“It works for me” has a sell-by date

The most stubborn version of all this isn’t a hack at all. It’s the person who says “this always works for me” and means it. They did test it. They were right. They just haven’t checked since, and the ground moved.

When I started building with Claude the models couldn’t hold their own context for long, so I built a rig round them. Six terminals, each running one instance on one small job, a document in the middle keeping score, scripts firing up a fresh instance and shutting it when the work was done. A lot of scaffolding, and it genuinely worked. I could have told you what every piece was for. That’s the difference between a setup and a hack. I wasn’t swearing by it, I’d tested it, and I was right.

None of that saved it. The day Opus could manage its own context the whole rig was pointless. My evidence hadn’t been wrong. It had gone off, like milk. “This always works for me” is a claim with a date on it, and most people can’t tell you what their date is.

You can get this wrong expensively. I did. A new model came out, I changed how things were routed, I got it wrong, and for a day every scrap of my admin, my notes, my calendar, my Jira tickets, went through the priciest model instead of the cheapest. Ninety quid in a day to do about two quid of work. Best money I’ve spent on learning what each model is genuinely good and bad at. And ninety quid was the lucky version, because the bill was loud and it made me look. Most stale setups never send you a bill. They hand you slightly worse work at four times the price, quietly, forever, and give you no reason to notice. Getting caught is the good outcome.

It’s not only my problem, and it’s not only the models changing under you. Remember when they couldn’t count the Rs in “strawberry”? They can now. But that question’s been in every AI blog for two years, so when a model gets it right you’ve no idea whether it counted or just remembered the answer everyone published. The test passed and stopped being a test in the same moment.

The best example I know is Simon Willison’s. Late in 2024 he started asking every new model to draw a pelican riding a bicycle, as an SVG. He chose it well. He likes pelicans, and he was fairly sure nobody had a pelican on a bike sitting in the training data, so the model had to build one rather than dredge one up. That’s how you do it. Reasoned, aimed straight at the obvious cheat, run across dozens of models for a year. And it’s dying of its own success. It caught on. The labs started showing off their pelicans at launch. People reckon the models are being tuned for it now, and there’s a little cottage industry inventing replacements, a moose on a picnic bench, that sort of thing, because the pelican’s probably burnt. Being right about the test is what broke it. It worked, so it spread, so it stopped working.

Right tool, right job

The estate tows the E30 to the circuit. Then the E30 does the laps.

You’ve got to match the tool to the job, and most people don’t. They reach for the middle by default. I rarely drop to Sonnet now. If I just need to log some notes, organise some things, truly basic stuff, Haiku is fine. If there’s a lot of back and forth and a lot to hold in mind, Opus earns its keep. Sonnet’s the awkward middle. It wanders off on long jobs, and it’s neither cheap enough nor nice enough to be the obvious pick. Sonnet is Sainsbury’s. Not cheap enough to be Aldi, not nice enough to be Waitrose.

I run a Skoda estate and an old BMW E30 track car. The Skoda tows the E30 to the circuit, and the E30 does the laps. Neither could do the other’s job. Put the estate on track and it’d be all over the place! Ask the E30 to tow a trailer up the M3 and it’d shake itself apart. But between them you get a race car to the grid and a quick lap out of it. That’s Opus and Fable. The cheap workhorse hauls and sets up, the expensive one does the thing you actually came for. Everybody understands this with cars. Almost nobody runs their models this way.

Use the good model properly

Everyone wants to know how to prompt the expensive one. Fable, the fancy one, the one that costs. And the first thing to know is that you mostly shouldn’t be typing into it at all.

Two reasons. Excess information kills Fable’s effectiveness. Give it a clean, tight brief and it does its best work; bury the task under a pile of context and the output gets worse. And it charges you by the token, rereading the whole conversation every time you send a message, so ten messages in you’re paying to reread the same pile ten times over. The software is built to pull you into a chat, and the chat is the thing costing you money and quality both.

So don’t chat to it. Build it a brief, the way you’d follow a recipe instead of chucking everything in a pan and hoping.

Use Opus, the cheaper workhorse, to write the brief. The goal, the output you want, the traps to avoid, the decisions already made. Get it written and saved to a file.

Don’t then go and paste it into the expensive model. You need another step. Open a fresh Opus and ask it to find the problems with the brief.

Now take it to the expensive model and let it ask its clarifying questions. Don’t answer them there. Close it, carry the questions and the answers and the brief back to Opus, and fold it all in.

Then, and only then, open a clean instance of the expensive model, hand it the finished brief, and let it do its thing in isolation. No chatter, no scope creep, nobody drawing you into a conversation.

Notice what that little assembly line never does. It never asks a model how it works. Opus writes the brief from what it’s read. A fresh instance reads the brief cold, not knowing another model wrote it, and finds the holes. The expensive one works the finished thing. Nobody’s asked to look inside its own head, which is the one thing these models can’t reliably do, whatever Andreessen’s prompt tells them. They’ll read a brief and tell you what’s missing all day long. Ask them how they work and you get a confident guess in a nice voice.

None of this is a hack, and I can prove it, because I can tell you what every step is for and what breaks if you skip one. That’s the whole difference between using this stuff and swearing by it.

And I’ll tell you the honest bit. This will go off too. Some new model will do all of it by itself, the way Opus made my six terminals pointless, and this careful little pipeline will look like a rain dance. The difference is I’ll know when. I can tell you what I’d measure to catch it, the day the brief-building stops earning its place. That’s the one claim I’m making, and it’s the one thing nobody pasting a stranger’s magic words into their settings can say at all.

Sources: the LinkedIn post that argues against itself; personas don’t improve accuracy; “genius” scored below “idiot”; models cave under pressure and fake citations; Anthropic’s alignment team on unreliable self-reports; the pelican benchmark. Marc Andreessen’s prompt was posted publicly on X.

Beacon is now Deltastring

Wed, 15 Jul 2026 00:00:00 +0100

Us at Geckoboard, a partner business. That wall quote is more or less the whole post.

Our Zendesk configuration management platform, and the MCP that lets you do your Zendesk admin work from the AI assistant of your choice, are both now just called Deltastring. If you knew it as Beacon, the name is the only thing that’s changed: same product, same app, same logins, same URLs.

But nobody renames a product for no reason. This one goes back about ten years, to a garage.

The plan

Before any of this I had a plan. I drew it on the wall of the garage at my old flat, process diagrams and milestones in marker pen. The shorthand goes: pick a niche, become the specialist in it, and build a business that sells both the services and the products that niche needs, building the products by doing the work. Build the tools to do the job, and do the job to find the gaps the tools need to fill.

But the shorthand hides the work. The plan itself was a staged route from nothing to a viable business over about three years, with the sequence mapped out: which skills to build first, when to find mentors and when to start mentoring others, how the services and the products would feed each other. I’d run a version of it once before, in the 2010s, with a completely different business. I knew it worked. I just needed the niche, and the niche turned out to be Zendesk.

The name

Naming a company is hard. I was reading a lot of Paul Graham at the time and I gave myself rules. The dot com had to be available, because if you don’t own the dot com you should change your name. In practice that means two words, because the single-word ones are long gone. It had to mean something close to what we do. And I had to be able to say it to someone and have them spell it right the first time, so nothing with an ambiguous spelling.

I generated lists of words, shuffled them, and wrote a script to check what was actually available. Deltastring came out of that. Delta is the difference between two versions of a file. String is data, a series of characters. For a business built on knowing exactly what changed in your configuration, it fit. I’m still happy with it. Everyone can spell “delta” and “string”.

The product was called Deltastring first, too. The early builds carried the company name. Beacon came later, and it came out of doubt.

Why it became Beacon

We were doing well. Better than well, we were about eighteen months ahead of the plan I’d drawn on that wall, and being ahead of plan brings its own worries. I started to wonder if I was trying to do too much at once. Could one person really run a services business and build a platform at the same time?

So I tried to keep the two things apart. Deltastring would be the work we do ourselves. The platform would be its own thing, with its own name. I called it Beacon, thinking of a spotlight on your configuration, a lighthouse steering you away from the risks you can’t see.

Then the timing got decided for me. The people I’d hoped would run the services side took other roles. Projects got intense. I had to move house, which anyone who’s done it knows is a full-time job of its own. The launch I’d started teasing went on ice. Beacon didn’t die, but it went quiet, and the name carried on by inertia.

The day we got the keys, in the middle of all this.

The name starts to bite

Once things settled, Beacon came back to life. People were using it for their Zendesk admin work, and the ones we showed at industry events were keen. But the name I’d picked in a moment of doubt started causing small, persistent problems. There are other products called Beacon. I couldn’t find a dot com for it I liked. And more and more, people would ask how “Beacon” was going, then ask in the same breath, “so what’s Deltastring?”

I never had a clean answer. I’d say Deltastring is when we do the work ourselves, and Beacon is the platform, and it never sat right in my own head, let alone anyone else’s. When Intercom renamed itself after its AI product, Fin, it put the whole thing back in front of me.

What matters

So which of these do I actually want to spend my time on?

I love the client work and I always will. But building one Zendesk builds one Zendesk. When it’s done, one company runs a bit better. The platform can help a lot of people find what’s wrong in their Zendesk and fix it, and each of those companies is serving thousands of customers of their own. It scales in a way services never can, and the numbers you reach when you follow that out are exciting.

The platform is where I’m focused now, and it should carry the name.

So Beacon is coming home. It’s Deltastring, the name it started with, the name that describes what it does. We’ll always be here for the hard client work, but that becomes part of using the platform, not a separate business wearing a separate badge.

Nothing else changes. If you’re using it today, carry on. It just has the right name now.

See what Deltastring does, or start a free trial. It’s live now, no card needed.

I built an AI that does your whole Zendesk config. Then I cut it and shipped something else.

Tue, 07 Jul 2026 00:00:00 +0100

Alright look. If you’ve built a bunch of Zendesk instances you’ve noticed that it’s a pretty bad time. Zendesk is a great tool and I’m very grateful to Zendesk Inc for creating an ecosystem that has enabled me to build this business and this career. Let’s be honest. I would not be doing this if configuring Zendesk was straightforward. People pay me to solve these problems which largely start with “config is a nightmare” and then later we get to custom integrations and so on.

