The AI ​​code wars are heating up

This is it Step back, A weekly newsletter covering one essential story from the world of technology. To learn more about AI coding mutations and bio-coding, Follow David Pearce. Step back It arrives in our subscribers’ inboxes at 8 a.m. ET. Subscribe to Step back here.

Writing code was a killer application for AI even before anyone was really talking about AI. In the spring of 2021, 18 months before the world knew the word “ChatGPT,” Microsoft launched the first product of a partnership with a non-profit called OpenAI: A tool called GitHub Copilot Which saw developers writing code and trying to auto-complete snippets and lines for them. It wasn’t very good at all, and was only a “restricted technical preview,” but over a million developers signed up to try it anyway.

Clearly, large language models are poised to make software development simpler and faster. Most of the code is relatively structured and straightforward; Coding languages ​​are generally very well documented; There is a huge amount of code available online for use in training models (albeit through sometimes questionable means). Unlike a lot of other information you might get from an LLM, you can also check the quality of the code by simply trying to run it. Initially, some companies thought that LLM holders might be able to write code faster By predicting the next word in the same way as Google Autocomplete. But they were hoping that it would soon be able to do some of the programming for you. Maybe even all of that.

For many years, companies in the technology industry have also followed the idea of ​​“low-code” and “no-code” software. Instead of presenting users with endless lists of settings and unparseable menus, the idea was to effectively let people build software themselves. For a long time, this was kind of innovative: you had things like Zapier and Apple Shortcuts, which were actually super-sophisticated workflow creation tools; Or you had programs like Notion and Airtable, which were very flexible at the cost of being difficult to discover.

Even in those early days, it was also clear why AI programming tools might one day become good business. Developers are expensive; It takes a long time to create a product. Any tool that could mean companies can hire fewer developers, or help developers be more productive, would certainly be an easy proposition for software companies around the world. If the technology works, the products will practically sell themselves. Companies like Cursor and Windsurf have raised huge sums of money trying to build companies around AI programming tools, while OpenAI, Google, Anthropic and others have started building new products for developers.

Initially, AI encryption tools could not be trusted. For a few years, maybe they can complete a few lines of code, but they should always be checked. In late 2023, Simon Willison, a programmer and blogger, called LLMs “Weird coding internsHe wondered whether these interns would make programmers more versatile and powerful than ever before, or eventually begin to replace them.

In early 2025, Anthropic released a product called Claude Code that should make this question more pressing for many people.

In late 2025, Anthropic released a new version of Claude LLM, called Opus 4.5. By Anthropic standards, this was Claude’s best model yet, but it doesn’t seem like it represents some amazing advancement in AI technology. However, after a few weeks, a lot of developers with a few free hours over the holidays started testing the new model in Claude Code, and they seemed to almost universally come to the same conclusion: It works. Suddenly, a tool that you previously had to prompt and carefully review can turn a few sentences into a working prototype. Boris Cherny, the creator of CloudCode, says he has already had AI write 100% of his code. “It was as surprising to me as it was to others.” He said Edge Earlier this year. In a way that seemed impossible for a programming tool, Cloud Code spread widely.

CloudCode may have captured much of the software world’s imagination, but Anthropic’s competition wasn’t far behind. OpenAI’s Codex, which was launched in 2025 a few months after Claude Code, has received a series of updates and is also a powerful and popular tool for writing code. Google has rolled out a command-line interface for its Gemini model, and has also recently put more programming features into its AI Studio app.

Increasingly, AI coding looks like the first truly mainstream use case for AI — not to mention the first major potential work in AI. Claude Code’s moment coincided with Absolute explosion in revenue for anthropic; A senior executive at OpenAI recently asked her team to stop doing “side quests” and instead focus on competing with Anthropic and Claude Code. Both OpenAI and Anthropic are reportedly planning to go public this year, which means both companies will need something to show for the billions they’ve raised in capital, and the billions they’ve burned on computing. It seems like everyone’s best idea is to write code.

In fairness, that seems like a pretty reasonable guess. Suddenly, companies all over Silicon Valley are seeing employees vying to use the most code, using GPU access as a recruiting tool, and openly bragging about their AI bills. Nvidia CEO Jensen Huang recently said he was concerned about any overpaid engineers He wasn’t spending $250,000 a year On artificial intelligence codes. Although developers fear that AI programming tools could spell the end of their careers and livelihoods, the race is on to adopt them as quickly as possible. One study 2025 It found that 98% of participants said they used AI coding tools “several times a week.”

It’s not just developers. In February 2025, Andrei Karpathy, a veteran expert in the field of artificial intelligence, He coined the term “atmosphere coding.”“I’m building a project or a web app,” he wrote on

It’s a shame Karpathy didn’t come up with a catchy phrase, because the enthusiastic programming stuck. The name as well as the phenomenon: a lot of people who couldn’t or didn’t even write code suddenly started making their way into workable software. For many of these people, who could have been making slide decks or mockups from Figma, a barely working prototype was too much, and these programming tools have proven to be more than capable of building barely working prototypes. However, Vibe coding comes with risks, both in terms of the problems that bad code can cause and the risks you take by giving these tools access to your computer and data. It’s one thing to trust a system when you can verify its output, but it’s another thing to do so when you can’t speak its language.

The crisis for software developers is just beginning. Companies across Silicon Valley are laying off employees by the thousands, usually citing artificial intelligence as the reason. “A much smaller team, using the tools we build, can do more and do it better,” Block CEO Jack Dorsey He wrote in a note Announcing the layoff of 40 percent of the company. “The capabilities of intelligence tools are multiplying faster every week.” In the case of Block and many others, AI is likely, at least in part, just a cover for pandemic-era overhiring, but the tech industry is clearly relying on AI as a way to boost productivity — and reduce headcount.

As AI coding tools continue to improve, they may also reshape the rest of the software business. Why pay a fortune for someone else’s software when Cloudcode can build it for you, exactly the way you want it? Some call this SaaSpocalypsepredicting a fundamental rethinking of the way we value software. Others believe we deserve a new generation of successful startups, offering AI-driven ways of doing everything. Still others think it’s overrated and that Salesforce will be fine. Whatever the outcome, the software industry, which has grown to unimaginable heights and valuations, seems to many to suddenly stand on shaky ground.

On the other end of the developer spectrum are active programmers. For most people, even the simplest current AI coding tools are too much. They make you read the code; They require terminal access; They ask a lot of questions and no one is expected to know how to answer them. AI coding still comes with a lot of bugs, big privacy questions, and plenty of ways bad actors can exploit both.

With products like Claude Cowork, Anthropic set out to see if it could make Claude Code’s technology more intuitive and less intimidating — just give it access to a bunch of files on your computer and let it get to work. Products like Perplexity Computer are exploring whether people could give MBAs access to everything on their devices, allowing AI tools to organize files, respond to messages, and even buy things on their behalf. The underlying technology is starting to work, but it’s not at all clear how people are supposed to use it, and whether they’ll even want to.