AI Agents

This is part 4 of the Agent production system series. Previous: Docker homelab ops. Constellation: the whole agent production map — especially Orchestrator, Eval gates, and Human approval. I am a former Senior Technical Program Manager at Xbox/Microsoft. That sentence is past-tense on purpose — and it still explains how I design agent systems. People hear “TPM” and picture a calendar full of status meetings. At its best, platform TPM work is the opposite of that caricature: critical path, risk, multi-team alignment, and SLA performance under real blast radius. Someone has to know what is actually on the path to “done,” what can kill the ship window, which teams are blocked on which contract, and when a green dashboard is lying. That is the job. ...

This is part 3 of the Agent production system series. Previous: Human approval. Constellation nodes: Docker host, Azure / ADO, Dashboards. Words I use below Homelab — the small, always-on computer setup in my house that runs real services for the household and for agents, not a laptop demo that dies when I close the lid Container — a boxed-up app (and its dependencies) that runs the same way on different machines; if one misbehaves, it fails in a smaller blast radius than “the whole computer” Compose — a config file that says which containers to start, how they connect, and what secrets they get, so you recreate the world from files instead of folklore OrbStack — the Docker runtime I use on the Mac Mini (think “Docker Desktop, but lighter on Apple silicon”) so containers actually run on that host Reverse proxy — one front door for many services; you hit one place instead of a pile of bookmarked :port URLs VPN gateway — a container that holds the download traffic behind a tunnel; if the tunnel drops, that traffic drops with it instead of leaking out the front door Watchtower — a helper that updates container images on a schedule, but only for services I’ve labeled as allowed to drift Agent-operated ops — agents can propose and run routine operations under the same gates as code — not root plus good intentions Those eight words are the toolkit. Now the point. ...

This is part 2 of the Agent production system series. Previous: Eval gates are not optional theater. Constellation node: Human approval. Words I use below Agent — a program that writes code or words for you, and can use tools, in a chat or a job queue Pull request (PR) — a proposed change to a code repository; someone (or something) asks to land a diff, and reviewers say yes or no Merge — accepting that pull request into the main line of code, so the change becomes part of what ships Force-push — overwriting history on a branch; useful in cleanup, dangerous when it erases work other people (or agents) already built on Irreversible action — something hard or expensive to undo (delete data, rotate a secret into the void, change production infrastructure, send a public message you cannot unsend) Approval identity — a separate login or app credential whose only job is to approve and merge a reviewed pull request (not the same credential the agent uses to write code day to day) Decision brief — a short package an agent hands a human: what it wants to do, why the gate fired, what it already checked, and what it needs you to decide — not a guilt trip and not a demand Action class — a bucket of similar actions that share one autonomy rule; routine merges after checks can be auto, irreversible carve-outs always pause Those eight words are the toolkit. Now the point. ...

This is part 1 of the Agent production system series. Start with the system map if you haven’t read it yet. On the interactive diagram: Eval gates. Demo agents look smart until they touch a real repo. In a demo, the agent writes the code, the code runs once, everyone claps. Nobody checks what happens the second time, or the tenth, or the time the agent decides the fastest way past a failing test is to delete the test. Then you learn the hard lesson: intelligence without a gate is just a faster way to ship a bad change. ...

Most “AI agent” demos stop at a chat box. What I care about is the opposite: a production system with routing, tools, gates, hosts, and a human who still owns irreversible decisions. This post is the written version of the interactive map on my About page — how the fleet is wired, what each layer is for, and what I refuse to automate. The mental model Think factory floor, not chatbot: ...