Felix Geelhaar — Field NotesLong-form essays on shipping software with AI agents. Plan-of-record, evidence layers, agent runtimes, and the anti-patterns engineering teams ship when a model is in the loop.https://felixgeelhaar.de/en21 patterns I keep seeing in teams shipping with AI coding agents.https://felixgeelhaar.de/writing/21-patterns-shipping-with-agents/https://felixgeelhaar.de/writing/21-patterns-shipping-with-agents/A partial, in-progress list of the recurring dysfunctions, anti-patterns, and quiet wins from teams that ship software with AI coding agents day-to-day. Working notes, not a framework.Sun, 17 May 2026 00:00:00 GMTDeploy is not release. Release is not launch.https://felixgeelhaar.de/writing/deploy-release-launch/https://felixgeelhaar.de/writing/deploy-release-launch/Three different verbs, three different owners, three different days. Decoupling them is the only honest way to ship software in 2026.Sun, 17 May 2026 00:00:00 GMTCloud coding agents read your repo. Yours doesn't have to.https://felixgeelhaar.de/writing/local-first-coding-agents/https://felixgeelhaar.de/writing/local-first-coding-agents/Why local-first is the right default for AI coding agents — and what choosing it actually costs.Sun, 17 May 2026 00:00:00 GMTAgents forget two ways. Most memory tools only fix one.https://felixgeelhaar.de/writing/plan-and-evidence-two-memories/https://felixgeelhaar.de/writing/plan-and-evidence-two-memories/Plan-of-record and evidence layer are different primitives, solving different forms of forgetting. Most agent runtimes collapse them into one — or skip both — and pay the cost in the wrong direction.Sun, 17 May 2026 00:00:00 GMTYour AI subscription has a rate limit. Your agent doesn't know about it.https://felixgeelhaar.de/writing/agent-rate-limit-headroom/https://felixgeelhaar.de/writing/agent-rate-limit-headroom/Flat-rate AI plans hide a cliff. The agent burns through your weekly Claude Max window, gets throttled mid-task, loses an hour of context. The fix isn't a different model — it's giving the agent visibility into the cap before it hits.Sat, 16 May 2026 00:00:00 GMTAgents forget. The plan should remember.https://felixgeelhaar.de/writing/agents-forget-plan-remembers/https://felixgeelhaar.de/writing/agents-forget-plan-remembers/AI coding agents lose context every session. The work doesn't. What survives the reset is the plan-of-record — and most teams don't have one.Sat, 16 May 2026 00:00:00 GMTYour AI's memory is someone else's database.https://felixgeelhaar.de/writing/ai-memory-yours/https://felixgeelhaar.de/writing/ai-memory-yours/Hosted memory layers and vector DBs hand the user's preferences, conversation history, and inferred claims to a vendor's infrastructure. The feature ships; the ownership doesn't.Tue, 12 May 2026 00:00:00 GMTLLM calls broke your resilience playbook. Add these back.https://felixgeelhaar.de/writing/llm-resilience-playbook/https://felixgeelhaar.de/writing/llm-resilience-playbook/Retry, circuit breaker, timeout — the patterns that kept services up for two decades don't cover the failure modes of services that call LLMs and tools. Here's what's missing.Tue, 12 May 2026 00:00:00 GMTThree MCP tools in, you've started rebuilding Gin. Stop.https://felixgeelhaar.de/writing/mcp-server-application/https://felixgeelhaar.de/writing/mcp-server-application/Hand-rolling MCP servers on raw SDKs means re-solving input decoding, validation, schema generation, errors, middleware, and transports — five times, in five repos, each slightly wrong. The protocol is the easy part.Tue, 12 May 2026 00:00:00 GMTYour security scanner can't see your AI code.https://felixgeelhaar.de/writing/security-scanner-ai-blind/https://felixgeelhaar.de/writing/security-scanner-ai-blind/SAST tools stopped evolving at HTTP request validation. AI features live on top of HTTP, but the failure modes — prompt injection, embedding leakage, agent over-privilege, MCP hardening — happen at the LLM call site, in places traditional scanners don't look.Sun, 10 May 2026 00:00:00 GMTSwitch statements are a state machine in denial.https://felixgeelhaar.de/writing/statecharts-over-switch/https://felixgeelhaar.de/writing/statecharts-over-switch/Every team modeling order lifecycles, payment sagas, incident workflows — and now AI agent runtimes — ends up with the same 400-line switch block, three boolean flags, and a Slack channel where the bugs surface. The fix is to make the FSM a first-class artifact.Sun, 10 May 2026 00:00:00 GMTThe Logger's Trilemma was a sampling artifact.https://felixgeelhaar.de/writing/loggers-trilemma/https://felixgeelhaar.de/writing/loggers-trilemma/The Logger's Trilemma says you can have two of speed, developer experience, and observability. The trilemma was a sampling problem; modern slog handlers can ship all three.Sat, 09 May 2026 00:00:00 GMTAn agent runtime is five primitives. Most fake at least three.https://felixgeelhaar.de/writing/cognitive-stack/https://felixgeelhaar.de/writing/cognitive-stack/Memory, time perception, commitment tracking, typed action, control plane — the missing infrastructure beneath the model. Most agent frameworks glue substitutes together and call it done.Sun, 03 May 2026 00:00:00 GMTEvery 200-team rollout ends in a Google Sheet. Fix it in Jira.https://felixgeelhaar.de/writing/jira-campaign-orchestration/https://felixgeelhaar.de/writing/jira-campaign-orchestration/Security patches, migrations, compliance audits — large-scale initiatives are fan-out problems Jira's defaults don't solve. Teams reach for spreadsheets and lose every benefit of the workflow they left.Sun, 01 Feb 2026 00:00:00 GMT