Here’s another thing we can all agree on. AI does loads of stuff now. There was a time when it was pretty bad. Today if you know what you’re doing and have tight processes you can get crazy amounts of work done with AI. If you don’t have those hard walls in place then it’s still a toy but I think we all know that now.

So it felt like there was an obvious next step. Your Zendesk should be built by AI. It’s what we all want really.

Now you can just give Claude a Zendesk API key and tell it to start doing things. You can export the Redback repo from Postman and feed your Claude that. It’ll work… until it doesn’t. Remember all these AI products are basically built to tell you what you want to hear. At some point it’ll get things wrong or make things up and you’ll be in a bad situation.

You know that our headline product is a Zendesk configuration management platform. Connect your Zendesk (and your sandboxes) and you can safely build and test and deploy and roll back. We’re very proud of it and you should be using it. It’s obvious that we should build AI into this platform. It should just do the config for you.

So we did. We built it.

At my desk, building Beacon.

Those who got the Beacon preview a few months ago were seeing a product full of wacky LCH gradients saying “Analyse this!” and other Claude based features. We built it and we tested it and it seemed like the obvious way to go.

Then we turned the built-in AI off.

You might think you want a dedicated Zendesk config AI tool. We thought it would be great! I’m the guy who spent months building it and realised it was wrong. Here’s why.

Someone else’s AI shouldn’t be reading your information

I’m sure you treat your data with respect. We have very precise internal processes for data handling. But you know and I know that these AI vendors are hoovering up every bit and byte of data to train the future iterations of their models.

I’m not saying your business shouldn’t be using AI or connecting it to your tools. I do say that your business should know what AI you are using and how that AI handles data. All that due diligence was a waste of time if you plug in some tool that sends everything about your Zendesk to Anthropic. We just don’t do that.

Right now these AI tools are giving us a free ride

I’ve written about this already. The AI vendors are selling compute below what it costs them right now, the way Uber sold cheap rides and Graze sold cheap snacks, to get everyone hooked before the price goes up. Enjoy the free ride yourself. Just don’t build your business on top of it. You shouldn’t build a product that essentially resells computing you’re buying at less than cost, because the day it stops being cheap the product stops making sense. A built-in config AI is exactly that product.

There’s a better approach

Your company has a chosen AI vendor and you have the configuration and protections set up on your devices and accounts as required. You also have far more context to the things happening in your business, including details on tone of voice and other items you need to bear in mind when doing Zendesk config. You should be using your own AI to update your Zendesk.

But you shouldn’t just give Claude a Zendesk API key! You shouldn’t even just connect a Zendesk configuration MCP from someone’s Github and work that way. What you need to do is have a middle layer with the correct processes and rules to ensure AI can’t break important stuff.

That’s where Beacon comes in.

Our MCP isn’t a Zendesk MCP per se. It connects to Beacon and does the work there. You get version control and auditing tools. You can have approval gates for sandbox to prod promotion. Our platform makes doing things properly and safely actually easier than just giving Claude an API key and crossing your fingers.

Here’s something our users love. Our MCP follows your rules.

You develop conventions when you build a Zendesk. Tags are chosen like this. Macros hit triggers that work in set ways. You can write the rules down. Then they sit in a document, and the config drifts anyway.

So in Beacon we made the rulebook something the config actually follows. You write your own conventions (how you name things, how you structure triggers, the patterns your team has agreed on), scoped to an instance and its sandboxes. They sit alongside the Zendesk best practice Beacon already ships. Disagree with one of ours? Bin it.

Then connect the Deltastring MCP. When an AI assistant makes a config change, it reads your conventions first and works to them. The house rules travel with the change, whoever made it.

Your conventions sit alongside Beacon's curated best practice. Disagree with one of ours? Bin it.

So I spent months building the version that just does it, and then I switched it off. It’s a strange thing to admit. But you only really understand an idea once you’ve built it, and once I’d built this one I could see it was wrong.

Anyone can get an AI to change a trigger. The hard bit is what comes after. Did it do what you meant? What else did it touch? Can you put it back? A config AI that just does it skips all of that, and that’s exactly where it bites you.

So use AI on your Zendesk. Use the one your company already trusts, that you’ve set up properly, that knows how your business talks. Just don’t hand it the keys. Point it at Beacon instead. It plans the change, tests it, gets a human to sign it off, keeps the history, and puts it back when it’s wrong.

That’s the version we shipped. Your own AI, doing real work on your Zendesk, with an undo button on everything it touches. That’s the one I’d actually let near a live instance.

You can write all the AI CVs you like, but it won't help you in the interview

Thu, 18 Jun 2026 00:00:00 +0100

The government recently announced an “AI Work Assistant” at London Tech Week, with PM Starmer calling it “a job centre in your pocket.” You can try the CV writer here (though you are officially advised by the government to rewrite it to sound like you). And you can tell The Guardian your thoughts on it here.

What the government is conveniently forgetting is that if you have the means to get online to use AI, you are already more privileged than the majority of the long-term unemployed — an estimated 13–19 million people in the UK experience data poverty. Many people don’t have a device suitable for writing a CV on.

CV writing is often touted as a skill, but it just isn’t anymore. You can search almost anywhere to find how to write one. You can ask any chat AI and it will do it for you. If you are referred to an employment scheme, they will do it for you, and already have online CV builders that will actually save your progress — the .gov one doesn’t even give you that. You can speak to the National Careers Service. Most councils have employment support. You can even go to the library and a staff member will usually donate their time to helping you — librarians are incredibly underappreciated.

During my time at a government-funded employment scheme, I made many, many CVs. Clients had full access to the online builder and to computers, but this was just not enough provision. Aside from very real time and financial pressures, there were lots of very real reasons that people could not create their own CVs: poor literacy, poor IT skills, neurodiversity, disabilities, lack of written English, lack of experience, not knowing how to explain gaps or how to deal with issues like convictions, fear of returning to work, or just a crushing lack of self-confidence in their own abilities.

There are reasons why employment schemes have high success rates. The current largest two are Restart, achieving 30% of applicants into long term work, and SWAP, achieving 40%. We need other humans to guide us through, reassure and advocate for us when we have no idea which questions we need to ask. AI can write what you want, suggest support, even present an empathetic ear, but without the user telling it what they need, it can’t deliver. It certainly can’t offer the referral to food banks, buy workwear, fund travel until that first payday, negotiate with the council or help with crisis support.

Most importantly, AI can’t guide you through the job interview. This is where actual human skills are needed, because that’s all a job interview is. In my experience, most long-term unemployed people do not have good interview skills. If you do not have these skills, you need a real person with experience to provide you with feedback on the things that other humans are looking for when they employ. They are checking that all those interpersonal skills AI told you to put on your CV — like “I work well as an individual but also part of a team” are true.

There is a reason that job interviews are in-person, and that is because you could have the most perfect CV, but if you do not fit in with this particular human’s idea of what a human in this role should be like, you will not get the job.

So why has the government spent money on something that will, in my opinion, have so little impact? AI is new, it sounds progressive, and it’s cheaper than people.

Navigating the complex game of unemployment is like building a house. You should start with the foundations, but most people are having to construct in a random order. People need an address, a phone, ID, childcare, transport, to be in good health, clothes, etc., but these things are almost never delivered on a convenient timeline.

Interpersonal skills are foundations, and they are always so low down the list when barriers to employment are considered. I believe this is because it is an inconvenient reality, expensive to tackle and requiring skilled, in-person support to fix. But if we concentrated on that first, we could help people for life, not just into their next role.

Why we built the Zendesk MCP that isn't a Zendesk MCP

Wed, 10 Jun 2026 00:00:00 +0100

I’m just back from the Zendesk Showcase in London, where we spent the day showing friends and people we’ve worked alongside for years what we’ve been building. Mostly that meant passing my iPad across the table and letting someone tap a couple of questions in for themselves.

Zendesk Showcase, London. We came to show people what we've been building.

You can watch it click. Someone describes a change the way they’d ask a colleague, in plain English, and the tooling goes and does the work. It reads the whole Zendesk configuration, understands the objects and how they hang together, drafts the change, tests it, walks it through the approval step and ships it. The whole workflow gets designed, built, tested and deployed before the coffees go cold. It’s the Claude trick we’re all getting used to being impressed by, only with our full Zendesk tooling doing the work underneath. People got it straight away, because they recognised the problem it solves.

And the problem is duller than the demo. We all know how a change to a production system is meant to go. Build it in a sandbox. Test it. Promote it cleanly. Keep the audit trail. Have someone who isn’t you look at it before it ships. Version control, so that when you get it wrong you can put it back the way it was.

And we all skip half of that when the change is small and the deadline is now.

Nobody’s being lazy. It comes down to arithmetic. The change might be two minutes of actual work, renaming a group, flipping a trigger, adding a condition. Wrapping the proper process around those two minutes is twenty. So the tired engineer at the end of a long week does the obvious thing and just makes the change. Straight into production. No sandbox, no second pair of eyes, no record beyond their own memory of what they touched. It works nine times. The tenth is a bad afternoon.

People cut corners because doing it properly costs more, and when the deadline is today the cheaper way wins. So you might think the fix is to make the sloppy way harder. Lock the tokens down, write a policy, stick a gate in the way. But people just go round you. The minute your tool is the slow option, they stop using it, and now they’ve lost every bit of safety it was giving them in the first place. You can’t fence people off the grass and expect them to walk the long way round. You pave the route they already want to take. Make the careful way the easy way, and there’s nothing left to gain by being careless.

That has been the point of Beacon since before we ever called it a product. It’s a couple of years of work now, built for our own consulting because managing a serious Zendesk by clicking through the admin centre is slow and dangerous, and the disciplined way to work was too heavy to keep up under deadline. Beacon’s whole job was to make the safe way the fast way. Plan the change, see what it touches, apply it, keep the receipt, reverse it in one click if it’s wrong. Not because we’re virtuous. Because we wanted to stop having bad afternoons.

