Single‑Player PRs

The AI‑Powered Shift

One of the biggest surprises I've had this year was learning that some startups, let a subset of trusted engineers merge code directly into non‑sensitive isolated codeareas.

On my first day that felt reckless. My gut tightened every time I hit Merge without a second human. A week later the discomfort faded to curiosity, then to the quieter question: why didn't we do this sooner?

After a career in early‑stage companies + my own Bountycaster I've learned a humbling truth: 90 % of what you ship lands with a shrug. The hit rate is brutally unintuitive, so the only lever you really control is how many shots you take and how quickly you iterate.

AI is a massive game changer. We now have large context windows that let tools like Claude Code see your entire codebase, not just a diff. I run /review on 90% of my changes—it's become muscle memory. The AI catches naming inconsistencies across files, suggests better abstractions I missed, and answers follow-up questions like "would this break if we scale to 10k users?" Before, human reviewers would skim diffs and rubber-stamp. Now I get comprehensive feedback in seconds: type-check, unit-test suggestions, alternative implementations—all before I finish my espresso. Programming turns into a single-player sport, and teams that adopt it gain an advantage over rivals still waiting on human approvals.

What we gain

• Flow over friction. Ship when you are the most productive (for me it's early mornings), not when a reviewer wakes up.
• True ownership. Chase the highest‑leverage work instead of camping in a PR queue.

Where to draw the line

Not all code is created equal. We self-merge for feature flags, UI tweaks, internal tools, and new endpoints that don't touch existing data. But we still require reviews for database migrations, payment logic, authentication flows, and core infrastructure. The heuristic is simple: if a bug would page someone at 3 AM or cost money, get eyes on it.

What we lose

• Fewer organic mentoring moments. The best learning often happens in PR comments. We're experimenting with weekly code review sessions where people can voluntarily share interesting problems they solved.
• Greater risk of duplicate work. Without PR visibility, two engineers might solve the same problem differently. We combat this with a #shipped channel where everyone posts what they merged.

I expect this to become the default for every company still searching for product‑market fit. Startups that ignore this golden opportunity may simply run out of time.