Posts tagged AI

Posts on using AI in real projects: coding agents, open-weight models, and the AI features behind ergy.coach. 15 posts with this tag.

2026

  1. Agent quality gate, part 5

    40 pull requests in one evening with an agent: the decisions I kept

    One evening, five repositories, 40 merged pull requests and 7 releases with a coding agent. The speed came from the agent. What made it safe was the human in the loop: plans, break points and the decisions it never made alone.

  2. Agent quality gate, part 4

    Prove it: measuring that a refactor changed nothing

    An agent says its refactor doesn't change the rendering. I asked it to prove it: every element, 46 computed styles, 16 pages, 0 differences. The measurement had to be tested first.

  3. Agent quality gate, part 1

    Your coding agents need feedback in one second, not one minute

    I asked whether SonarQube could review my agents' pull requests. The answer was a stack of checks ordered by speed: hooks for fast feedback, CI for enforcement.

  4. An always-on box for my coding agents, on Railway

    How I set up a cloud server to drive Claude Code and Codex from my phone: Railway, T3 Code, signed commits, Vite+ as the toolchain, and a hard line between client and personal code.

  5. Engineering agents, part 7

    The toolkit: Agent Skills from this series

    Every Agent Skill from the engineering-agents series, grouped by step, on one page.

  6. Engineering agents, part 6

    Don't break what works: Chesterton, Hyrum, and the supply chain

    The simplifier agent removed 400 lines. Another agent added a dependency that pulls in forty others. Each change looked harmless.

  7. Engineering agents, part 5

    Scale without chaos: Brooks, Conway, Lehman

    Six agents in parallel, three agents split by technical layer, six months of nobody deleting anything. Three laws saw it coming.

  8. Engineering agents, part 4

    Verify: test first, mutation gates, independent review

    "47 tests added, all passing, 94% coverage." Then the same agent reviews its own code and finds nothing.

  9. Engineering agents, part 3

    Make wrong code impossible: structure, deep modules, contracts

    A 180-line function, a rounding helper imported from eleven files, an optional discount that can be -15. None of these agents made a syntax error.

  10. Engineering agents, part 2

    Start right: thin slices, one job per agent, no premature optimization

    You ask an agent to "build the billing module." Twenty minutes later, it has generated 3,000 lines, and nothing compiles.

  11. Engineering agents, part 1

    Why fundamentals still matter: your agents aren't dumb, they're amnesiac

    On Monday, you explain to an agent that a 200-line function is not acceptable. On Tuesday, a fresh agent writes one with 250 lines.

  12. Hacktoberfest 2026 isn’t about PRs anymore, and that’s good news

    Pull requests no longer count toward Hacktoberfest rewards. The scoreboard moved to learning and building with open-weight models, stickers included.

  13. Improvement: analyzing rowing technique with AI

    An evening proof of concept: pose estimation in the browser, stroke metrics from video, and an agent that coaches from numbers not thresholds.

  14. Data-driven: pulling metrics my rower won't give me

    My rower exposes no API and Strava bans AI use. I pulled live metrics over Bluetooth FTMS instead.

  15. I built the tool I needed for my own training

    First post in a 3-part build-in-public series: why I built ergy.coach for my own sub-8:00 2k obsession, with an agent that actually remembers.