Thorsten Ball says drop branches, push straight to main
The argument is that agents absorb the feature flags and incremental migrations that made working on main expensive.
Thorsten Ball followed his case against pull requests with a harder one, that you do not need branches either. His argument is that trunk-based development, continuously integrating small commits into main, always worked, and that the thing stopping people was the amount of fiddly work it required.
Nah, I can merge nearly anything continuously in small commits to main.
Don't just use agents to replace your typing.
His list of that work is specific. Make the first small change compile and pass tests, merge it, ship it, watch it. Push still-unused dead code, add a feature flag, add a config variable, keep two configurations alive for a day, migrate call sites one at a time. Ball says that is the reason people reached for a big fat branch instead, and that agents are good at exactly those chores.
He is also dismissive of the usual objection, the change that supposedly cannot go on main because it has a database migration or reshapes the architecture. Add the column first, merge, deploy, then add the helpers, then migrate the old pattern to the new one one by one. In his framing the genuinely unsplittable change is 0.5% of cases, not 50%.
The pushback
Ada Lundhe replied with the short version of every compliance team's objection, "SOC2. Fedramp. Working with more than two other people." Ball answered that his team has 21 people, has SOC2, and pushes to main.
Jordan Webster described the other end of the scale from experience. His org had Diffs, which work like a PR, and Stacks, which work like a branch with no ceremony, where you make commits on main and that is your stack. Pulling always rebased your stack on top of main, and he says thousands of developers were all working on main.
For working engineers, the practice here is decades old and Ball says so himself, pointing to Ruby and to someone who ran it in an org of 400 engineers. The only new claim is that agents make the incremental path cheap enough to be the default, and nobody has put a number on that yet. Ball's actual pitch is broader than branching anyway, which is to stop using agents as faster typing and re-examine the workflows you only adopted because writing code was expensive.
Thousands of developers, all working on main.
Have the good opinion, not the loud one
One email a day. Under three minutes. What working engineers found out about coding with AI.