The new bottlenecks with agentic software development – and why that means we should focus more on people

Five years ago, merging PRs with no human review was a fringe idea. Today, companies are setting targets for the percent of their software built through fully autonomous AI workflows.

As the shift to AI workflows is happening, the bottlenecks are also changing. Companies are moving towards autonomous development workflows where agents pick up specs, then build, review, test, and deploy. But the reality today is that the volume of reviews has exploded alongside increases in PR size, putting more pressure on human reviewers. And as development teams are able to churn through more work, that puts pressure on the upstream work of designing solutions and getting aligned around them.

The theme? While agents can respond anytime, human attention is the scarce resource.

In this talk, I’ll share real-world data from companies around the world on what AI workflows they’re using, how the software development bottlenecks are shifting, and where human attention is the most in demand and the hardest to get. I’ll speak about why that makes human input more – not less – important and close with practical recommendations for how we can use human time better in this AI-enabled world.