Intro #

Ever since the supposed “AI coding inflection point” in November 2025, everyone’s been moving fast. There’s been a wave of Substack think pieces, tweets, and LinkedIn posts about how productive everyone has become thanks to the democratisation of software development. Apparently anyone can build anything now. We’re entering a new golden age of productivity.

But, to what end?

If AI Works As Advertised #

Let’s assume, for a second, that AI coding tools can match the hype. In this world, Claude Code, Copilot, Cursor, etc. are all genuinely transformative tools. They crush bugs that used to take days in hours, ship complex features in an afternoon, and so on. Junior developers with an AI subscription are shipping code at the pace of an experienced senior engineer.

If this happens – and I am quite sceptical that it will in the near future, especially at present price points – then the important question becomes: what separates successful software companies from unsuccessful ones?

After all, if everyone can quickly ship features and bugs, then engineering speed as a competitive advantage essentially evaporates. I have already replaced a few tools that I may have been forced to pay for with homemade versions because I can code and AI makes me faster. Sure, they are stripped down versions with only the features I need, but that’s precisely the point. The barrier to “good enough” is collapsing.

Anyone with even a mid-tier Claude Code subscription can clone your product if all it is is AI-generated features. You could spend months developing and polishing a feature. Your competition could replicate it in days after seeing it in production.

The Panic Response #

What I have seen (and it terrifies me) is that the panic response from most companies to this “AI moment” is to demand more. More hours. More features. More velocity.

It makes sense, right? If everyone has access to the same AI tools, then competitive advantage comes from who can grind the hardest. Who’s willing to work 80-hour weeks? Who’s willing to sink thousands of dollars into API tokens? i.e. who can ship the fastest!

But, this is absolutely insane. It’s a race to the bottom, a war of attrition where the prize for winning is burnout and technical debt.

When I look at the Big Tech companies and how they actually succeeded, it’s clear that it was never about out-hiring the competition or working the most hours. Google didn’t beat Yahoo because they had more engineers. Facebook didn’t beat MySpace through sheer volume of code. The advantage was quality of talent. Having the best developers will always give you an advantage.

So, what happens then when the playing field is levelled by everyone using the same AI models from OpenAI, Anthropic, and Google? You can’t hire “better” AI. You can’t poach Claude Code from your competitor.

That leaves human effort as the only variable left. So companies default to the metrics they can control: hours worked, lines of code pushed, etc. But this is pointless because everyone is using the same tools. Your 80-hour week isn’t materially different from your competitor’s 80-hour week when you’re both prompting the same model.

Confusing Motion With Progress #

My thesis is this: AI is creating an illusion of progress that could destroy the very advantages that actually matter in software businesses.

Everyone is so giddy about what they can build that they’re not stopping to think about what exactly they should build. The rhetoric right now is “everyone is moving fast with AI, we need to move just as fast to keep up”. But is there an actual plan or are you just shipping features for the sake of it?

In this new age, developers must become picky. I, too, know how intoxicating it is to have all this power in your hands and feeling like every “good” idea you have ever had is finally within reach. It’s always been true that knowing what to build was a scarce resource, but there were always physical constraints stopping developers from chasing every squirrel up a blind alley. Now, bad ideas can be built just as fast as good ones.

It’s also important to understand that there’s more to software than just the code. If, let’s say, Claude Code can completely replicate anything – take Stripe, for example – what do you think is stopping Anthropic from doing so and putting every company out of business? I’d wager that it’s that you can’t clone the fact that millions of businesses trust Stripe with their payments. Software is more than just code. To quote Ed Zitron:

Every single piece of SaaS anyone pays for is paying for both access to the product and a transfer of the inherent risk or chaos of running software that involves people or money. - Ed Zitron, The Hater’s Guide to Private Equity Link

Most importantly, especially for somewhat established software companies, what happens when your rushed, AI-coded features introduce critical production bugs? Because, no matter how many precautions you take, bugs in code are an inevitability, and writing and shipping more code just increases the surface area for bugs to appear. As seasoned software engineers will tell you: code is a liability.

We are still in the honeymoon phase where everyone is celebrating their newfound velocity and the promise of escaping human limitations. But production bugs from insufficiently understood AI-generated code are coming. As Simon Willison predicted, a Challenger disaster for AI is coming, and the reputational damage will be catastrophic in ways that far outweigh any temporary speed advantage.

What Actually Matters Now #

If AI does commoditise implementation – and that’s still a big if – then software businesses won’t die. They will just be forced to compete on the things that actually matter such as deep relationships with customers, reliability, and strategic clarity about what problems are worth solving.

I understand the anxiety, especially for companies in the middle, because they are the most vulnerable right now. Most of them are established enough to feel threatened by faster-moving AI-native start-ups, but not established enough to have the reputation and customer lock-in to weather mistakes (only Microsoft can do that). Those are the companies most likely to fall into the velocity trap.

The companies that will thrive aren’t the ones grinding out 100-hour weeks trying to out-ship everyone else. These companies will recognise that in a world where anyone can build anything, the more important question is “what should exist, and why would someone choose ours?”

Uncomfortable truths #

Maybe the real change isn’t that AI eliminates engineering advantage. Maybe it’s that AI exposes how little of software business success was ever really about engineering in the first place.

“Code has always been the easy part”, Kellan Elliott-McCrea, Link

We are about to find out which companies were winning because they had superior execution, and which were winning because they had superior judgement, positioning, and relationships.

I reckon the velocity addicts will learn the difference the hard way.