<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>Bearing Out</title><description>Jeff Breece on systems architecture, engineering, the homelab, the trail and the kitchen.</description><link>https://bearingout.com/</link><item><title>The Agent Keeps a Ledger. Someone Else Has to Read It.</title><link>https://bearingout.com/engineering/ai/2026/10/10/the-agent-keeps-a-ledger-someone-else-has-to-read-it.html</link><guid isPermaLink="true">https://bearingout.com/engineering/ai/2026/10/10/the-agent-keeps-a-ledger-someone-else-has-to-read-it.html</guid><description>A follow-up on agent-ready delivery: the change log my coding agent writes on every turn, the drift check that reads it, and why the research suggests that check should come from somewhere other than the agent that did the work.</description><pubDate>Sat, 10 Oct 2026 16:00:00 GMT</pubDate><category>agentic-ai</category><category>architecture</category><category>devops</category><category>governance</category><category>drift-detection</category></item><item><title>I Pointed an AI Coach at My Own Agent Work</title><link>https://bearingout.com/engineering/ai/2026/10/04/i-pointed-an-ai-coach-at-my-own-agent-work.html</link><guid isPermaLink="true">https://bearingout.com/engineering/ai/2026/10/04/i-pointed-an-ai-coach-at-my-own-agent-work.html</guid><description>I ran Microsoft&apos;s open-source AI Engineer Coach over 90 days of my own agent sessions. My prompts were mostly fine. The structure around them usually wasn&apos;t, some of the scoring rules were built for a different way of working, and the local half of my stack never showed up at all.</description><pubDate>Sun, 04 Oct 2026 11:00:00 GMT</pubDate><category>agentic-ai</category><category>coding-agents</category><category>local-ai</category><category>developer-productivity</category><category>standards</category><category>homelab</category><category>observability</category></item><item><title>The Agent Wasn&apos;t the Innovation. The Standards Were.</title><link>https://bearingout.com/engineering/ai/2026/09/27/the-agent-wasnt-the-innovation-the-standards-were.html</link><guid isPermaLink="true">https://bearingout.com/engineering/ai/2026/09/27/the-agent-wasnt-the-innovation-the-standards-were.html</guid><description>Two years building a small cloud delivery platform, and what a coding agent did and did not change. The speed came from standards written down the year before; the agent mostly used them.</description><pubDate>Sun, 27 Sep 2026 16:00:00 GMT</pubDate><category>agentic-ai</category><category>architecture</category><category>devops</category><category>gitops</category><category>adr</category><category>delivery</category><category>leadership</category></item><item><title>I Striped Two Drives for Space I Never Used</title><link>https://bearingout.com/project/2026/09/20/raid0-to-raid1-storage-greed.html</link><guid isPermaLink="true">https://bearingout.com/project/2026/09/20/raid0-to-raid1-storage-greed.html</guid><description>My homelab bulk store ran for months as a two-disk RAID 0 stripe for capacity I never touch. Here is why I converted it to a RAID 1 mirror, and why I am not fooling myself that the mirror is a backup.</description><pubDate>Sun, 20 Sep 2026 20:03:00 GMT</pubDate><category>homelab</category><category>storage</category><category>raid</category><category>backup</category><category>disaster-recovery</category><category>data</category></item><item><title>Taking the Home AI Lab Hybrid, on a Leash</title><link>https://bearingout.com/project/2026/09/14/taking-the-home-ai-lab-hybrid-on-a-leash.html</link><guid isPermaLink="true">https://bearingout.com/project/2026/09/14/taking-the-home-ai-lab-hybrid-on-a-leash.html</guid><description>How I joined a fully local AI lab to a small, budget-capped slice of Azure AI Foundry over a private tunnel, without making the cloud a dependency, and what a security audit of the result taught me.</description><pubDate>Mon, 14 Sep 2026 11:00:00 GMT</pubDate><category>homelab</category><category>azure</category><category>wireguard</category><category>agentic-ai</category><category>mcp</category><category>security</category><category>architecture</category></item><item><title>Agent of Change, Agent of Chaos</title><link>https://bearingout.com/architecture/leadership/2026/09/02/agent-of-change-agent-of-chaos.html</link><guid isPermaLink="true">https://bearingout.com/architecture/leadership/2026/09/02/agent-of-change-agent-of-chaos.html</guid><description>Solution architecture is change delivery, one solution at a time. Which face of the coin people see often has little to do with the quality of the work, and the emotional cost tends to land on one person. Here&apos;s what I&apos;ve found helps, and what I think has to be built at the org level instead.</description><pubDate>Wed, 02 Sep 2026 11:00:00 GMT</pubDate><category>architecture</category><category>change-management</category><category>leadership</category><category>governance</category><category>togaf</category></item><item><title>Putting the Lab in a Chat Window: Agent Skills, Flows, and CI/CD for AnythingLLM</title><link>https://bearingout.com/homelab/engineering/ai/2026/08/19/anythingllm-agent-toolkit.html</link><guid isPermaLink="true">https://bearingout.com/homelab/engineering/ai/2026/08/19/anythingllm-agent-toolkit.html</guid><description>Fifteen custom agent skills and five flows in AnythingLLM Desktop, the CI gate and pull-based sync that deploy them without me, and an honest accounting of the value I expect to get back.