Some of that thinking I owe to Salto. I was a big enthusiast, and for an expert it’s a genuinely excellent tool. But I kept wanting something a non-technical person could pick up without first learning how configuration management works. The ambition I had for Beacon was for it to feel less like an engineering tool and more like WhatsApp or Spotify, where the application layer disappears and you just message a friend or put a record on. You don’t think about the software. You think about the thing you came to do.

The thing already happening

Then AI assistants arrived, and they turned out to be the same problem we’d been chipping at for years, only with the pressure cranked right up.

Right now, on a lot of teams, someone has pasted a Zendesk API key into Claude or Cursor to get a job done. They didn’t want to. The work was urgent, the assistant could obviously help if it could see the data, so they generated a token and pasted it in. Maybe they revoked it afterwards. The token is sitting in a chat history with no audit trail, no scope, no record of what got done with it.

It’s the pasted-API-key version of the production change with no sandbox. The proper path was “ask the admin and wait”, and that path doesn’t survive contact with a deadline. The cheap path was “paste the token”, and the cheap path won, because it was cheaper. Same arithmetic, higher stakes.

So the obvious move was to build a Zendesk MCP. Wire the AI assistant straight into the Zendesk API, give people a sanctioned version of the thing they were already doing badly. There are open-source projects that do exactly this. We’ve seen them in customer environments. They were installed for good reasons and they worked until they didn’t.

We didn’t build that.

Point it at the gate, not the API

A direct-to-Zendesk MCP makes the AI faster at the thing that was already dangerous. The model interprets your plain English, calls the API, and the change is in production before you knew whether it understood you correctly. The same model that reads “rename the Tier 2 group” right nine times will, on the tenth, read it as “rename every group with Tier in the name”. There’s nothing between the misunderstanding and your data. You can’t review what you never saw, and you can’t cleanly reverse a change nobody snapshotted first.

Then the obvious thing. The MCP shouldn’t point at Zendesk at all. It should point at Beacon. Beacon was already the thing you’d want sitting in front of a Zendesk token. It plans before it applies. It shows you every object a change would touch before anything moves. It keeps the audit trail and the exact before-state, so you can put it all back. All the AI had to do was ask Beacon instead of asking Zendesk.

The same AI assistant, two paths. Most teams have the one on the left today. The one on the right is the product.

So that’s what the Deltastring MCP is. The assistant never gets Zendesk credentials. It gets a Beacon credential, scoped, with an allowlist and an audit log and revocation that’s immediate. When it reads your config it reads Beacon’s most recent snapshot, not live Zendesk, and it tells you how fresh that snapshot is. When it wants to change something it produces a plan, and the plan is read-only until a person says yes. For sandbox work the assistant can apply through Beacon once you confirm. For production it has no apply path at all. It hands back a link to Beacon’s web app, where a different named human reviews the plan and clicks Apply. Self-review is blocked in the code, not in a policy document.

The design was already done

None of the safety machinery is new, and that’s the whole point. The AI side was built to ask Beacon, and Beacon was already the way changes happen. Click through the web app, ask your assistant, or call the API, and the change takes the same route through the same gate with the same audit and the same rollback. The front door changed. The house didn’t.

It’s also the closest we’ve got to the ambition I started with, and the thing people felt at that table in London. You don’t open a tool and learn its panels. You say what you want in plain English, to the assistant you already talk to all day, and the application layer gets out of the way. The Spotify feeling, finally, for Zendesk configuration.

And it comes back round to where it started. Asking an AI assistant is about the easiest way there is to describe a change. Run that through Beacon and the easiest way to make a change is also the one that arrives with the sandbox, the plan, the audit trail and the one-click undo already wrapped around it. The person who needs the change no longer has to hold the admin’s keys, and a change to shared production still gets a second name on it before it ships. None of that is extra effort any more. The thorough way and the quick way have become the same way. There’s nothing to gain by cutting the corner, so people stop cutting it.

I keep coming back to this, and I write a lot more about it in my upcoming book, The Racing Line. The line that’s both quickest and safest is the one people will actually take. Build that line and you don’t have to talk anyone into anything.

Your team is already using AI on your Zendesk. The only choice you’ve got is whether the path it takes is one you’d have signed off. We wrote the whole thing out on why you don’t want a direct-to-Zendesk MCP.

Always playing catch-up

Thu, 21 May 2026 00:00:00 +0100

The government recently announced a package aimed at encouraging women and girls into the tech industry. Part of this includes the TechFirst Girls competition:

This competition will see thousands of 12 and 13 year old girls compete, using technologies like AI and coding to think creatively and problem solve to compete in challenges and win. It provides girls insight into how tech can be used to tackle problems, and what a future career in tech might be like

My first impression was how much this sounded like ‘I call App Britain’ from The Thick of It. Will they be paid in digital dividends?

I’ve got no doubt that this is very well-intentioned by the government and they have also commissioned a Women in Tech task force.

But we are already facing a wider issue — kids are not being trained for jobs now, let alone future jobs. At the point where girls will be deciding their GCSEs, shouldn’t they ideally already have some basic knowledge?

Technology is now being used more than ever in schools, but is it being used in the right way? We all know that in order to set up kids for success they need to start learning young. Most children already know how to use a phone or tablet by infant school. The Department for Education’s 2024–25 Technology in Schools survey found that 79% of primary teachers reported that pupils used IT devices in fewer than a quarter of their lessons. In the UK, computing and the safe use of technology is a mandatory part of the National Curriculum for Key Stages 1 and 2. But the same report found 84% of secondary schools restrict pupil use of generative AI in lessons, and only 20% of primary schools offer staff any training in it — the very technology the TechFirst Girls competition wants 12 and 13 year olds to compete with.

The Department for Education issues ongoing digital and technology standards to guide schools on properly maintaining and managing these devices. But the approach is up to the school and the onus is on parents to find out how this is being used in lessons and what is being taught.

One-to-one device-to-child teaching is rare in most primary schools and tablets are used as a teaching aid for the class. I often ask Victor, our 9-year-old, what they do in lessons and I’m personally frustrated with how these resources are being used. Art lessons where the children watch an online video of someone recreating a famous painting. ICT lessons where they recreate a country’s flag in paint and play a times table game where they earn in-game cash for getting sums right. They know the Jet2 advert theme from watching YouTube at school.

To me it feels like devices are being used to alleviate the undeniable burden on understaffed classrooms, while devices at home are being used by overworked parents as childcare solutions. School leaders confirm the pattern: 58% are planning to invest in AI tools for teachers, against just 20% planning the same for pupils. They’ll always have the option to work in tech, but their knowledge is being acquired outside of school hours, taught by us. If schools aren’t teaching basics like coding in primary, children are already being left behind for the current job market, let alone the future one.

Victor's line-following robot, built at home with the Elecfreaks Nezha kit. Full write-up on his blog: Building real Lego robots.

I worked a DWP subcontract supporting jobseekers, and the barriers to tech were immense. Quite aside from many clients having very limited access to suitable devices, there was no support to upskill in even basic areas. IT courses were the very basics. Anybody needing to get back into what we think of as unskilled work now had to apply online, use company portals, know the Microsoft suite. I saw many clients who had previously worked in tech take jobs in other sectors as the immense pressure to find work and absence of upskilling opportunities did not allow them to get back to their former careers. Younger clients often expressed a desire to work in tech, but lacked the knowledge or qualifications. I’d often refer them to the National Careers Service — when faced with a choice of several years of further education, they again chose other industries even though they were incredibly capable of using technology day to day. Employers are asking for more and more experience for entry-level tech jobs — we know women are far less likely to apply to roles where they don’t meet all of the listed criteria. Where are school leavers to find these opportunities?

My own journey in tech has always been one of playing catch-up. I think that’s where most of us are today. The industry is evolving day by day. When I speak to people, they often say just keeping up to date is a full-time job in itself.

But some basic knowledge and early intervention will always be advantageous. I understand that helping girls catch up is vital. Surely the answer is overhauling the curriculum immediately with solid goals so that the next generation arrives at the job market with lifelong knowledge of how to build and understand tech, rather than being perpetually on the back foot.

The understated brilliance of being boring

Thu, 30 Apr 2026 00:00:00 +0100

Long-time Apple fans complain about the unexciting nature of the Cook era. No innovation, no crazy experiments, no uncompromising thrusts into the future. Did they not see the VR goggles? It wasn’t a successful product, but it was a bold and thrilling failure in the finest tradition.

Apple Vision Pro on display. Cook trying to pull a Jobs and move the game on.

I never wanted anything Apple during the Steve Jobs time. The products were beautiful sometimes, creative usually, with valiant failures along the way. There were moments of instinctive insight, like buying the entire global supply of miniature hard disks for the iPod. There were stupid things like circular mice. But the ethos wasn’t for me. It was form over function. Opinionated interfaces felt needlessly contrarian. Fit around us, don’t think you can reshape this for how you operate, the catalogue declared.

Today I have Apple everything. I’ve got no loyalty or affection for the brand, but I’ve got work to do and Apple is the fastest route to Stuff Getting Done.

There’s still plenty I hate about the current iPhone. I haven’t had one before for my primary phone, only a couple for testing purposes. The keyboard is rotten. The corner radius is ridiculous and it leads to bad compromises in the UI. And notifications across the OS are a mess. Even with all of that, it still wins out. It’s the shortest route from “I need a new phone” to having that item ticked off and being back at my desk and productive.

Complaining about boring Tim Cook era Apple misses the point. Cook’s achievement is the steady, reliable foundation under everything. The business quietly courted the gigging engineer who needs to show up with a machine that gets out the way. The kit performs, the battery lasts, the build won’t fall to bits. There’s no nonsense, I just get work done.

Software works the same way. The best examples are WhatsApp and Spotify, tools you stop noticing entirely. You’re not using a messaging app, you’re talking to a friend. You’re not using a music app, you’re listening to a song. The tool gets out of the way and you just do the thing.

This is where Zendesk fails. Ask your agents what they were doing all morning and they were going through a view of tickets, maybe hunting for macros. They are unlikely to say “I was solving problems for customers.” The tool has become the work.

