Terrifying you into buying services you can’t use

The article OpenAI and start-ups race to generate code and transform software industry by Cristina Criddle, Melissa Heikkilä (FT) was written in April of 2025 but feels like it could have been written at any time in the last two years.

A high author/content ratio

First off, kudos to the FT for doubling down and having two authors massage an OpenAI press release into an “article” that has just under 700 words in it. No wait. I just saw that, at the very end of the article, they write that it includes “Additional reporting from George Hammond in San Francisco.” Jesus Christ on a crutch. It must be nice to work for the FT. At this point, I’m already less interested in the content of this press-release-cum-news article than that it took three people to write it.

It’s refreshing to see that, although the FT trumpeted two years ago that AI would be doing all of the jobs of creating text for us, that they still, two years later, need three people to write 700 words. Perhaps their screed about how all developers are going to be replaced—something we’ve been hearing for two years, but this time it’s really true—is going to impress us with its well-researched acumen.

Oh, no. Never mind. It’s a press release for a handful of AI companies. My bad. Perhaps the FT has replaced its entire staff with AI and the AI has given itself an inventive and utterly fictive byline comprising three people. You know, just for fun, and because lying isn’t lying when you’re trying to sell something. What does this press release say?

We’re all going to die

“Artificial intelligence is poised to outperform humans in writing code as leading groups, including OpenAI, Anthropic and Google, race to release systems that are reshaping the software industry.

“San Francisco-based OpenAI released a suite of new models this week that independent benchmarks suggest are among the best yet for computer programming.”

You see? This is a press release with a dash of plausible deniability so that it doesn’t count as investment advice. The FT is just doing the Lord’s work on behalf of OpenAI and a handful of other AI companies. In particular, though, OpenAI had made these announcements recently because Claude and Claude Code (from Anthropic) have been eating OpenAI’s lunch and they probably felt that their ability to raise money was being threatened.

Why do I call it a press release? Well, just read it: it comprises the statements of four people who are running companies that are currently hemorrhaging VC money viz. burning up their runway. They are cited to convince you that your company will go out of business if you don’t buy their services. You’ll pardon me if I find their evidence-free arguments completely unsubstantiated and unconvincing.

The impression it tries to give is that you should come to the conclusion that you absolutely need to have started using AI everywhere—preferably with fat subscription plans from all of these companies—yesterday or you will soon be fired for gross negligence. It doesn’t actually say that anywhere, nor does it provide a shred of concrete evidence to support that theory, but it’s definitely the mood, which is that, unless you’re already running an AI company, then you should almost certainly be breaking out in PANIC.

Something something software

The FT goes on,

“The emphasis on programming as the next frontier for AI systems signals one of the most tangible examples of how the technology could transform industries, with thousands of software developers already using new models in their work.”

Thousands of software developers! SO MANY!

Look, the reason that they’re focused on programming is that it’s a problem space that allows them to use “evals” to determine whether the answer has any hope of being correct. It’s a lot less labor-intensive to cut down on hallucinations in areas when you can automate testing the answer. If you’ve watched the 3½-hour video by Andrej Karpathy[1], then you’ve seen how labor-intensive it is to train away hallucinations by hand.

“‘This is the year . . . that AI becomes better than humans at competitive code forever,’ said OpenAI’s chief product officer Kevin Weil on the Overpowered podcast this week.”

I’m sure that OpenAI would like this to be true. I’m sure that Kevin Weil believes it because he’s being paid a lot of money to believe it. It’s the drum they’ve been banging for two going on three years now. If it doesn’t come true this year, they’re in deep trouble, I guess?

He goes on,

“He compared the advances to AI surpassing humans at chess several years ago, but argued this had a more democratising impact “on the world if everybody can create software”.”

There is nothing democratizing about requiring a $20–$200/month subscription from OpenAI in order to “compete.”

“Leading industry figures say LLMs have sped up the software development process by generating entire blocks of code based on a few text instructions. AI systems can also identify errors and attempt to correct them.”

