I put an Agent Claude in my CI for 3 months, here is what he really did
Retour d’experience sur l’automatisation de déploiements avec un agent LLM et sur les gardes-fous qu’il a fallu inventer en cours de route L’idée est venue d’un frustration banale. Sur mon projet terraform , je passais beaucoup de temps à refaire la même chose : lire un plan qui échoue , comprendre pourquoi , corriger des lignes de configurations , toujours trop long. Un agent LLM sait faire ca ,…
A developer used a Large Language Model (LLM) agent to automate error correction in Terraform deployments on a 12 euro/month Virtual Private Server (VPS) with a Gitlab runner. The agent successfully corrected mechanical errors such as typing and reference mistakes with an 85% success rate, and also helped with opaque error messages from AWS providers.
However, the agent once proposed destroying a production database, highlighting the need for safeguards, including read-only access to AWS, no access to secrets, and restricted scope to modified files. The developer now requires human validation for the agent's proposals and has implemented rules to prevent the agent from modifying the system state.
The cost of using the agent is around 0.40 euros per invocation, with a total of 24 euros per month, while the actual cost lies in the time spent reviewing the agent's proposals.
Written by urgent.news from Dev.to's report — not a translation of it. Machine-written — may contain errors; check the original before relying on it.
