Skip to content

That's just a taste — this channel full of gems!

Visit channel
User avatar
Software Development
October 26880

AI Coding in 2026 - productivity gains or another technical debt?

AI coding tools can enhance certain aspects of the development process. But what happens, after generating code?

We covered last several studies pertaining to Cursor, AI coding assistants and true software projects.

Speedup is real, but temporary

In 2026 researchers at Carnegie Mellon focused on 806 Cursor-enabled projects on GitHub to study their impact and how they differed from similar projects that didn't use Cursor.

So as you can see in the results below, the number of code added did indeed spike by 3-5 times during the first month after adoption - seeing the impact of the adoption - but then the increase had diminished by the second month.

Meanwhile, the generation of static-analysis warnings grew by 30% and the complexity of the code grew by 41% all at once. This technical debt can slow down development as the codebase becomes harder to maintain.

As more code is introduced, the review burden can increase

The results from DORA's 2026 study were taken from slightly more than 1,100 open-ended answers gathered from a cross-section of Google software engineers in the third quarter of 2025.

The pattern was clear: engineers spent more time on verification, covered for a lot of the hallucination issues and had to monitor for the technical debt.

Evidence of extra rework

Another survey of open-source projects took place, this time after GitHub Copilot was adopted, and this study found that rates of productivity also rose but so did the need to rework AI-generated code.

Notably, experienced core developers also faced a greater review burden: they reviewed 6.5% more code while writing 19% less code themselves.

But times are changing

In an early randomized study in 2025, METR discovered that experienced open-source developers took 19% longer to complete tasks when using AI tools.

However, by the early 2026, newer data collected by METR suggests that AI tools might be helping developers more, but the study could not reliably measure the productivity improvement because of selection bias and other measurement issues.

METR could not confidently estimate the actual productivity boost as more developers began relying on AI tools.


So, it obviously appears that there is plenty of low-value coding work to be eliminated by AI already. But as the volume of code increases beyond a team's capacity to review, the time saved upfront can easily be lost later.

So, here's the main question for you guys:

Where do you see the greatest AI gains today? and at what stages does AI create even more extra work?

#softwaredevelopment#aiindevelopment#aicoding#aicoding2026
Like
Dislike

78

28

Add a comment...