We Had the Map. We Just Couldn’t Build the Machine. Until Now.

For two years the conversation has been some version of the same thing. AI is coming for the jobs. Which jobs keeps changing, but the anxiety has been pretty consistent.

That conversation isn’t wrong exactly. But I don’t think it’s the interesting part anymore.

What I’m actually living through, and watching others start to figure out, is what happens when you use AI to build something you couldn’t have justified building eighteen months ago. Not a tool. A system. Something that runs your business the way your business actually runs, not the way some SaaS company decided professional services should work.

We’ve been doing this at Bespoke. We had the whole thing mapped out. We knew exactly how the business should run. The problem was it lived across five systems and seven subscriptions that had nothing to say to each other. The map existed. The machine didn’t.

What’s been genuinely strange, in a good way, is that I built it myself. Not a developer. Me. That sentence would have been nonsense two years ago. Now it’s just true. And the part that’s hard to explain to people who haven’t done it is how much you can fine-tune it when you built it. Not adapting to someone else’s logic. Not working around the thing the software won’t let you do. Just this is how we work, so this is how it works.

What we’re working toward is a platform where marketing and sales and production and finance aren’t four separate conversations that someone has to manually connect. Where a new project brief starts populating fields downstream without anyone copying and pasting. Where the numbers you’re looking at on any given Tuesday reflect what’s actually happening, not what was accurate three weeks ago when someone last updated the spreadsheet.

Not groundbreaking as a concept. But getting there used to require a budget and an engineering team. It doesn’t anymore.

McKinsey’s State of AI report made a point that stuck with me. The businesses seeing the best returns from AI weren’t the ones with the most tools. They were the ones with connected systems. AI embedded in how the work flows, not sitting alongside it in a separate tab.

Most businesses are still in the separate tab phase. A growing collection of monthly subscriptions that each do something useful and share no data with each other. That’s fine as far as it goes, but it’s also what everyone else has. You can’t build a real advantage on tools that are available to every competitor at the same price.

Klarna is the case study everyone cites when they talk about AI reducing headcount. Replaced 700 support staff, revenue per employee went up, made for a clean IPO story. What got less airtime was that they started hiring again. The work didn’t shrink. The ambition did what ambition does when you give it more capacity.

That’s been our experience too. You go into it expecting to end up with more bandwidth. What you actually get is a higher ceiling, and the ceiling gets filled. Doors open that weren’t there before. Work becomes possible that you couldn’t have taken on or even pitched six months ago. We’re not looking at a smaller team. If anything I’m starting to think about roles we’ll need to hire for that didn’t really exist as a category until now.

I’m not sure the jobs being replaced story has played out the way anyone predicted. What’s shifted is the standard those jobs have to be done to, and the kind of work worth going after in the first place.

There’s a version of this that doesn’t work, and it’s the version where you use AI to do the same thing cheaper rather than to do something better. Cut the team, hold the output flat, protect the margin. Which looks good until someone else uses the same tools to do considerably more than you, at the same price, probably faster, in ways clients eventually notice.

Cheaper isn’t better, it’s just cheaper.

The way I’ve started thinking about it: AI is a translator. Between human ideas and creativity and what computers can actually do. That gap used to require a developer, a budget, a project, a timeline. Now it’s just you explaining what you need, and something building it. Which means an enormous amount of work that previously couldn’t exist, now can. New roles. New services. New kinds of businesses that aren’t possible without that translation layer.

That’s not a threat to jobs. That’s what every major technology shift has done. The question is just whether you’re on the side of it that’s building something new.

The question worth sitting with is... if you could build the exact system your business actually needs, not the closest available approximation, not the tool that mostly fits, what would it do?

That used to be a question for companies with engineering departments. It’s not anymore. Which is either exciting or a bit destabilising depending on where you’re sitting.

Probably both.

Curious where others are with this. Still mostly at the tools stage, or have you started building something more connected? And if you have, has it changed what kind of work you’re going after?