</description><pubDate>Wed, 19 Aug 2026 08:00:00 GMT</pubDate><category>homelab</category><category>resonancelab</category><category>localai</category><category>anythingllm</category><category>automation</category><category>ci-cd</category></item><item><title>The Meter Is Not the Bill</title><link>https://bearingout.com/engineering/ai/tooling/2026/08/11/the-meter-is-not-the-bill.html</link><guid isPermaLink="true">https://bearingout.com/engineering/ai/tooling/2026/08/11/the-meter-is-not-the-bill.html</guid><description>How billing actually works across Anthropic&apos;s Claude Max, OpenAI&apos;s Codex, GitHub Copilot, and Microsoft Copilot - four meters, one wall, and the local tools that read them.</description><pubDate>Tue, 11 Aug 2026 13:00:00 GMT</pubDate><category>ai-tooling</category><category>billing</category><category>claude</category><category>codex</category><category>copilot</category><category>ccusage</category><category>developer-experience</category></item><item><title>Helen Gets a Command Surface</title><link>https://bearingout.com/homelab/engineering/ai/2026/08/11/helen-voice-commands.html</link><guid isPermaLink="true">https://bearingout.com/homelab/engineering/ai/2026/08/11/helen-voice-commands.html</guid><description>Extending the lab&apos;s voice assistant from answering questions to running a fixed set of operations by voice - plus a real clock, general questions, and a menu that a test keeps honest.</description><pubDate>Tue, 11 Aug 2026 11:00:00 GMT</pubDate><category>homelab</category><category>resonancelab</category><category>localai</category><category>voice</category><category>automation</category><category>python</category></item><item><title>Seven in the Morning: What Helen Says, and Why No Model Writes It</title><link>https://bearingout.com/homelab/engineering/ai/2026/08/05/captain-awareness-reports.html</link><guid isPermaLink="true">https://bearingout.com/homelab/engineering/ai/2026/08/05/captain-awareness-reports.html</guid><description>The last post in the series. A twice-daily report on the state of the lab, assembled deterministically, fail-closed, and always saying what changed since you last heard from her - including when the answer is nothing.</description><pubDate>Wed, 05 Aug 2026 11:00:00 GMT</pubDate><category>homelab</category><category>resonancelab</category><category>localai</category><category>observability</category><category>automation</category><category>python</category></item><item><title>A Refusal a Model Can&apos;t Be Talked Out Of</title><link>https://bearingout.com/homelab/engineering/ai/2026/08/03/helen-tier-system.html</link><guid isPermaLink="true">https://bearingout.com/homelab/engineering/ai/2026/08/03/helen-tier-system.html</guid><description>Giving an assistant real authority over real machines, and then bounding it in code rather than in a prompt. On tiers enforced at the tool level, a two-attempt rule, and readings that cannot be reported without their age.</description><pubDate>Mon, 03 Aug 2026 11:00:00 GMT</pubDate><category>homelab</category><category>resonancelab</category><category>localai</category><category>safety</category><category>automation</category><category>python</category></item><item><title>How Helen Came to Be: Four Voices and a State Machine</title><link>https://bearingout.com/homelab/engineering/ai/2026/08/01/langgraph-va-graph-personas.html</link><guid isPermaLink="true">https://bearingout.com/homelab/engineering/ai/2026/08/01/langgraph-va-graph-personas.html</guid><description>The lab stopped being one assistant and became four, plus a branch that isn&apos;t a personality at all. On stickiness over topic-detection, why the house doesn&apos;t get a character, and what changes when the thing answering has a name.</description><pubDate>Sat, 01 Aug 2026 11:00:00 GMT</pubDate><category>homelab</category><category>resonancelab</category><category>langgraph</category><category>localai</category><category>personas</category><category>python</category></item><item><title>The Coach That Doesn&apos;t Use a Model</title><link>https://bearingout.com/homelab/engineering/ai/2026/07/30/health-coach-deterministic-workouts.html</link><guid isPermaLink="true">https://bearingout.com/homelab/engineering/ai/2026/07/30/health-coach-deterministic-workouts.html</guid><description>Training plans generated from equipment and rules rather than by a language model - fast, available when the GPU is busy, and identical every time. On knowing which parts of a personal system should never be improvised.</description><pubDate>Thu, 30 Jul 2026 11:00:00 GMT</pubDate><category>homelab</category><category>resonancelab</category><category>localai</category><category>health</category><category>fastapi</category><category>python</category></item><item><title>Looking Outward: A Small Data Lake for Public Sources</title><link>https://bearingout.com/homelab/engineering/data/2026/07/28/data-hub-v1-parquet-duckdb.html</link><guid isPermaLink="true">https://bearingout.com/homelab/engineering/data/2026/07/28/data-hub-v1-parquet-duckdb.html</guid><description>Earthquakes, economic series, market data, river levels and weather alerts, collected into Parquet on a home server and queried with DuckDB. On why a file format and a query engine beat a database here, and the trap of collecting more than you ask.