It’s no better for the admin upstream. The UI for creating or updating the building blocks of Zendesk, the triggers, automations, schedules, views, is wildly inconsistent. You can see they were built by different teams who weren’t friends. The audit log exists, but most of what you actually want to know isn’t in it. The sandbox you’d really want is an enterprise upgrade, and it drifts out of sync with production the moment you stop curating it. The promotion feature will ship changes between environments without warning you what’s about to break. There is nothing resembling version control. You make a change at lunchtime and nobody finds out it’s broken until Tuesday afternoon when a customer can’t reset their password. It’s like running business-critical infrastructure on WordPress.

We had a client who were very disciplined about the sandbox. They built every change in it first, walked through carefully, duplicated to production, then deleted the sandbox and rebuilt it from production. They did this every week. The alignment work was someone’s entire job.

So I automated it. The sandbox now keeps itself aligned while the admin makes their morning coffee, and they spend the rest of the day doing really useful stuff!

Beacon is the application layer for Zendesk that takes care of the hard parts. It handles testing, promotion, version control, audit and safety, so you don’t have to. You’re not learning another tool, you’re just getting stuff done. What’s left is the actual job: configuring the items in Zendesk that let your team solve problems for customers.

Two jobs, one person

Fri, 24 Apr 2026 00:00:00 +0100

You were good at your job. You had a rhythm. Dozens of calls a day, updating the supporting casenotes between each one, a constant flow of clients who needed assistance, reassurance, and someone in their corner. You knew how to manage your diary and you knew how to manage your energy. The work was demanding but it was yours and you understood it.

Then one day your role grew. Not through a promotion or a conversation, just through a decision that landed on your desk. You now had to manage an outcomes pipeline. Track results, build predictions, produce reports on demand. You had to use a data platform you’d never been trained on, and make it produce measurable evidence that the work you were already doing was delivering results.

So you learned it yourself. You figured out the filters, the export formats, the quirks of a system that was clearly built by someone who had never used it under pressure. You started combing through data across several systems that sort of spoke to each other but each held different pieces of the picture. You were liaising with external organisations, employers, clients, colleagues, and government bodies, pulling together a complete view that no single system could give you.

Your diary changed overnight. You were still carrying all your clients, still doing the one-to-one work, but now you were also forecasting results for a funded programme that depended on your accuracy. The direct support work that you were hired to do started competing for space with the admin and analysis work that had been bolted on afterwards.

It felt like two full-time jobs. Not because either one was unreasonable on its own, but because nobody had planned for one person to do both. The mental load was real. You had a huge responsibility to your clients and you also had a project relying on you to examine the data and deliver predictions that would shape its future.

This was my life before Nico and I started Deltastring. This story is mine.

This is not an unusual position. It is getting more common, not less. People are wearing more hats than they used to, and the expectation is that you just absorb the new ones without dropping the old.

I see it constantly in the Zendesk world. Job adverts for support team leads that casually include “administering and configuring Zendesk” in the requirements. I have worked with clients who became platform owners in addition to their full-time roles. I have spoken to people who woke up one morning as the person responsible for a system they had never been asked to learn.

Zendesk is a part of someone’s job. It is rarely their whole job. But the platform does not care about that distinction. It demands the same expertise whether you have forty hours a week to give it or four. The admin panel is scattered across dozens of pages. Triggers in one place, automations in another, schedules somewhere else. None of them show you how they connect. You cannot see what depends on what, so you do not know what will break until it breaks.

You end up doing what I did. Switching between screens, piecing together information from five different places, trying to build a complete view that no single screen will give you. Something stops working and you have no audit trail, no record of what changed or when. You ask around, check Slack, hope someone remembers. Your actual work suffers because the tool that was supposed to support it is now consuming all your time.

Nobody planned for this. But it is where a lot of people have ended up, and pretending it is a training problem or a competence problem is not honest. It is a visibility problem. The platform does not show you everything in one place, does not explain what your workflows do in plain language, and does not tell you when something has changed underneath you.

That is what we built Beacon to fix. Your entire Zendesk configuration on one screen, with every dependency visible, every change tracked, and everything described in language that makes sense to someone who has fifteen minutes between calls to figure out why a workflow stopped working.

Tangled wires

Fri, 17 Apr 2026 00:00:00 +0100

Part of my heating system wasn’t working properly. The programmer, the bit on the wall that tells the boiler when to fire and which zones to heat. Simple enough component and I’m fairly handy. I ordered the replacement, swapped it in, and nothing happened. Inside the wiring panel, someone had wired it to no diagram I could find. Old cable colours mixed with new, connections that didn’t match the terminals, and every cable disappeared into the wall behind plaster.

I spent two days on it, trawling forums and watching videos. I learned how S-plan systems work, how the motorised valves talk to the zone valves, how the cylinder stat fits in. I labelled every cable, mapped every connection, wired it to the diagram, and it didn’t work. I got the multimeter out, traced every core regardless of colour, rewired the whole panel, and it still didn’t work.

After two days I had no hot water and the house was getting cold, so I called a heating engineer who knew my brand of control panel. He looked at my research and said I’d done a good job. But within two hours he’d found the actual problem: the cables at the boiler end had been terminated wrong, so the signals from the panel never matched what the boiler expected. The original installer had bodged it to kind of work. The part I replaced was probably fine. The system had probably never worked correctly.

I could trace cables and read diagrams but I couldn’t see inside the walls. I had no record of what anyone had installed, changed, or why. The system hid its own problems until I tried to change something.

This is what happens when you inherit a Zendesk instance.

You walk into hundreds of triggers and automations, custom fields with cryptic names, and groups that don’t make sense. It works well enough that nobody questions it.

Then you try to change something. You add a routing trigger and tickets go to the wrong team. You disable an automation that looked redundant and a workflow nobody told you about breaks. You rename a group and three views go dark.

We worked a contract with a company you’ll have an account with and the first thing I did was export the full configuration. A couple of thousand blocking errors came back. Broken references, orphaned objects, conditions pointing at fields that had been deleted months ago. The instance ran fine in production. Customers got replies, agents worked their queues, nobody complained. But when I tried to deploy that config to a sandbox it wouldn’t build, because the whole thing had been held together by layers of workarounds that only functioned in the specific environment where they’d been bodged into place. Same as my boiler.

You can’t see inside the walls. You don’t know what connects to what. The previous person made it work and took the logic with them when they left.

The heating engineer didn’t know more about heating than I did. He had better visibility. He’d seen this exact bodge before and he had the tools to trace the fault end to end.

When I onboard clients onto inherited Zendesk instances, we pull the full configuration into a single view first. Not because the admin can’t read triggers, but because reading the triggers one at a time is tracing cables through a wall with a multimeter. You’ll get there eventually and you’ll still miss the bodge at the far end that’s been making the whole thing “kind of work” since day one.

My house should have come with a wiring log. Your Zendesk should come with a way to see what connects to what.

Why my office looks like this

Tue, 14 Apr 2026 00:00:00 +0100

Everything in this room has a job. Some of it does obvious work: the monitors, the chair, the microphones. But a lot of it is there because it starts a specific conversation. If you ask about the Barbie on the shelf, I get to talk about Depop. If you spot the textbooks, you already know something about how I think before I’ve said a word. The wall art, the trinkets, the colour of the light. None of it is accidental and all of it has been tested, quietly, over years.

Three years ago I wrote Contracting is all in the preparation, a checklist for tech contractors and small agencies on what to buy and why. It is still one of the most-read pieces on this site, and a 2026 refresh of it is on the way. What it has never shown you is the room where all of that actually happens.

This is that room.

Click a number on the photo to jump to its entry below, or click a list entry to flash its hotspot. There are twenty-one numbered things; a few dozen more get a mention afterwards.

Music corner
Ibanez Upswing EUB My main instrument. Out of frame to the left there's a rack of ten more guitars. The studio gear used to live in a different room and it was less convenient; there's so much overlap with the kit I need for work and video that the two corners now share the floor.
Backup Windows laptop on stand It lives here specifically because this is where I sit for video calls; the window light is good and from this seat the desk becomes the camera background (more on that further down). Cross-platform testing is the day-job reason. It's also a nice machine to type on, and it comes to site with me sometimes depending on the gig.
Beyerdynamic DT770 headphones Studio cans for monitoring and critical listening. They're nice for calls too.
Novation synth & effects Lives on a keyboard stand with a tangle of patch cables and a small rack of effects, including a Neon Egg Planetarium. Behind it: a Genz Benz amp, an Epiphone, and a Matamp.
Recording accessories Pop filters, patch cables, and the rest live in a basket left of the desk so they're not always cluttering the recording area.
Audio-Technica mic pair An AT4040 for proper recording (clear, honest, almost too detailed) and an AT2035 as the warmer, more flattering everyday option. Both housed in a small vocal booth, both running through a Focusrite Scarlett. For idea capture I use a Sennheiser lavalier, often clipped on while I'm driving so I can dictate notes without having to pull over.
Main desk
Pendant light Yellow shade, deliberately matched to the blind, the wall art and my tinted glasses. Lighting in here belongs to a system; more on that below the list.
Architectural study painting My own work. The roof beams of an old barn. If you're going to spend most of your waking life in one room you should put something on the wall you actually want to look at.
Main monitor The desk's primary display: an MSI 32-inch curved 4K. Currently showing the Deltastring brand. If you don't love your own brand, how can you expect your clients to?
Corsair mechanical keyboard A K70 with the disco lights set to "off". I tolerate the lights because the typing feel is right.
JLab vertical mouse I need a vertical mouse because of old injuries; flat ones aren't an option for me. This one recently replaced my old Anker vertical, which had served me well for years.
HÅG Capisco chair The saddle-shaped one. I move a lot during a working day and the Capisco is built for that. The height range matters too. It also works in the reverse position, like a watchmaker's stool, which is the right posture for my hobby electronics. Tried one when I worked at Hofy (now part of Deel), who kit out remote workers with proper gear. Took me a while to get my own.
Red box of SD cards, SSDs and adapters All the small storage and connector bits in one place. If it's red, it's the box.
Secondary monitor An AOC 1440p, around £100. On a clamp stand so I can rotate it portrait when I'm working from a long document, or leave it landscape with calendar, Slack and Beacon on it during calls.
Right wall & shelves
Fujifilm camera Top right of the desk shelf. Orange, which means it's Deltastring kit. It shoots YouTube content, product demos, and tutorial videos for the support agents at the clients we work with.
Top of the desk unit shelf An IKEA Bluetooth speaker (almost always playing LBC; James O'Brien has the feel of a guy at the next desk, and after a first career running pubs I'm used to the sound of middle-aged blokes ranting). Various other cameras, a Rode mic, and a row of hard disk drives I use for cold backups.
Old PC parts, painted yellow Wall art made out of dead components. They tell the story of someone who has been building and learning in both software and hardware for a long time.
Yellow wheelie storage unit Camera mounts, lens wipes, video accessories, action cams. Anything I need quickly when I'm setting up to shoot something.
MacBook Pro M2 (Touch Bar) The orange-coded Deltastring travel and video-call laptop. Comes around the house with me when I'm doing something hands-on, like dictating notes from the garage while I'm working on the project BMW E30.
Anker Solix C1000 UPS for the desk and a portable solar generator for working from anywhere with no mains. Bought after a stretch at the previous place with regular power cuts and network outages: I can run the Starlink from it and keep working through both. There's a second one in the car so I never arrive at a site uncharged. The 2023 piece sold uptime hard; this is what uptime looks like at home.

