François Chollet says agents now write nearly all code yet engineers work harder and company results are not budging
Some observations about the "total" automation of software engineering: 1. There are more software engineers now than in 2025, and demand is growing. 2. Software engineers have never worked harder than they do now. 3. They largely don't write code anymore, agents write ~100% of the code. 4. The bottom line of the companies that employs these engineers and agents (i.e. the terminal purpose of all this code) is typically not budging as a result of AI adoption. This is why you have to think in terms of systems. Automating one aspect of a system creates its own friction and moves the bottlenecks elsewhere.
Context
Chollet's point is that automating coding moves bottlenecks elsewhere, which challenges the idea that AI agents will quickly replace software engineers.
This is an opinion post from Chollet, the creator of the Keras library and cofounder of the ARC Prize. He gives no data source for his observations.
Deep dive
On October 17, François Chollet posted four observations about what he called the total automation of software engineering.
He wrote that there are more software engineers now than in 2025 and demand is growing, and that software engineers have never worked harder than they do now. He wrote that they largely do not write code anymore, with agents writing about 100% of it.
His fourth point was that the bottom line of the companies employing these engineers and agents, which he called the terminal purpose of all this code, is typically not budging as a result of AI adoption.
He concluded that this is why one has to think in terms of systems, because automating one part of a system creates its own friction and moves the bottlenecks elsewhere.
The post drew more than 200 replies. Chollet did not cite figures for engineering employment or company results, so the observations are his own read of the industry. His view echoes a debate covered elsewhere in these decks, including Anthropic's statement that Claude writes about 90% of Claude Code's own code.
What to expect next
Watch for data that tests Chollet's claims, such as software engineering job postings, company earnings that cite AI productivity, and surveys on how much code agents write.
Also watch where the new bottlenecks he describes appear, such as code review, testing, product decisions and coordination, and whether tools aimed at those stages gain traction. Responses from AI lab leaders who predict faster gains will show how far apart the two camps remain.