Your experience is the best thing you bring to AI

If you spent ten or twenty years doing one kind of work, AI doesn't wipe that out. It makes it portable. The habits you built in your first career can now run projects you couldn't have taken on alone a few years ago.

I've spent more than 17 years implementing enterprise software and leading projects. At night I build games, small tools and AI experiments under TecodIA. I didn't expect the day job to matter much at night. It turned out to matter most.

The barrier moved

Starting something on your own used to mean finding a co-founder, a developer, or a lot of spare years. That's changing. Carta's data shows the share of new US startups on its platform with a solo founder rose from 23.7% in 2019 to 36.3% in the first half of 2025, and its report names AI as one of several possible factors, since AI has expanded what one person can get done in limited time (Carta, Solo Founders Report 2025, December 2025).

So the better question is what that person brings that the tools don't.

What experience actually does

Experience is holding its value. A Stanford working paper using US payroll data found that workers aged 22 to 25 in the most AI-exposed jobs saw a 13% relative drop in employment after generative AI took off, while employment for more experienced workers in the same jobs held steady or grew (Brynjolfsson, Chandar and Chen, Stanford SIEPR, August 2025).

Employers surveyed by the World Economic Forum expect 39% of core skills to change by 2030, and the skills on the rise include more than tech. Leadership, analytical thinking, resilience and flexibility sit next to AI and big data on the list (WEF, Future of Jobs Report 2025, January 2025).

Those are the things you don't learn from a tutorial. You learn them from projects that went wrong.

My example: every agent team gets a PMO

When I organise AI agents, the agents aren't the hard part. Coordinating them is. Agents drift out of scope, repeat work, mark tasks "done" that aren't, and nobody notices until something breaks.

I've seen that movie before, with people. So whatever method I use to organise agents, I add a PMO (project management office) layer on top:

• Visibility. One status view, updated as the agents work. At any moment I can see what's in progress, what's blocked and what's finished.
• Risk. A running risk list. When an agent hits something uncertain, it gets logged with an owner and a mitigation, not buried in a log file.
• Deliverables. Nothing is "done" until there's something to check against the acceptance criteria. Not a summary of the work. The work.

None of this is new. It's standard project management. What's new is that I can apply it to a team of agents at 11pm, alone, and get the kind of control that used to take a project office.

There's a reason I bother. Gartner predicts that over 40% of agentic AI projects will be cancelled by the end of 2027, because of escalating costs, unclear business value or inadequate risk controls (Gartner, June 2025). Two of those three are project management problems.

The counter-argument: AI levels the field

The best objection is that AI makes experience matter less, not more. There's good data for it. In a study of 5,179 customer support agents (the human kind), generative AI raised productivity by 14% on average and by 34% for novice and low-skilled workers, with minimal impact on the most experienced (Brynjolfsson, Li and Raymond, The Quarterly Journal of Economics, 2025). On a well-defined task, AI closes the gap between a beginner and an expert.

There's a second objection: experience doesn't protect you from fooling yourself. In a 2025 trial with 16 experienced open-source developers, they took 19% longer on tasks when they could use AI tools, while believing they'd been sped up by 20% (METR, July 2025). METR has since said that result is out of date and that newer tools likely help more, but the lesson about not trusting your own impression still stands.

I think both findings are right, and neither breaks the argument. They measure how fast someone finishes a task they were given. They don't measure choosing what to build, splitting it into work an agent can do, or noticing that the plan is about to fail. That's where years of experience show up. AI flattened the doing. It didn't flatten the deciding.

What to do with this

If you're thinking about building something new with AI, don't start by learning to prompt. Start by writing down the three or four methods from your first career that you'd never drop: how you plan, how you check work, how you handle risk. Then give them to your agents.

The work you've already done is the operating system for whatever you build next.

What's the one habit from your first career that you'd bring into an AI project? Reply and tell me, or email [email protected]. Happy to compare notes.