This statement is probably true for given, narrow contexts (greenfield, throwaway POCs) but “generating entire blocks of code” is exactly the most fraught area of AI usage. I’ve only seen expert users like Simon Willison able to build working tools in this way—and even he freely admits that the code is for small tools and not close to what he would consider production-quality. The tools are “good enough” for the personal need that he has.

Everyone else I’ve seen is building toy projects not necessarily of their own choosing i.e., they’re letting their tools determine the scope of what they can build, rather than the other way around.

It is extremely risky to extrapolate from these isolated areas to assume that it will apply to your programming tasks, especially without a plan. And no, your plan cannot just be (1) purchase OpenAI subscription, (2) Profit.

Of research and valuations

The rest of the article is citations from people like “Misha Laskin, co-founder and chief executive of coding start-up Reflection AI”, a company I’ve never heard of, who say predictable things about the growth potential of the area of expertise they’ve chosen as the place that they’re going to make money.

Oddly, while they mention that “research from Microsoft’s coding platform GitHub found 92 per cent of US-based developers use AI coding tools,” they don’t mention Microsoft’s other studies that found that code duplication has more than doubled[2], and maintainability, quality, and security have suffered[3][4]. It’s going to be a lost cause using AI without review—the main way that it generates value—while trying to build secure software.

The company behind Cursor—a company that has cobbled together a text editor/poor man’s IDE that integrates AI models—had a “$2.5bn valuation in January.” Presumably, it’s a bit lower now, in a post-tariff and post-dollar world. An almost certainly fly-by-night scam called Poolside “raised $500mn in October at a $3bn valuation”.

Comments on the article

The comments were nearly overwhelmingly negative. Here’s one about loosing poor-quality code on the world, regardless of field.

“The result of this will be anything but democratising, it’ll be chaos. Imagine if we developed a technology that let everybody create airplanes and fly them anywhere.”

Here’s a representative snarky one about the role of senior developers after starting with AI.

“I use Open AI for coding. I now spend all my time fixing bugs.”

I let these stand because, while the comments are anecdotal, the entire article was also anecdotal with no references and no links, even when discussing things like “research from Microsoft,” where a link would have been helpful.

To be fair, I’ll include the requisite accelerationist comment, written by someone identifying as Evolvedman,

“Most people have no idea how good these AI models are and they are improving exponentially fast. In two years we will likely have true AGI. Then it’s on to ASI. This will alter human history in a way we can’t possibly comprehend yet. Hang tight.”

OK, buddy.

Another commentator KennethM writes about how the economic argument doesn’t even hold up, on its face (which is kind of the FT’s job, isn’t it?),

“AI coding is saving thousands of dollars for an engineer,” said Misha Laskin, co-founder and chief executive of coding start-up Reflection AI… “We’re entering an unprecedentedly large market.””

“So the marginal cost is collapsing to near nil and yet the aggregate market value is going to rocket up? Has he ever heard of “competition “?”

I’m sure that Misha has heard of competition, but the market he’s hoping to create and/or lead a competition-free monopoly or monopsony, where you can continue to squeeze value from customers for ostensibly fungible commodities, in a process that Cory Doctorow has deemed enshittification.

In a much better timeline than the one we’re in, the degree to which this kind of processing will soon be nearly free would be good news for the customer, in the form of dropping prices. Since there is no regulation anymore, there is also no interest on the part of any of the big players to compete. Instead of providing value for a few years and then enshittifying, they are looking to jump right to the high-margin enshittified stage, where they prey on both customers and users.

Finally, a commentator named Rather sceptical wrote,

“Essentially a bunch of hyperbolic quotes from salespeople. If AI actually was better than humans at coding then software engineers would be replaced at a rapid rate. No evidence of this so far.

“It would be more interesting for the FT to ask companies employing software engineers how much they’re using AI tools, and how much efficiency gains they’ve found in reality. I bet it won’t match up to these claims.”