AI tooling and the closing window

We're in a unique window right now. One that I think will look obvious in hindsight, but that most people — even technical people — haven't fully grasped yet. The cost of building software is collapsing, and it's happening faster than anyone expected.

I'm not talking about some theoretical future. I'm talking about right now, today, in March 2026. I've been building things with AI tooling for months and the acceleration is visceral. What used to take a team weeks can be scaffolded in an afternoon. What used to require deep specialist knowledge can be handled by someone with decent judgement and the right prompts.

The simple stuff is basically solved

Let me give you a concrete example. I recently built ScubaAtlas.cc — a dive site and dive school directory with searchable listings, geo filtering, enquiry forms, an admin panel, and review moderation. Nuxt 4, Prisma, PostgreSQL, deployed to Vercel. Thirty-two pages, fifty-two API routes, full admin dashboard.

The initial MVP took a couple of days. Not a couple of sprints. Days. And I wasn't writing most of the code — I was directing an AI agent, reviewing its output, course-correcting when it went sideways, and making architectural decisions. The ratio of thinking to typing has completely inverted.

Same story with NPSI Fleets, an EVE Online fleet management tool. Database schema, API routes, an EFT fitting parser, ESI integration, real-time fleet composition tracking — all stood up in a single session. The kind of project that would have been a solid month of evenings a few years ago.

But the hard stuff is where it gets interesting

Simple projects are one thing. What about the genuinely complex ones?

At Cartridge Save, we're working through a migration of our main e-commerce platform from Nuxt 2 to Nuxt 4. If you're not in the Vue ecosystem, just know this: it's not a version bump. It's closer to a philosophical shift. Almost everything about how you structure a Vue application changed between these versions.

On top of that, we're migrating from Vue Storefront's older API layer to modern equivalents. The VSF middleware that used to handle API abstraction, the composables that wrapped their SDK — all of it needs rethinking. Custom integrations, payment flows, search, the lot. You can't just find and replace your way through it.

This is where AI tooling gets genuinely useful — and where it still needs a human at the wheel. An AI agent can convert an Options API component to Composition API in seconds. It can refactor a Vuex module to Pinia beautifully. It understands the patterns. What it can't do — yet — is understand why a piece of code exists in a legacy codebase, what business logic is hiding in that computed property, or what happens downstream when you change how the cart state is structured.

That's the oversight window I'm talking about. Right now, in 2026, you still need someone who understands the domain, the architecture, and the consequences. The AI is an extraordinarily capable pair of hands, but it needs a brain directing it. The question is: how long does that remain true?

The window is closing

Six months ago, AI coding tools could autocomplete functions and suggest boilerplate. Today, I'm having an agent build entire features end-to-end — database schema, API routes, frontend pages, tests — while I review the output and steer. I spawn sub-agents for parallel workstreams. I describe what I want in plain English and get back working, deployable code.

The trajectory is clear. The things that still require human oversight today — complex migrations, nuanced architectural decisions, understanding legacy business logic — are exactly the kinds of problems these models are getting better at. Context windows are growing. Reasoning is improving. The ability to hold an entire codebase in working memory and understand the ripple effects of a change is getting closer.

What this means if you lead a team

If you're a tech lead or CTO reading this, the implications are significant and they're arriving faster than your planning cycle accounts for.

The value of technical oversight is at a premium — temporarily. Right now, the people who can direct AI tools effectively and review their output critically are absurdly productive. A senior developer with good AI tooling can outpace a small team. But this advantage is a feature of the transition, not the destination. As the tools improve, the bar for useful oversight drops.

Legacy migrations are the immediate opportunity. If you've been putting off a major framework migration or platform modernisation because of the cost, the equation has changed. What would have been a six-month project with a dedicated team can now be tackled in weeks with AI-assisted development and experienced oversight. The ROI on technical debt reduction just got dramatically better.

Hire for judgement, not just output. The developers who will thrive aren't necessarily the fastest typists or the ones who've memorised the most APIs. They're the ones who can evaluate whether a solution is correct, maintainable, and appropriate for the context. Code review skills matter more than code writing skills. That's a weird sentence to write, but it's where we are.

Cautious optimism

I'm genuinely excited about what's possible. Building ScubaAtlas and NPSI Fleets has been fun in a way that development hasn't been for a while. The tedious parts are handled. The creative, architectural, problem-solving parts remain — and those are the parts I actually enjoy.

But I'm also watching this with clear eyes. The window where human technical oversight is essential is closing. Not tomorrow, not next month, but it's visibly narrowing. The question isn't whether AI will be able to handle complex migrations and architectural decisions autonomously — it's when.

For now, we're in the sweet spot. The tools are powerful enough to be transformative but not yet capable enough to be autonomous. If you're technical, lean into it. If you lead a team, tool up. If you've been putting off that migration — this is your moment.

The window's open. It won't be forever.