How Ramp engineers work with AI agents at every step
- Source
- youtube.com
- Author
- Claude
- Date
Context
In a Claude-hosted discussion, Ramp engineers describe using Fable, a model from Anthropic, on hard problems in their large Python codebase. One says it made merged progress on import cycles and on making the app start lazily. He adds that no earlier model had passed every one of the tests he gives each new release, and that Fable did.
They distinguish two tools. A loop is repetitive work, such as babysitting pull requests or deleting dead code daily. A dynamic workflow has the model orchestrate many sub-agents when steps are unknown, which one speaker likens to extra . For CI, the speaker says an profiled the system, landed changes, scheduled itself to check real production data a day later, and repeated this, taking median CI time from about 18 minutes to about 6. That figure is the speaker's recollection, not a published measurement.
- test-time compute — Spending more computation when the model answers — thinking longer, trying multiple attempts — to buy accuracy without training a bigger model.
- AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
Checking sign-in…
Loading comments…





