Getting AI and automation into production, and keeping them alive there. Notes on AI agents, automation and engineering leadership by Miguel Côrte.https://mcorte.com/https://mcorte.com/posts/your-agent-isnt-bad-at-coding-your-spec-is/https://mcorte.com/posts/your-agent-isnt-bad-at-coding-your-spec-is/AI coding agents do exactly what you ask and guess the rest. A good spec isn't a long document. It's five short things: the outcome, the context it can't guess, the limits, how you'll check it, and when it should stop and ask.Sun, 01 Nov 2026 00:00:00 GMTAI AgentsEngineering LeadershipAI-Assisted CodingSpecificationsCode Reviewhttps://mcorte.com/posts/anthropic-outgrew-git-your-bottleneck-is-next/https://mcorte.com/posts/anthropic-outgrew-git-your-bottleneck-is-next/Anthropic moved its main codebase off Git after daily merges went from about 2,500 to over 30,000 in four months. You're not Anthropic. But the lesson applies to every team using AI to write code: the slow part is no longer writing it.Mon, 26 Oct 2026 00:00:00 GMTAI-Assisted CodingEngineering LeadershipDeveloper ProductivityCode ReviewCI/CDhttps://mcorte.com/posts/your-ai-needs-a-plan-b/https://mcorte.com/posts/your-ai-needs-a-plan-b/Every AI provider has bad days. Most companies never decided what their business does on one. Four questions to answer before the next outage, in plain English.Thu, 15 Oct 2026 00:00:00 GMTAIAI AgentsReliabilityOperationsEngineering Leadershiphttps://mcorte.com/posts/ai-layer-is-the-last-thing-you-build/https://mcorte.com/posts/ai-layer-is-the-last-thing-you-build/Everyone wants to start with the AI. It works when it goes on last. The three floors to build under it first, in plain English, and a quick test for where your company is.Tue, 06 Oct 2026 00:00:00 GMTAIDataAutomationOperationsEngineering Leadershiphttps://mcorte.com/posts/jev-confidence-is-not-a-decision/https://mcorte.com/posts/jev-confidence-is-not-a-decision/Jev answers like a form and tells you how sure it is. How it works in plain English, how it compares to LLMs like ChatGPT or Claude, when to use each or both, and the four questions no model answers for you.Mon, 28 Sep 2026 00:00:00 GMTAIAutomationJevLLMDecision MakingAI Governancehttps://mcorte.com/posts/debug-agents-you-cant-reproduce/https://mcorte.com/posts/debug-agents-you-cant-reproduce/Teams instrument their agents properly and still can't debug them. The problem isn't missing traces - it's that a trace tells you what happened, and an agent bug is a question about why it chose that. The four levels of traceability, and the one most teams never reach.Tue, 01 Sep 2026 00:00:00 GMTAI AgentsObservabilityOpenTelemetryProductionDebugginghttps://mcorte.com/posts/em-job-when-agents-write-code/https://mcorte.com/posts/em-job-when-agents-write-code/When agents write most of the code, the engineering manager job doesn't shrink - it moves one station down the line, to review, coherence, and restraint. Here's what actually changes, and what to do in the next 30 days.Mon, 24 Aug 2026 00:00:00 GMTEngineering LeadershipAI AgentsEngineering ManagementCode Reviewhttps://mcorte.com/posts/building-ai-agents-that-actually-work/https://mcorte.com/posts/building-ai-agents-that-actually-work/AI agents don't fail because of bad models — they fail because of weak foundations. Learn the four pillars (data, processes, ROI, and architecture) needed to build scalable AI systems that actually work in production.Wed, 08 Apr 2026 00:00:00 GMTAI AgentsArtificial IntelligenceAutomationAI Architecture