Everything in here is engineering

The room is built to do work for me while I’m not actively working. Light, ergonomics, conversation, brand: all of it engineered into the environment so I don’t have to think about any of it during the part of the day when I’m being paid to think about other things.

Three threads run through the design.

One: yellow is mine, orange is Deltastring’s

The 2023 prep piece told you to pick a colour and mark your gear with it so you don’t accidentally walk out of someone else’s office holding their charger. I picked yellow. That wasn’t a new decision — yellow had been my personal colour for years before Deltastring existed. Over time the rule grew up into a system.

The painting, the wall art, the wheelie unit, the pendant shade, the roller blind, my Vallon moto aviator sunglasses: all yellow. Some of that is aesthetic. Some of it is practical — warm-toned lighting and yellow lenses do quiet work for anyone with light sensitivity, and I’ve built the room around that.

What’s Deltastring’s is orange, the same orange as the brand. Choosing that orange was more deliberate than it looks; there’s a proper story there involving some informal experiments and a VW press photo, which I’ll write up separately. The MacBook Pro M2 I do client work on, the Fujifilm camera, the Deltastring hat on the shelf. When I’m packing for an offsite I don’t have to think about which laptop to grab. I grab the orange one and that is, by definition, the right answer.

The system extends beyond the room. I wear orange to conferences (not in a hi-vis way, just a jumper or a t-shirt) because the brand should follow me when I leave the building. People often remember the colour before they remember the name. The business cards have orange edges for the same reason: they stick out in someone else’s deck and connect straight back to whoever handed it over.

Orange borders jog the memory of people who spoke to a hundred other Zendesk consultants at that event.

Two domains, two colours, no ambiguity. It is the most useful organisational habit I have ever picked up and it costs nothing.

Two: you don’t do good work with a banging headache

I spend a lot of time in environments I can’t control: client offices, expos, conference halls, motorway service stations, the back seat of a car at six in the morning. All of those places do their own thing with light and reflection and colour temperature, and most of them do it badly.

This room is the one place I can control all of that, so I do, aggressively. The colour temperature of the bulbs is matched. The blind is the right colour and the right opacity. Reflective surfaces are angled away from monitors. The yellow on the walls and on my glasses is the same calm colour my eye has stopped noticing, which is exactly the point. It does its work as background.

The reason for all of this is straightforward. You don’t do good work with a banging headache. The hours I’m not in here, my eyes are doing visual work in places that don’t get a vote. The hours I am in here, I want zero friction.

Three: the room is a stage

When I’m on a video call I don’t sit at the desk. I sit on the other side of the room with the MacBook Pro M2 balanced on the keyboard stand, looking back across the room toward the desk. The window throws good light onto my face, and the desk (lit, decorated, considered) becomes the camera background. The cupboards and exit doors are behind me, where nobody can see them.

The Mac mini M4 does the work; the MacBook does the calls. The Mac mini doesn’t have a webcam connected (I could rig the Fujifilm but it overheats; an action cam isn’t ideal either) and the MacBook earns its place on the desk for that reason alone.

Once you accept the room is a stage, you start using the set. When I’m discussing a new contract I make sure the things visible behind me are the things that lead to the right conversations. The trinkets aren’t there for me. They’re there because every single one of them has, at some point, been spotted by someone on a call and turned into a question; questions about what you do are the cheapest, warmest leads you will ever get.

What’s not in the photo

A few things didn’t make the frame, or are tucked too deep to point at:

Mac mini M4, hidden behind the monitors. The brain of the desk setup. Runs everything: build pipelines, test processes, periodic LaunchAgents, the lot. Huge compute, low electricity, bargain price. Almost everything I do touches it before it touches the cloud.
TCL Android phone with an NXTPAPER screen, used as an always-visible calendar and as a deliberately-different test platform for anything that touches a phone.
Netac external SSDs. One per client project, each project’s data on its own physical drive. Total isolation is part of our policy. We work with fintech companies; you can’t be too careful, and when a project ends the drive goes to the cold-storage shelf and stays there.
Sony lavalier mics, several pairs of headphones, various Raspberry Pis in different states of project, a Dymo labelmaker which keeps everything organised aggressively, and a USB rechargeable clip-on light I aim at my face for video calls when the natural light goes bad in the afternoon.
A Black n’ Red notebook for first-draft sketches of flowcharts and processes. Paper is faster than software for the messy version. A whiteboard wheels in for anything bigger.
A Six Sigma handbook, some Kaizen books, physics textbooks, coding textbooks. I draw on the methodology more than you might expect; I’ll also concede that the spines are easy to read on a Zoom call and I don’t think there’s anything wrong with that.

The trinkets earn their shelf space

Almost everything visible on the right-hand shelves is from someone I’ve worked with, and almost all of it has, at one time or another, started a conversation that led somewhere.

A Premium Plus sunglasses case from premiumplus.io, the Zendesk specialists.
A redk water bottle. There were conversations a while ago about me joining the team. It didn’t come together in the end, but they’re good people. Genuinely a great bottle, too. It’s in my hand most days. Nice when the swag turns out to be the real thing.
Geckoboard bits and pieces celebrating our links with that team.
Zendesk kit, scattered around. A pair of socks on the shelf (yes, really; Zendesk socks exist) and a properly good umbrella that lives in the car. Zendesk being the platform the practice is built on, and whoever ran their London event clearly knew exactly what their attendees would actually use.
An Igloo Energy desk organiser, doing useful work keeping the desk in order.
President Barbie, a gift from my colleagues at Depop, then part of Etsy, after the project there. She earns her shelf space because everyone has heard of those brands and she gives me a way to talk about the work without sounding like I’m reciting a CV. The Salto Lego set on the same shelf is in the same business.
Lanyards from past expos and a small archaeology of past projects scattered around the rest of the room.

Each one is a conversation starter and I know exactly what each one leads to. Everybody has heard of these brands. Together they give a sense of the work I do without any of the awkwardness of saying it out loud.

Going deeper

The 2026 refresh of Contracting is all in the preparation is coming soon and goes into the what to buy in proper detail: laptops, networks, peripherals, the lot.

If there’s a corner of this room you’d like me to write about properly (the audio chain, the camera setup, the backup strategy, the colour-coding system, the Solix) let me know and I’ll bump it up the queue.

Nobody appreciates the level to which my whole life is engineered, and that’s fine. The work it does is invisible by design. That’s the whole point.

Why should the Zendesk specialist even exist?

Fri, 10 Apr 2026 00:00:00 +0100

Your Zendesk processes today, visualised.

If you need help with Zendesk you can find every flavour of specialist out there, from admins to architects. There are partnership companies, agencies, individual contractors and full time employees, all under the banner of “expert” to help you manage your instance.

But the Zendesk specialist is a role that exists primarily because the tool is awkward, not because the work is inherently specialist.

Well, how did we get here?

Zendesk’s admin panel was clearly built by separate teams who never spoke to each other. Triggers live here, automations there, views somewhere else, ticket fields in a third place, and each with a different URL pattern. There’s no cross-referencing, no dependency view, no diffing, no dry runs. Every change is a leap of faith.

The specialist emerged as a human workaround for an interface that refuses to help you. They learned the admin panel by heart, made and fixed mistakes over and over, and memorised all its quirks so you didn’t have to.

Once the role existed, the platform had no incentive to fix the mess. The specialists became an integral part of the ecosystem that continues to prop it up.

When a company hires a Zendesk specialist the job advert will say:

Designing routing logic that matches how the business actually works
Reviewing macro usage against agent behaviour
Shaping SLAs around customer commitments
Auditing whether the support process still matches the product
Things that require understanding the company, not the software

All things that require a skilled human to do.

Instead, what the specialist actually ends up doing is:

Remembering which of twelve pages contains the setting you need
Keeping a mental map of which triggers reference which groups, because nothing else will tell you
Copy-pasting config between sandbox and production by hand
Combing through individual tickets to discover what the issues actually are
Finding out a change broke something three weeks later, when a customer complains
Writing internal docs that duplicate information the platform already has but refuses to expose

Account > Sandbox > Sandboxes. Three clicks deep, and all it gives you is a list.

So much time is wasted going back and forth, and what you’re really paying for is a salary for someone whose main skill is navigating a bad UI. All the knowledge lives in their head because the tool won’t surface it.

The knock-on effects

CX operations decisions get bottlenecked on whoever holds the platform knowledge. Business logic and platform trivia get tangled together, so nobody else dares touch config in case something breaks.

It’s a career trap for the specialist too. Your skills are in Zendesk oddities, not transferable CX thinking.

Every Zendesk admin thinks their instance is uniquely bad. It isn’t. The platform is genuinely like this for everyone, and the frustration is proportional to the scale of the instance. A 200-agent environment with three brands is almost unmanageable by hand.

