At SimpleKYC , the #devops_requests Slack channel is where every team goes when something breaks — or when they think something might be about to break. Pod stuck in CrashLoopBackOff? Pipeline failed? 502 from a service that was fine yesterday? It all lands there. A colleague analyzed the channel and found that 70% of requests could be self-resolved by querying the same tools the DevOps team uses: kubectl, Azure CLI, Bitbucket pipelines, health checks. The knowledge wasn't secret — it was just scattered across terminals, runbooks, and people's heads. We were already using Claude Code to investigate issues ourselves. The next step was obvious: put an agent in the channel that could do the same thing, for everyone, automatically. The Agent Architecture The bot is a generic LLM agent — not a collection of pre-built workflows. When someone mentions @DevOpsBot in a thread, GPT-5.4 (via Azure OpenAI) decides which tools to call, in what order, with what arguments.…