An autonomous Claude Code agent ships 340+ production deploys behind a three-stage gate it cannot edit
The author runs a "fully autonomous coding agent (built on Claude Code)" that plans, implements, opens PRs and merges them if a verifier sub-agent approves. Asked to "reduce p95 latency on the cart summary endpoint", it added a cache in front of a pricing lookup. Tests passed "because the tests mocked the pricing service", but "the cache key didn't include the currency", and "Customers in the EU saw USD prices for 11 minutes until an alert fired and a teammate rolled back by hand." The author then built a three-stage deploy gate: a pre-flight contract the agent writes into the PR body, a 5% canary for 10 minutes with hard metric thresholds, and an automatic rollback the agent cannot override. The gate lives in a separate repository the agent can read but not push to, and each rollback becomes a task asking the agent why tests passed.
The numbers (as reported)
| Deploys since the gate | "340+ production deploys with zero human-paged incidents" over six months |
|---|---|
| Rollbacks | "23 rollbacks into 23 closed test gaps" |
| Original incident | EU customers saw USD prices for 11 minutes |
| Canary | 5% of traffic for 10 minutes, hard thresholds against a live baseline |
| Pre-flight failures | "maybe one PR in twelve", mostly under-declared blast radius |
| False breaches | 4 in six months, all during one upstream payment provider incident |
| Cost per deploy | "about 12 minutes of wall-clock time and a few cents of metrics queries" |
| Gate size | "maybe 400 lines total" |
Facts
- Company / subject
- Yureki Lab (dev.to author)
- Tools
- Claude Code
- Codebase
- brownfield
- Work type
- Feature
- Stack
- Python 3.13, Node.js 22.x
- Published
- 2026-10-01
- Last checked
- 2026-10-21
What the source claims Practitioner (own blog/paper)
Claimed by the builder, writing about their own work. Not checked independently.
- 340+ production deploys over six months with zero human-paged incidents 340+ deploys, 0 pages
“Six months later it has shipped 340+ production deploys with zero human-paged incidents.”
- The gate costs about 12 minutes of wall-clock time and a few cents per deploy ~12 min per deploy
“The extra cost per deploy is about 12 minutes of wall-clock time and a few cents of metrics queries.”
Caveats
Self-reported by a single pseudonymous author on dev.to about one team and one system. The company, codebase size and traffic are not named, and none of the numbers are independently checked. The author says every incident the gate caught was a wrong answer ("None of them produced a single 5xx"); commenters point out a 5% canary over 10 minutes may miss low-volume regressions, and that the canary thresholds come from the deploy block the agent writes, so the agent still chooses how strict its own check is. No figures are given for the agent itself (token cost, time per task) or for how long the gate took to build.
More stories by tool: Claude Code
Related topics: Brownfield codebases