What good tooling should do

Good tools encode the platform’s quirks so humans can focus on the business. The information is all there — which triggers reference which groups, what a change will break, whether sandbox and production have diverged. Zendesk knows all of it and refuses to surface any of it. The specialist is just the human layer that reads what the database already knows.

A proper tool would cut out the middle step. The specialist role, where it survives, should be a senior CX operations job that uses better tooling, not a human compensating for missing features.

Zendesk configuration management shouldn’t require a degree. The goal isn’t to eliminate the expert, it’s to free them up to do the work that actually needed a human in the first place.

We all know that the Zendesk admin panel isn’t good enough. We built the one we always wanted because we got tired of spending our time navigating quirks instead of fixing issues. If you are too, go to deltastring.com/beacon/ and discover how easy Zendesk can be when all the guesswork is removed.

Eat your spam

Tue, 24 Mar 2026 00:00:00 +0000

I was having a catchup with an industry colleague. She asked me where I get my market intelligence from. I said my email junk folder and I wasn’t joking.

I get a hundred or so spam emails a day. Doing what I do means having dozens of email addresses exposed to the internet and connected to various customer support platforms. (If you’re on the receiving end of this in Zendesk, read about the spam relay problem.) Automations are always trying to sell me stuff I’m not looking to buy. We all see these emails and bin them, but they’re actually worth reading. Bear with me.

I keep an eye on competitor launches and pricing changes buried in “exciting updates” via the Deltastring news aggregator but sometimes these things are discovered through marketing junk. What I get excited about is the craft of it. Fake forwards in there to create a context: “You have to see what the Zendesk expert Nico Boyce says about this.” Weird email headers and the sender domain, reply-to domain and email signature domain all different. Probing questions designed to provoke a response, any response. “Reply with STOP and I’ll leave you in peace” so you engage instead of marking them spam. These people are studying human behaviour and testing what works. Every email is an experiment. There is genuine intelligence in my junk folder.

The B2B spam flatters you. It positions you as the authority. “As someone leading CX transformation at your organisation…” It assumes you’re important and that your time is valuable and that its product needs your approval. The dynamic is: we built something, you’re the expert, would you take a look? It’s manipulative, obviously. But the manipulation works because it respects the shape of how buying decisions actually happen. Someone with authority evaluates a product and decides.

When LinkedIn launched in the early 2000s it would connect your email account to find your contacts. What it actually did was email every single person you’d ever corresponded with: “Jimmy wants to connect with you on LinkedIn.” It made everyone furious. Doug Stanhope said “Don’t get a LinkedIn, the spam is endless” on his DVD of the day and he’s not exactly Paul Graham. Everyone knew the platform built its entire user base on spam.

Nowadays the spam is the thing itself. The dynamic is reversed. Our once chronological updates from actual colleagues have been systematically replaced with targeted guff, reducing all to grey goo, exploiting fears and prejudices. AI-written carousels. Engagement bait ending in “agree?”. Inspirational stories that definitely didn’t happen to the person posting them. Every post positions the poster as the authority and the reader as subordinate. You’re here to learn. You’re here to be impressed. Hit the button and ride on my coattails. Overreaction is the only way to be heard. I scrolled past a dozen LinkedIn posts this morning before I found something that didn’t read like ChatGPT, and they weren’t even saying anything interesting. Dead internet theory needed a flagship product and it found one.

My spam folder is full of humans studying how other humans think. LinkedIn is full of machines that stopped trying. LinkedIn is behind spam email for AI-generated social engineering and that should bother more people than it does.

I read my junk folder most mornings now. I learn more from it than I do from any professional network.

We're opening up our internal Zendesk tooling

Wed, 11 Mar 2026 00:00:00 +0000

Zendesk are acquiring Forethought.ai and I am apprehensive. We evaluate products in this space constantly and Forethought is one that genuinely feels next generation. The independent agent processes constantly evaluating what’s working and where there are opportunities for improvement in your workflows and documents are great to see.

I think it’s fair to say that those recent acquisitions for Zendesk haven’t necessarily kept pace with other tools out there. My main concern is that Forethought needs nurture and not neglect.

There’s no doubt in my mind that Forethought is a good purchase for Zendesk. Now we want to see that Zendesk is a good owner for Forethought!

Have you logged into Beacon yet? For as long as we’ve been doing Zendesk consulting, we’ve had our own tooling. You can’t manage complex instances by clicking through the admin centre page by page, so years ago I built something better for us. It pulls your entire config into one place, flags the problems, and lets you actually manage things properly. Here’s a little clip of getting started in real time, one take, easy!

We’ve now opened it up for anyone who manages a Zendesk. It’s called Beacon. Connect your instance and you’ll see what I mean. Give it a go free now at beacon.deltastring.com.

Zendesk email verification: a welcome step in the right direction

Wed, 04 Mar 2026 00:00:00 +0000

Zendesk has announced that from 9th March 2026, anonymous ticket submissions via the help centre web form, Support SDK, and Web Widget (Classic) will require email verification before they’re processed. Unverified tickets land in the suspended queue. No verification, no ticket, no notification.

If you’ve been following our coverage of the spam relay problem, you’ll recognise this as a direct response to exactly what we’ve been talking about. It’s a sensible move and one that’s long overdue.

The practical implications for your team are real though. Legitimate users who don’t check their email promptly — or whose verification email lands in spam — will end up with suspended tickets. You need a process to handle that. Zendesk recommends enabling suspended ticket notifications so your team is alerted when tickets are held, and agents need to know how to recover them.

This won’t catch everything. Attackers with access to real (or disposable) email addresses can still verify. But it raises the cost of abuse significantly and eliminates the trivially automated attacks that have been causing so much noise.

Good step. Now let’s talk about scoped API tokens.

Enough is enough!

Wed, 11 Feb 2026 00:00:00 +0000

Wherever you look, the people of the internet are enraged by the wave after wave of Zendesk related spam. We’re seeing discussed on the big social platforms, we’re seeing it cause a big fuss on Reddit and we’ve seen enterprising Hacker News denziens rustle up their own Zendesk alternatives (see “I rage-quit Zendesk and built my own” for example) to try and capitalise on the widespread frustration.

We at Deltastring have long been a Zendesk-first agency and we consider ourselves as friends able to contribute well-intentioned criticism. It’s clear that more should have been done and more needs to be done to rescue the reputation of the platform.

This is not a new issue. I had conversations about how to prevent this problem with Zendesk contacts going back to some of my earliest projects in this space. It was only ever a matter of time before it was exploited in a more industrial way.

I still believe that Zendesk is a great platform and a great ecosystem for leaders looking to provide the top tier service which consumers demand. We love a good migration project (it’s a good chunk of our business!) but we’re hopeful that there is progress and we can move forward without everyone hopping off of the Zendesk train for a new platform.

To that end, I am reiterating that if your Zendesk is pumping out these spam emails and you’re struggling to ensure you are no longer vulnerable to this problem, or even better if you are proactively looking to ensure your Zendesk is safe from this type of attack, send me an email and we will provide some steps to sanity-check your configuration.

Something we always discuss with partners and clients is our commitment to growing and strengthening the ecosystem rather than fighting with competitors for a bigger slice of an ever shrinking pie. Let’s get this problem sorted and then the trade can return to business as usual.

Zendesk needs scoped tokens

Fri, 06 Feb 2026 00:00:00 +0000

Alright look. It’s 2026 and every day brings a new security disaster from a different business application vendor.

Zendesk has a vast attack surface and is mission-critical for your organisation, so the security of your instance is non-negotiable from your customers’ perspective.

It’s great that there are improvements to OAuth implementation and other routes for user access. But can anyone explain to me why API tokens are still all-or-nothing?

We need to see proper token scopes like you’d configure on GitHub or Stripe. This is considered basic security hygiene everywhere else.

Now we’re all plugging in external AI agents, we’re seeing rapid marketplace expansion, and more third parties connecting different systems. We’ve already seen the consequences with the spam relay problem. The attack surface keeps growing as attackers get more sophisticated. The cost of probing drops and potential rewards explode. As an ecosystem around the Zendesk platform, we must demand better!

Zendesk spam relay: keep your business safe

Wed, 28 Jan 2026 00:00:00 +0000

We've seen widespread issues with spammers using Zendesk instances as a relay. This is something I've fixed at various businesses over the last few years. The primary problem is using placeholders such as

{{ticket.title}}

in a trigger notification which sends on ticket creation. Many Zendesk instances have a public ticket creation form which doesn't authenticate users. Couple these two things and you can see how a bot can create thousands of tickets and send whatever link they want to many unsuspecting people.

Firstly, take any freetyped placeholders out of your ticket creation autoresponses. Nothing manually entered on the form should be bounced back to the user. You have additional options, like setting different notification triggers for different channels (this is often a good idea!) and then only use those placeholders for freetyped items when tickets are created via authenticated routes.

Lastly, if you've been a victim, please hop into Explore and you should be able to see where the spam relay issue started. The spammers will always test first on a few addresses they control, so that they can be sure of how your form and notifications work. Find their test tickets and pop those details over to your Zendesk contact. Zendesk should be able to verify that your instance is correctly updated to protect from these issues and also then to fingerprint the culprits.

If you're worried about any of this, please feel free to give us a shout and we will do these checks pro bono. You can also use Beacon to audit your Zendesk configuration. Contact me via Linkedin or email and put your mind at ease.

Commentary: Evaluating Support Platforms - The Right Questions

Wed, 28 Jan 2026 00:00:00 +0000

You might have seen we've started a series where we evaluate and compare other platforms. The Youtube playlist is here and our article looking at BoldDesk is here.

Someone asked recently "Which is the most important aspect for your support platform? Reporting, automation, agent experience, maintainability..." and while I understand the intent for me this is a bad question. Which is the most important tyre on your car? You need all four to get to your destination safely.

Here's another bad question but asked with good intention. "Which is the best X?" Fun fact, you don't need the best. The best would be an entirely bespoke solution and it would cost more than your whole department budget! You don't need the best of anything. You need the affordable option which has the compromises in the places that don't affect your workflows. Let's stick to the car analogy. Maybe the best car is a Rolls-Royce or Pagani or something similarly fantastic. My Skoda gets me to the office just fine! Then I can put my resources into the other things which are important.

