How to maintain a codebase – when everyone can code
summary
Vercel demonstrates a concrete pipeline: observability as machine input where logs, metrics, traces, feature‑flag telemetry and experiment results feed JSON blobs to agents that trigger on anomalies, generate code fixes, open preview PRs and run fixes in duplicate production. Humans remain for PR review and merge while agents handle investigation and patching. The stack uses session replays, full logging (no sampling) and feature flags to decouple release from deployment enabling continuous preview deploys and automatic repair agents.
AI-written summary · about
chapters (8)
- 0:00Introduction
- 1:29AI transformation at Teada Pharma
- 3:40Human PRs to AI fixes
- 5:27Anomaly fixes with AI agents
- 6:51PR review and merge queues
- 8:17Guardrails for AI maintenance
- 8:59AI development principles
- 11:51Fully automated deployments
related talks
36mAgentic CI/CD: from Pull Request to Production without the Paper Cuts
42mRevisit Old Problems with New Eyes!
59mEverything as a Code
50mPanel: Beyond The Vibe: Code Quality First
22mCTO panel: Internal apps and agents in production
17mCooking with Agents in VS Code
35mTesting in Production: there is a better way
27mConcept to Code to Runtime - Simplify, Enable, Exploit