The Same Job, A Different Team

My time at Provet Cloud reached its natural conclusion, and I had to figure out what came next.

I'm immensely proud of what we achieved there. We didn't just ship features; we rebuilt the product engine. We moved the organisation toward an outcome-based strategy and launched a Clinical AI line that didn't just work, it opened a meaningful new revenue path for the business.

But that chapter was complete.

As I looked at what came next, I set two clear goals. First, to move beyond the strategic theory of the AI shift and get back into the mechanics of it, building and testing from the inside out. Second, to be unapologetically selective about my next home: it had to be somewhere that hands-on conviction would actually matter, not just get nodded at over a coffee and forgotten about.

Four months later, I'm starting as VP of Product at Medius. This is the story of the space in between.

An education I'd been putting off

Throughout 2024, and especially into 2025, I'd been watching AI change the texture of product work in small but meaningful ways. Better specs. Faster synthesis. Cheaper prototypes. Real progress, but mostly incremental.

I understood the landscape. Now I had the chance to live inside it.

So I started where a builder starts: at the keyboard.

I bought a Mac and went back to writing code properly. Not to become an engineer again, but to understand how engineers are now working.

The first thing I shipped was my own site: a clean home for the writing and the work, end to end on the live web. It wasn't much. But it was the first time I'd taken something out of the scratch-pad and put it somewhere real, and that mattered more than the output.

That was month one.

Then I moved up a layer into agentic coding: not AI that autocompletes, but AI that plans, executes, reviews, and corrects itself across a task. Different muscles. Different mental models.

By month three, I was building my own agent on top of OpenClaw, and learning how an agent stops being a tool the moment you give it the right context, boundaries, and access.

When I coupled that agent to an MCP server pointed at a live SaaS product, something shifted. The agent wasn't calling the software. It was operating it. The way a new hire operates in week three, without needing the first two.

That was the moment I stopped thinking about AI as a tool I was using, and started thinking about it as a team I was running.

From there, the model changed quickly. I went from directing AI step by step to orchestrating a team of autonomous, skilled agent personas: setting the problem, the constraints, and the shape of the team needed to solve it.

If you've led product teams for twenty years, that should feel familiar, because it is.

It's the same job.

You're still framing problems, setting constraints, composing teams, and judging the work that comes back. What's changed is who (or what) is on the other side of the brief, and how cheap it is to try the next version.

Six weeks, one niche, one proof point

Then I did the thing every product leader should do at least once: I stopped theorising and shipped something real.

I picked a niche the software industry has largely ignored, kennel management, and over roughly six weeks I built and launched a working SaaS product. Bookings, billing, customer records: the lot.

It runs with minimal intervention and generates a small, steady stream of income.

Six weeks, solo, should not be possible. The only reason it was is that I wasn't building alone. I was briefing a team.

The product itself is a side project I'm proud of. But it's not the point.

The point is that the economics of building have changed, and I felt that change directly.

What changes for product leaders in this mode

I'm wary of anyone returning from a sabbatical with a manifesto. But a few things feel genuinely different now, and they're worth saying out loud.

The unit of work has moved from a ticket to a problem with constraints. You're no longer handing an engineer a job; you're handing an agent a problem, a guardrail, and a definition of done. The more granular your instructions, the less leverage you get. If you've been writing good PRDs for twenty years, you've been practising for this job without knowing it.

Team shape has stopped being an annual headcount decision. Working solo with agents, I was reshaping it daily. Sometimes hourly, whenever the problem in front of me demanded a different kind of help. At team scale the clock runs slower, but the direction is the same: team composition is something you decide as the problem arrives, not a headcount you negotiate once a year.

And the cost of trying something has collapsed to near zero. When everything is cheap to try, the scarce resource becomes knowing which thing is worth trying.

If I were putting this on a whiteboard, it would look like four shifts:

• Briefs over tickets.
• Guardrails over specs.
• Team shape over headcount.
• Judgement over iteration.

Building with agentic AI isn't a replacement for a product team. It's a multiplier on the team that knows how to use it.

Why Medius, and why now

When the time came to choose where next, Medius was the one I kept coming back to. They're a Swedish-founded spend management platform covering the full source-to-pay process, and they've been training AI on real transaction data since 2016. That's a nine-year head start most companies can't buy.

They're now a Gartner Magic Quadrant Leader and deploying agentic AI across the full suite. Very few companies have both the data depth and the distribution to turn agentic AI into something customers trust with their money.

Medius is one of them. I took my time with it, and chose it carefully.

The kennel SaaS runs quietly in the background now. It was never the point. The point was the education, and that's what I'm starting with.

If you're a product leader watching this shift from a distance, my only advice is the one I gave myself four months ago. Buy the Mac. Open the laptop. Go in. You'll understand more in a month of hands-on than a year of reading about it.

That's what four months of agents taught me.

John