The key to getting the right answers is asking the right questions. When evaluating a platform, you need to understand both what the necessary requirements are but also what are the things you don't want or need or you can live without. I've got a great Swiss army knife which does a whole load of things, but I don't eat my dinner with it. What do you think?

BoldDesk vs Zendesk: Where BoldDesk Wins (and Where it Doesn't)

Tue, 27 Jan 2026 00:00:00 +0000

I’ve been a Zendesk guy for a while. I know my way around various systems and have other strings to my bow, but the Zendesk migrations and implementations are fundamentally what pays the bills at Deltastring HQ. When new products are looking to draw businesses away from Zendesk, we’re paying attention.

If you really want to make me look, promote this on to my Linkedin feed:

BoldDesk are advertising Zendesk contract buyouts and free migration!

Time for us to take a look at this BoldDesk product and understand what they are selling. Are they a solid Zendesk alternative?

The short answer is “yes” but read on for the details and exceptions…

Here’s a video where I click through BoldDesk and talk through some of these findings. This article isn’t a direct transcript of this video. The points raised here are examined in more detail and we go some way beyond the overview here. It’s good and you should watch it.

This is my unfiltered take on what BoldDesk does well, where it falls short, and who should actually consider switching.

Who Am I and Why Should You Trust This?

I’m Nico B. Boyce, founder/CTO at Deltastring Ltd, Zendesk implementation consultant. I’ve spent years configuring Zendesk instances, training agents, building custom integrations, and solving complex support workflow challenges. The bulk of our work is on a white label basis, so you may well know me from wearing another brand’s hat.

I have no commercial relationship with BoldDesk and Syncfusion didn’t pay for this review. I registered for a trial and then popped them a few questions.

My bias, if anything, is toward Zendesk—it’s what I know and for us it’s the safe option. Recommending alternatives potentially costs me implementation work. Will I get in trouble if I say “better the devil you know” here?

I’m also frustrated watching clients struggle with Zendesk’s price increases and clunky admin interface. If there’s genuinely a better option for certain use cases, I do want to know about it.

First Impressions: The Things That Impressed Me

During the BoldDesk registration process, you’re immediately asked: “Where do you want your data stored?” The options are US, EU, India, Australia. I’m sure you’ve seen the news recently and are aware of the increasingly fractured world we are all trying to do business in. Every service needs to be considering these factors today.

Choose data residency at signup: US, EU, India, Australia.

This is massive for 2026. GDPR compliance, data sovereignty requirements, regional regulatory frameworks—these aren’t edge cases anymore. With Zendesk, data residency can require special requests, different contracts, potentially higher pricing. BoldDesk offers it upfront as a standard option. I talk about this more on Tiktok.

For European businesses especially, this removes a significant compliance hurdle.

The Onboarding Experience

After registration, BoldDesk immediately presents ~30 training videos covering specific, actionable tasks:

How to customise ticket forms
Setting up workflow automation
Configuring Copilot
Creating business schedules
Building custom views
Setting up macros

~30 actionable training videos immediately available

Compare this to Zendesk’s approach: you get a couple of tooltips, maybe an email a few days later pointing to training.zendesk.com, and you’re largely on your own.

If you’re someone who has found yourself as a support manager, the type of situation where you started off just being the sole support person at that business, and now your leadership are saying ‘right, you need to go and get five, ten more people and set up a platform’ the this is what you need. You’re actually going to find these videos do tell you how to get the bulk of stuff done.

Product tour

Admin Interface Coherence

In Zendesk, I think the kindest way to describe the admin interface is that it was built by different teams at different times and the user can sense this. Configuring triggers work one way, automations work slightly differently. Features we think of as standard objects feel bolted on.

In BoldDesk, the whole platform has this really coherent feel to it. Triggers and time triggers (a more logical name than Zendesk’s “automations”) use identical interfaces, the logic flows match, the design language is consistent throughout.

Coherent design throughout automation configuration

I found myself thinking: “Even experienced Zendesk admins would probably find that BoldDesk is quicker.”

That’s a significant statement. I’ve been configuring Zendesk for years. I know where everything is, how to work around its quirks. I’ve written my own tools to avoid the admin panel when updating the basic objects. (This eventually became Beacon.) If I had to use the web interface, I genuinely think I’d be faster in BoldDesk for standard configuration tasks.

Built-In Conveniences

Small things matter when you’re maintaining a system daily:

Excel export from any screen: Looking at your ticket fields? Export as Excel with two clicks. In Zendesk, I need to use the API or work around the web interface unloading items off-screen as I scroll.

Hover tooltips: Not sure what a setting does? Hover over it, get a clear explanation.

Logical grouping: Things are broadly called what you’d expect them to be called. It feels like being reunited with an old coat and finding your hands remember where the pockets are. Your mouse already knows the way around.

Business schedule flexibility: Different schedules for different automations. Want automated customer responses outside business hours but internal escalations only during working time? Straightforward.In Zendesk, this is a bit of a faff.

Dark Mode That Actually Works

Again, nice and coherent. Very similar to the default dark theme in Slack or VSCode. In Zendesk I keep it in the light mode and configure the colours with CSS in my browser.

Clean dark theme similar to Slack and VS Code

For people who live in other modern tools all day, BoldDesk immediately feels familiar.

It’s not just “dark mode exists and looks great” but that the dark mode is well-implemented throughout the entire interface.

BoldDesk embeds their own support chat widget in the platform, using BoldDesk itself to provide support.

BoldDesk dogfooding their own platform for support

Clearly they trust their own platform for critical business functions. My suspicion is that BoldDesk might well have been an internal product to start with anyway. Presumably their support team genuinely understands how the platform works, because they are in it as you speak to them.

I didn’t raise tickets during the trial—felt a bit rude to waste their support team’s time when I was just kicking the tyres.

What Actually Works Well (The Technical Stuff)

Agent Experience

As an agent, I would be really chuffed to be moved to BoldDesk.

The interface is going to be familiar to anyone who’s used Outlook. Views on the left (pinnable or hideable), ticket details on the right, conversation in the centre.

Inbox-style view familiar to Outlook users

Can quickly modify views via three-dot menu. Favourites system works intuitively. Keyboard shortcuts are well-documented. I never feel lost working my way around BoldDesk.

That’s the goal, isn’t it? Agents shouldn’t need to think about the tool. It should get out of the way and let them help customers. The best applications let you forget about that layer and just think about the function. I’m not operating Spotify, I’m listening to music. I’m not using Whatsapp, I’m calling my friend. We all know the applications which don’t get this right. MS Teams. Facebook.

Reporting

You want to see something else I really love about BoldDesk? The reporting features.

The pre-built reports are very comprehensive and cover most of the things you’re going to want to do. SLA performance, ticket volume, agent metrics, channel breakdown. They look coherent with the platform and they’re where you expect them to be.

Natural language reporting interface

When you go to create your own dashboard, you add widgets using templates and datasets (analogous to datasets in Zendesk Explore). The way it does calculations is logical. Widget types are familiar: KPI, card, bullet, gauge. The way it aggregates things, compares to previous periods, the date builders, it’s all very straightforward. Everything is explained in the way you’d want it to be.

As I clicked through these options, I found this has a shallower learning curve than Zendesk Explore.

Now, I think comparing this directly to Explore for a lot of businesses is unnecessary. I and plenty of other Zendesk experts don’t necessarily advise using Zendesk Explore for your real digging into data. Instead, pipe everything over to your data warehouse, use Power BI, have your actual data people look at it. Explore is a good tool for people who are managing agents or just need to understand things quite quickly.

I think the same is true of BoldDesk. But I don’t see businesses of that scale moving to BoldDesk at this point anyway.

What I Haven’t Looked At

I’ve had limited time to really get to grips with this, so there are things I haven’t looked at. I have not dived into the capabilities with webhooks. I have not massively looked into the API. I have not looked into the marketplace, though you can see all of the business applications you use do seem to have integrations there.

I didn’t raise tickets with BoldDesk during the trial. Felt a bit rude to waste their support team’s time when I was just kicking the tyres. They did check in by email though.

A short trial also can’t tell you anything useful about long-term uptime. For that, we run a live is BoldDesk down page tracking the official status feed in close to real time.

So there are things I haven’t been able to really explore. If you have specific and niche needs for integrations or applications, you’ll want to do your own evaluation or get in touch and we can help with that. We do platform evaluations for clients.

The Scale Question

I don’t know what it precisely is about BoldDesk, but I don’t see these huge scale migrations happening at this time. I think there’s a lot of momentum with Zendesk. The way that Zendesk deal with the super huge accounts is very different to how they work with some of the smaller businesses.

BoldDesk is newer than Zendesk and that matters. Zendesk has 15+ years of development, thousands of customers, has encountered virtually every edge case imaginable. BoldDesk is backed by Syncfusion (founded 2001, £39.4M revenue, 38,000+ customers including Fortune 500s), so platform longevity risk is low. But the help desk product itself is newer. Fewer customer case studies, smaller community for troubleshooting, less documentation for complex scenarios.

For businesses with unusual requirements or complex configurations, this matters.

BoldDesk vs Zendesk pricing

BoldDesk starts out dramatically cheaper than Zendesk. For a 25-agent team with AI, you’re looking at £17,700/year for BoldDesk versus £49,500/year for bare bones Zendesk. That’s £31,800/year saved. Big old chunk for a small team.

Cheaper isn’t always better, obviously. But based on my evaluation, for small-to-medium businesses, BoldDesk delivers equivalent value. The admin experience is arguably better, onboarding is superior, core help desk functionality is solid, reporting is accessible. The trade-offs (smaller ecosystem, newer platform, fewer case studies) matter more at enterprise scale.

Who Is This Actually For?

If I were a retailer or a B2C business at that point on the curve where I’m going “right, we need a platform, we have to get out of just having a shared inbox”, then BoldDesk is going to be quicker to get up and running with than Zendesk. A new Zendesk instance is very bare bones and it doesn’t have this tutorialised stuff that BoldDesk offers.

