When AI Codes: Are Developers Obsolete or Just Getting Started?

Will AI Replace Developers or Supercharge Them?

Whenever a new wave of tech shows up, the first question is always: does this kill jobs?

Some high-profile voices say yes. Mo Gawdat (ex-Google) recently claimed his company runs with 3 people and “a lot of AIs,” doing what might once have required 350 engineers. It’s a provocative statement, and it feeds the fear that coding jobs are next on the chopping block.

But most practitioners see it differently. The consensus among experienced engineers is: AI won’t replace developers but it changes what developers do. Less time typing boilerplate, more time thinking about architecture, edge cases, performance, security, and whether we’re building the right thing in the first place.

Think of it this way: developers shift from carpenters to architects. AI handles the routine labor. Humans handle the design, the tradeoffs, and the judgment calls. Or as the saying goes: AI won’t replace developers. Developers who use AI will replace those who don’t.

Can One Developer Do the Work of Five?

There’s definitely truth to it. Developers using AI tools report cutting feature build times from days to hours. A single engineer with strong instincts and good prompting skills can now deliver what used to take a team. Startups love this : suddenly, two founders and one technical hire can launch products that once required a squad of engineers.

But let’s be real: the “10X developer” myth didn’t magically become true because of AI. Simon Willison (who’s been hands-on with these tools from the start) estimates a 2–5× productivity boost in the parts of the job that involve typing code. That’s huge, but software isn’t only typing. Architecture, debugging, tests, and cross-team communication still take time.

There’s also a hidden trap: less experienced devs sometimes think AI makes them faster, but studies show they actually slow down once debugging and fixes kick in. AI can write 80% of the code, but if you can’t handle the missing 20%, the project stalls. So yes, one developer can feel like five, but only if they know how to steer and verify.

The Rise of the “Citizen Coder”

Here’s where things get really interesting: AI doesn’t just amplify engineers, it gives non-engineers a shot at building, too.

Product managers, designers, analysts, people who know the problem deeply but don’t know how to code, are suddenly shipping working tools. One PM recently described how he vibe-coded an entire internal dashboard by just explaining what he wanted and hitting “accept.” The code wasn’t elegant, but it worked. And that was enough.

This opens the door to a new class of AI-powered makers. People with product or QA backgrounds who can sketch solutions in natural language, then let AI do the technical heavy lifting. That’s empowering. But it comes with limits: building a prototype is one thing, maintaining and scaling it is another. Without engineering instincts, you hit walls fast.

The sweet spot will be hybrids: people with product taste and enough technical grounding to debug, verify, and refine AI’s output. These “AI orchestrators” won’t write every line of code, but they’ll be accountable for the outcome.

Developers Become Curators, Not Typists

For career developers, this shift is profound. The job is less about raw output and more about orchestration.

You set the framework, the architecture, the APIs. AI fills in the details. Then you step back in to review, test, and adapt. You’re less a code writer and more a code curator.

That changes the skill set. Prompting matters, sure, but what matters more is judgment:

• Can you spot when the AI is confidently wrong?

• Can you design the right constraints so the AI doesn’t go off-track?

• Can you catch the subtle bug that no autocomplete model would notice?

In practice, the “last mile” of development, the tricky 20%, still belongs to humans. AI is great at scaffolding and syntax, but it doesn’t yet solve novel, ambiguous problems. That’s where human creativity and critical thinking stay essential.

What This Means for Startups and Founders

For founders, this is a massive opportunity. A small team can now build at a pace that rivals mid-sized companies. You can move faster, iterate cheaper, and test more ideas.

But speed cuts both ways. If you’re generating features at warp speed, you also need to verify at warp speed. Testing, QA, and documentation can’t be an afterthought. AI makes it too easy to ship code you don’t fully understand.

Hiring also changes. You’re not just looking for language-specific coders. You’re looking for people with strong problem-solving instincts, who can think in systems, communicate clearly, and guide AI effectively. You’re looking for hybrids: product-minded engineers and engineering-minded product folks.

And the playing field shifts. If everyone has access to AI, then the differentiator isn’t who codes faster, it’s who chooses what to build, and who understands the user best. Strategy and taste matter more than ever.

My Take

I’m vibecoding almost daily now. And I’ve realized something: the technical knowledge I picked up working alongside engineers and startups for the past 20 years, combined with my product design background and obsession with user needs, gives me a real edge.

When the AI gets stuck, that’s when experience kicks in:

• Remembering the exact library or API that gets you unstuck.

• Knowing design patterns that are battle-tested.

• Understanding how to split a system into clean APIs so the AI has less room to mess things up.

• Anticipating edge cases and testing scenarios because you’ve lived through them before.

AI doesn’t erase the value of that knowledge, it makes it more valuable. Because when you combine judgment with speed, you can build faster and smarter than either AI or human alone.

So here’s my shortlist of tools if you want to start (or upgrade) your vibe:

• Lovable.dev – Great for spinning up small apps quickly. It’s improved a lot lately with “agent mode.” Paired with Supabase, you get DB, auth, and backend in one package. Think of this as your shovel.

• ChatGPT Codex – When you want to save on AI credits and work more surgically, Codex helps you plan and execute focused tasks, integrated with GitHub. This is your axe.

• Cursor – Probably the most respected vibe coding tool among engineers. It feels like a regular IDE, but with a buddy that writes code alongside you. Reliable, transparent, and great for reviewing. This is your fine chisel.

And on the model side:

• Claude (Anthropic) is rock-solid for coding.

• GPT-5 (OpenAI) is powerful but tends to over- or under-engineer at times.

• Tools like Lovable and Cursor add their own “secret sauce” on top.

I would be happy to share more of my ongoing experience, just send me a message and let's connect!

The Bigger Picture

Developers aren’t going anywhere. But the definition of “developer” is evolving.

AI handles the repetition. Humans handle the vision. Together, they build faster than either could alone.

The gap is already opening: teams that use AI well are shipping at a completely different pace. The question for each of us, founder, engineer, product person, isn’t “will AI take my job?” It’s “how do I use AI to stretch what I can deliver?”

So here’s the question I’ll leave you with:

👉 If AI cut 60–80% of your busywork, what would you build next, and who on your team would you trust with the last mile?