</description><pubDate>Tue, 28 Jul 2026 11:00:00 GMT</pubDate><category>homelab</category><category>resonancelab</category><category>duckdb</category><category>parquet</category><category>dataengineering</category><category>python</category></item><item><title>Two Stages and a Refusal: Routing Inside the Memory</title><link>https://bearingout.com/homelab/engineering/ai/2026/07/26/rag-service-architecture.html</link><guid isPermaLink="true">https://bearingout.com/homelab/engineering/ai/2026/07/26/rag-service-architecture.html</guid><description>A second router, one layer down - deciding whether a question about my own life is arithmetic, recall, or both. On sub-millisecond regexes, a 1.5-billion-parameter tiebreaker, and why saying nothing is a valid answer.</description><pubDate>Sun, 26 Jul 2026 11:00:00 GMT</pubDate><category>homelab</category><category>resonancelab</category><category>rag</category><category>localai</category><category>sqlite</category><category>python</category></item><item><title>The Index That Was Quietly Empty</title><link>https://bearingout.com/homelab/engineering/ai/2026/07/24/rag-indexing-sqlite-chromadb.html</link><guid isPermaLink="true">https://bearingout.com/homelab/engineering/ai/2026/07/24/rag-indexing-sqlite-chromadb.html</guid><description>Building the lab&apos;s memory with SQLite and a vector store - and the two silent bugs that meant it spent months confidently retrieving from a fraction of what I&apos;d given it.</description><pubDate>Fri, 24 Jul 2026 11:00:00 GMT</pubDate><category>homelab</category><category>resonancelab</category><category>rag</category><category>embeddings</category><category>chromadb</category><category>sqlite</category><category>python</category></item><item><title>Teaching the Lab to Know Me: Journals Into Structured Data</title><link>https://bearingout.com/homelab/engineering/ai/2026/07/22/journal-analytics-obsidian-to-csv.html</link><guid isPermaLink="true">https://bearingout.com/homelab/engineering/ai/2026/07/22/journal-analytics-obsidian-to-csv.html</guid><description>Turning years of daily notes into something a machine can reason over - booleans where the data is crisp, language models where it isn&apos;t, and a hard line about what never leaves the house.</description><pubDate>Wed, 22 Jul 2026 11:00:00 GMT</pubDate><category>homelab</category><category>resonancelab</category><category>obsidian</category><category>localai</category><category>personal-knowledge</category><category>python</category></item><item><title>Mostly Arithmetic: Classifying Intent Without Training Anything</title><link>https://bearingout.com/homelab/engineering/ai/2026/07/20/embedding-nearest-centroid-classifier.html</link><guid isPermaLink="true">https://bearingout.com/homelab/engineering/ai/2026/07/20/embedding-nearest-centroid-classifier.html</guid><description>How the lab decides what kind of question you asked, using an embedding model, about forty example sentences, and some cosine similarity. On thresholds, the margin rule, and the guard that keeps personal questions from being answered generically.</description><pubDate>Mon, 20 Jul 2026 11:00:00 GMT</pubDate><category>homelab</category><category>resonancelab</category><category>localai</category><category>embeddings</category><category>nlp</category><category>python</category></item><item><title>Higher Brain Monitor: A Physical UX for the Home AI Lab</title><link>https://bearingout.com/homelab/engineering/ai/ux/2026/07/19/higher-brain-monitor-physical-ux-home-ai-lab.html</link><guid isPermaLink="true">https://bearingout.com/homelab/engineering/ai/ux/2026/07/19/higher-brain-monitor-physical-ux-home-ai-lab.html</guid><description>A homelab visual UX experiment using Nanoleaf panels to make local AI activity visible in the room: brain, eyes, services, tools, warnings, and the strange comfort of machines with body language.</description><pubDate>Sun, 19 Jul 2026 11:00:00 GMT</pubDate><category>homelab</category><category>localai</category><category>resonancelab</category><category>observability</category><category>ambient-computing</category><category>ux</category><category>nanoleaf</category><category>ollama</category><category>systemd</category></item><item><title>Where Does This Sentence Go? The Gateway Problem</title><link>https://bearingout.com/homelab/engineering/ai/2026/07/18/gateway-service-architecture.html</link><guid isPermaLink="true">https://bearingout.com/homelab/engineering/ai/2026/07/18/gateway-service-architecture.html</guid><description>One assistant, seven kinds of question, and no obvious way to tell them apart. On building a router with three interchangeable brains - and why the dumbest one is the one that runs in CI.</description><pubDate>Sat, 18 Jul 2026 11:00:00 GMT</pubDate><category>homelab</category><category>resonancelab</category><category>localai</category><category>api-gateway</category><category>fastapi</category><category>python</category></item></channel></rss>