Habr AI: METR and Google Cloud See No Promised Developer Acceleration from AI

On Habr AI, an analysis of several studies was published about the impact of AI tools on developers' work. The market-unfriendly conclusion: developers often feel they work faster, but measurable metrics don't yet confirm the promised acceleration.

Where Growth Is Not Visible

One of the key arguments in the article is connected to an analysis of the Python ecosystem. The author cites Answer.AI's breakdown, where they researched PyPI dynamics in search of traces of "explosive" productivity.

If AI were truly accelerating development radically, it should have shown up in the number of new packages, release frequency, and the overall volume of published code. But the picture turned out much more modest: no notable spike in new packages, and the growth in update frequency started back in 2019 and is probably tied more to CI/CD practices than to generative models. An interesting detail is that the growth in activity is most visible in AI projects.

Such packages update more frequently, but the author interprets this not as a universal tool effect, but as a consequence of hype and the influx of investment into the segment. Similar logic applies to GitHub data: according to the review, there's no mass surge in new repositories either, though that could have been a simple indicator that launching side projects became considerably easier. The verdict is harsh: AI helps create impressive prototypes, but doesn't eliminate the real bottlenecks of development and product launch.

What Research Shows

The most striking gap is visible between team perceptions and operational metrics. The article cites Google Cloud's report on GenAI's impact on software development: 75% of developers say that AI gives them a sense of greater productivity. But with a 25% adoption rate of such tools, delivery throughput drops by 1.5%, and delivery stability drops by 7.2%. In other words, working with AI feels subjectively more pleasant, but doesn't necessarily mean more useful output.

• 75% of developers experience improved productivity
• delivery throughput decreases by 1.5%
• delivery stability falls by 7.2%
• activity surge is most noticeable in the AI segment

Even harsher is the METR research cited by the author. In it, experienced developers expected that AI tools like Cursor and Claude would speed up their work by roughly 20%. In reality, it turned out the opposite: task completion slowed down by about the same 20%. The explanation seems plausible: engineers type less code by hand, but spend more time checking, making fixes, waiting for model responses, and running tests again. In other words, some mechanical routine disappears, but a new layer of quality control takes its place.

"Today we have facts that AI makes many of us less, not more productive."

This gap between expectation and reality matters not just for engineers, but for business. The article also cites the example of Notion: after adding AI features, the product's margins, according to the company's CEO, dropped from 90% to 80%. The logic is clear: the market pushes AI integration into almost everything, but additional costs for inference and infrastructure don't guarantee either audience or revenue growth. For companies, this means that implementing AI in development and product cannot be evaluated by wow factor — metrics of speed, stability, and economics are needed.

What This Means

Habr AI's analysis effectively cools expectations around "10x" accelerations in programming. At the current stage, AI coding tends to improve the work experience and speed up certain pieces of the process, rather than guarantee an increase in team or business productivity. For managers, the conclusion is straightforward: before rolling out such tools at scale, measure not developer sentiment, but cycle time, delivery quality, and the final cost of changes.