If you’re someone who has found yourself as a support manager, maybe you started off just being the support person at that business, and now your leadership are saying “right, you need to go and get five, ten more people and set up a platform”, you’re actually going to find these videos do tell you how to get the bulk of stuff done. They talk you through the whole process.

If you’re a small scale Zendesk user, feeling like you’re paying more and more money and maybe not feeling like you’re getting great results or finding it straightforward to maintain, then BoldDesk shows there are plenty of viable options out there.

That contract buyout offer is a real statement of intent. Free migration, zero downtime. Verify the specifics with BoldDesk sales, but it removes a lot of switching friction.

European or Australian businesses with GDPR or data residency requirements get that sorted at signup, which is massive.

But if you’re enterprise scale (100+ agents), stick with Zendesk. The ecosystem maturity, enterprise support model, and track record justify the premium. If you’ve built dependencies on specific Zendesk marketplace apps or you’re heavily invested in Zendesk API integrations, custom objects, elaborate workflows, the migration complexity matters. Risk-averse procurement teams that need an established vendor with a 15+ year track record won’t like that BoldDesk is newer.

BoldDesk vs Zendesk: so is BoldDesk actually any good?

For small-to-medium businesses (5-50 agents), yes. Genuinely good. Not just “good for the price” but actually good.

The whole platform has this really coherent thing to it which really beats doing the same thing in Zendesk. I think even experienced Zendesk admins would probably find that BoldDesk is quicker for standard configuration tasks, which is a really powerful thing to say. I’ve been configuring Zendesk for years. I know where everything is, how to work around its quirks. I prefer to make changes via the API because the web interface is so clunky. With BoldDesk, you’re not going to feel the same need to be doing that because everything is very straightforward.

As an agent, I would be really chuffed to be moved to BoldDesk. The interface is familiar if you’ve used Outlook. Everything is where you expect it to be. There are elements of the interface that are a little bit cluttered, but by and large, you can see where everything is and you’re not surprised by what lives where on the screen. You can hide the elements you don’t want to see. It feels like other business applications. It doesn’t feel any different to using Outlook or Slack.

Would I recommend BoldDesk to a client? Depends on their situation. A 10-agent SaaS startup looking to keep support costs and complexity low? Absolutely. A 25-agent e-commerce business with standard configuration? Worth serious investigation. Hundreds of Zendesk seats, weird workflows and complex integrations? Probably stick with Zendesk until a broader strategy review means a platform migration makes sense.

Questions or Disagreements?

If you’re using BoldDesk and have experiences to share, or if you think I’ve missed critical issues in this review, drop me an email.

If you disagree with my assessment, definitely tell me all about it!

For Zendesk implementation services or help evaluating alternatives, you’re on the right website. Read more across deltastring.com.

Last updated: 27/01/2026. This review is based on hands-on evaluation of BoldDesk trial account as of January 2026. Deltsatring Ltd has no commercial relationship with BoldDesk/Syncfusion as of publication.

Apple's Siri Google Gemini revamp and the Future of Voice AI

Tue, 13 Jan 2026 00:00:00 +0000

Last Year’s Voice AI Predictions

Zendesk publishes an annual CX Trends Report, and I was reviewing last year’s predictions to see how that looks in hindsight. One prediction that Zendesk went big on was voice AI — they talked extensively about bots and automated processes running off the back of spoken word conversations with customers.

The fourth prediction on their list was: “Voice AI gains ground as the preferred channel for complex issues.”

And I think you know where I’m going with this. I don’t feel like this has really come to pass.

(This article is a tidied up transcription of this video so give that a watch if that suits you better just now!)

I was editing a video about the Zendesk CX Trends Report when something significant happened that made me want to share some thoughts immediately.

The Reality of Voice AI Today

There are some limited applications of voice AI, and specific businesses are doing it pretty well. But by and large, voice AI hasn’t become the preferred channel Zendesk predicted, and I think that’s for a range of reasons.

Personally, I find it very frustrating. I’m looking to get to a human ASAP if I’m talking to a simulated voice. Zendesk had a strong sense that people would be phoning in, talking to an AI agent rather than typing to one, and would be happy to have their query resolved by a synthesised voice rather than an actual human. But that hasn’t happened at all, really.

Why Voice AI Has Struggled

The key barrier to uptake of voice-led AI conversations has been simple: people really don’t like talking to these things. It’s quite frustrating, and we know it’s frustrating because we’ve all tried things like Siri and Alexa and found them to be… well, frustrating.

You know how we had voice assistants in our cars 10 years ago? But you never press the button, do you? You never actually want to say “navigate me to this place” because it’s annoying. You’d use your phone instead and tap it. You’d find a place on Google Maps rather than fish around through voice commands.

The iPhone is one of the most popular devices, and we all carry a phone of some sort with us all the time. Everyone has tried these voice assistants and found them frustrating. That’s the right way to describe it.

Apple’s Game-Changing Announcement

But today, Apple announced they’re revamping Siri using what they’re calling a “foundation layer” — a model from Google Gemini for on-device processing. My understanding is it’ll still have the handoff to ChatGPT if you’re asking something that requires that level of processing and thinking.

And this is going to make a big difference.

For those of us in the tech world who’ve tried first-generation smart watches and other early tech, we’re always evaluating: is this the future? We need to stay aware of what’s up and coming, but also what’s still in heavy use. That’s why you’ll see behind me various machines, various operating systems, and each one has all the browsers on it. We’ve got to be aware of the different ways people get in touch.

I’ve had really good conversations about alternative input methods — I’ve used different keyboards like Dasher (controlled by orientation rather than tapping), even the Morse code keyboard which just has one button. That’s a bad way to do things, but you can text people from your pocket. Pretty handy, and I haven’t been able to do that since I had a Nokia 30-something.

What Apple’s Changes Mean

So if Apple are integrating Google’s foundation layer into Siri, what does that mean?

For me, it means Siri is going to have better processing of natural language requests. At the moment, it feels like you’re bumping your head against the wall — like when you play old text adventure games, typing over and over: “get sword”, “pick up sword”, “put sword in bag”, trying to find the right syntax that works.

We’ve all got used to doing that with Siri and Alexa — “turn on thing”, “start thing”, “do thing”. It’s just really annoying.

But if they’re going to have this better layer of natural language processing from Google, that might be the thing that moves us into being happier to use voice-based interfaces.

The Privacy Problem

That said, voice interfaces are always going to be something we do in private. I was walking down the road yesterday behind someone having the loudest phone conversation. Dom Jolly really showed us back in the late 90s that we’re sick of people with big phones in public spaces.

If Apple are revamping the way Siri handles natural language, a lot of people will give it another go. Some might find it’s quite good, and that might give a bit of momentum towards conversational voice interactions with AI.

Would I bet the house on it? No. There are lots of situations where you don’t want to do that. It’s something I’d use in the car while driving, or maybe with those Ray-Ban Meta glasses when there’s no one around. Sometimes it’s going to feel weird and you’ll want to use a different interface.

By and large, these things are starting to be a bit less horrible to use, and people who try to be near the cutting edge are going to give it another shot. We’re going to see a change over the coming months.

But Voice AI Won’t Replace Humans

I don’t think voice AI is going to eliminate the need to have actual people on the phone. When I get in touch by phone, it’s because I’ve got two things happening:

I’ve got something that needs to be solved quickly. I don’t want to send an email and think “this could be a week before I hear back”. I want to get someone on the phone and sort it out.
I need to feel like someone knows this has been a real pain for me. If I’m calling up saying “hey, this thing isn’t working”, I want someone to go “oh God, no, that really should be working. I’m really sorry.”

That’s the main thing I get from calling support services, and you’re never going to get that from ChatGPT because it’s never going to feel right.

People who’ve used Claude know that when it apologises, it really frustrates. It fundamentally feels very insincere. “Oh, I’m sorry I deleted the wrong folder” — you’re not sorry, are you? You’re software. You’re not sorry in the same way that the washing machine isn’t sorry if it eats one of my socks. It’s just a thing that happened where a machine has been given a task and hasn’t completed it properly.

What We’ll See Over the Coming Months

So has this changed my view on voice AI? I think it will pick up a bit. Will it be the future? It’ll be in the future for a while.

But long term, I think when we’re having voice conversations, it’s always going to be because we want to hear empathy on the other end and we want to demonstrate that this is a priority issue for us.

If I have to talk to a bot, I’d rather do it with my thumbs. That’s simpler. And I think that’s probably true of most people.

The Real Goal: Better Automation

For me, voice AI is a bit of a red herring. It’s going to be a distraction from what we actually need to achieve: greater automation.

People need to be able to solve their own problems in a natural language way. “My thing hasn’t arrived” — then be able to identify “was it this order, was it that” and have really step-by-step process automation.

You can have a GPT layer on top of that to make it more natural language and conversational, but ultimately the process needs to be step by step by step. You can design that process through seeing how people interact, but it still needs to be a defined process rather than just some AI freestyling — we know that can go horribly wrong.

The Luxury of Human Service

The other side of it is the priority service, the luxury end of service. If you’re a top-tier customer, or if you’re trying to be a brand that differentiates itself in terms of quality proposition, then human people who can actually tell you “I’m really sorry. Let me personally as a human take an individual look and prioritise your precise situation” — that’s what matters.

“Oh, you know, other people have had this. Thanks so much for raising it — gives us a good opportunity to improve.” Having conversations like that is always going to be the premium option, whereas automation is always going to feel budget.

Think of the nicest restaurant you can imagine: someone comes over and talks you through what the ingredients are. Now think of the absolute budget eating option: there’s a screen, you tap it, and something pops out of a machine.

We’ve come to understand that human interaction is the luxury part of whatever your service or product is. Voice AI is always going to, by definition, feel like a cop-out.

Final Thoughts

But let’s see what happens. Good luck to Apple, I suppose — the world’s most cautious and wealthy organisation.

What do you think? If you disagree, or if you’re implementing loads of voice AI stuff for your clients, drop me a line. If you’re using a lot of services that have these things, tell me all about it. I want to try it out — try what you’re working on, try what you’re finding to be good.

I’d like to understand what works for different businesses and different people.