The constraint on building has shifted.
It’s not execution capacity any more. It’s judgement. This week made that obvious.
FATP is relaunching next week. The previous rebuild took two to three months. Custom content architecture, user portals, thousands of pages, each one richer than the last, auto-updating as the underlying data changes. This version took days, because I’m building differently. I direct agents that write the code. I describe what I want, check what comes back and iterate fast.
The gap between idea and execution has collapsed. I keep coming back to “creating at the speed of thought” because it’s the most accurate description I have of what’s happening.
When you can move from concept to working product in hours rather than months, it changes what you’re willing to try.
Tech-grade visibility
I built custom dashboards for the bakery this week. Sales data against labour costs. Rota software feeding in via API. Management finally able to see what’s actually happening in the business, in real time, without someone manually compiling a spreadsheet. Took a few hours.
I mention it as a data point. Custom software for a small business used to need a development agency and a five-figure invoice. Now it gets done in an afternoon. The bakery isn’t a tech company, but it now sees its own numbers the way a tech company would.
Do you still need investment?
I also went to a funding event this week. Recruitment founders/CEOs and investors. AI was the constant thread. Everyone talking about it was no surprise. That’s been true for two years. The subtext was more interesting.
Investors want AI in the mix. Traditional models are getting a cold reception.
My read is that for businesses that genuinely know how to build with agents, external capital is worth less than it was. Money still matters. But much of what capital used to buy (time, people, infrastructure) can now be had another way. The bakery dashboards and the FATP rebuild used to be five-figure line items. Now they’re afternoons.
That changes the negotiation. It changes what you’re willing to give up.
I’m not saying don’t raise. Before you do, ask what you’re actually buying with that money, and whether there’s a faster path.
Autonomy is earned
This is the part the AI conversation tends to skip.
Tiding is live. Sentinel Talent is live. Relay is live. We’re moving toward a fully autonomous recruitment loop. And then something broke. Model selection failures led to wrong emails going out across multiple recruitment projects. The model took its instructions too literally. No judgement applied. It did exactly what it was told, which was exactly wrong in context.
Update, October 2026: Tiding and Sentinel Talent were experiments at the time, and both have since closed as projects.
So, I put the brakes on.
Speed without judgement is just fast mistakes. Agents, for all their capability, don’t have the judgement a person builds through years of repetition and consequence. They don’t feel the weight of a bad call.
When the edge cases arrive, and they always do, someone has to catch it. Right now, that someone is me.
I didn’t come away thinking agents are dangerous. I came away with two rules: define your guardrails before you need them, and stay close enough to catch the errors. Autonomy gets earned, one workflow at a time.
Execution was never going to stay the constraint. What’s left is your ability to judge when the machine is right, when it’s wrong, and when it’s confidently, efficiently doing exactly the wrong thing.
The trade
I’m building at a speed that wasn’t possible two months ago. The bigger change is where the hard work now lives. Less of it is in the doing and more of it is in the deciding.
Less capital than we thought.
Less time than we thought.
More judgement than we thought.
That’s the trade. Make sure you’re getting the good end of it.
If you run a founder-led business and want one workflow, like the bakery’s sales and labour dashboard, built with the guardrails in place from day one, I work hands-on inside a small number of businesses. Here’s how that works.
