The bottleneck in software development isn't writing code anymore—it's waiting for someone else to read it.
The Summary
- Ramp engineers now use Codex powered by GPT-5.5 to review code, cutting feedback cycles from hours to minutes
- The shift isn't just about speed—it's about freeing senior engineers from review grunt work while maintaining quality standards
- This marks a clear inflection point: code review, long considered too nuanced for automation, is becoming an agent task
The Signal
Ramp's engineering team has turned Codex with GPT-5.5 into their first-pass code reviewer. The system catches bugs, flags style inconsistencies, and suggests optimizations before human eyes ever touch the diff. The feedback loop that used to take hours—waiting for a senior engineer to context-switch, load the codebase into their head, and write thoughtful comments—now happens in minutes.
The economics are brutal and obvious. A senior engineer costs $200K-plus per year. Code review is necessary but doesn't ship features. If you can automate even 60% of review work, you've just made every engineer on your team 15-20% more productive. Ramp isn't using AI to replace engineers. They're using it to let engineers do more engineering.
"Code review was the last place we expected AI to work—too much context, too much judgment. Turns out we were wrong."
What makes this different from earlier attempts at automated code review:
- GPT-5.5 understands business logic context, not just syntax
- It can reference Ramp's internal documentation and past decisions
- It learns what "good" looks like from the codebase itself
The model doesn't just lint. It reasons about whether a function is too complex, whether an API contract makes sense, whether error handling covers edge cases. The kinds of things that used to require a human with institutional knowledge.
The Implication
If Ramp is doing this in production, so are dozens of other engineering teams who just haven't written the blog post yet. The next wave won't be about replacing engineers—it will be about replacing the coordination overhead that keeps them from building. Code review today, sprint planning tomorrow, incident response the day after. Watch for engineering productivity metrics to decouple from headcount growth over the next 18 months.