I’ve been sharing what I’ve learned building SchoolsOut, my side project, about the agile product lifecycle in a startup context. The last piece: how AI showed up throughout.
Let’s state the obvious first. For a startup, AI’s impact on speed, quality and cost is genuinely game-changing. Without it, we’d have taken longer to build something worse, or spent a lot more doing it. It’s been a collaborator across the whole lifecycle: strategy, research, product design, solution design, UX, coding, testing, infrastructure. Not just code generation.
Three things stood out.
Context is everything. We talk a lot about data mattering to AI, but that framing is too narrow. It’s the context data provides that matters. What do you need to capture so AI can genuinely help, in discovery, in documenting features, in generating designs, in planning? It felt almost excessive for a two-person startup to be logging everything in Jira and Confluence like a corporate team. It paid off. That discipline is what let AI help across the lifecycle, not just when writing code.
Tooling isn’t optional. Setting up tools, connecting systems, building out a chain between them, felt like overkill for something this small. It wasn’t. It’s as critical here as in a large organisation. Tooling is what creates the context above, and more importantly, it’s what lets AI actually act on your behalf rather than just advise.
Craft is the missing ingredient. Garbage in, garbage out is as true as ever, it’s just harder to spot when the garbage comes out well-structured and confident-sounding. Human craft still matters, maybe it always will. When AI produces such compelling output, the discipline shifts: think critically, evaluate hard, know exactly what value only you can add, and make sure you add it.
I care about what we’re building and have real hope for it. But the biggest value so far has probably been something else: a low-risk space to experiment, test hypotheses, and learn, about startups, about agile product development, and about AI.