Everyone says AI will 10x your dev team. But what does the data actually say?

In this episode of Fractional, I walk through what the data says about AI and developer productivity—and why the answer is more nuanced than the usual 10x claims.

Our Stanford team built a model trained on expert code reviews to study the impact of AI coding tools. We found a productivity uplift of around 20% at best, with wide variation across programming languages, codebase complexity, and context.

We discuss:

• What productivity really means in software engineering

• Why legacy codebases often get worse with AI

• How LLMs quietly replace pair programming for junior devs

• What makes engineering teams more effective in reality—not theory

• Lessons from building Crunchyroll’s global tech org

This isn’t an AI hypefest. It’s a reality check.

https://www.swarm.work/podcasts/simon-obstbaum