Arc · 9 posts

From Control Tower to First Product

One question, asked after a project spent 25 days producing documentation and no working code: why does a factory that can build individual tools still struggle to coordinate them into one real product? The answer took three working sessions to build and prove: trust moved out of the agents doing the work and into the contracts between them, so a status can be checked instead of just believed, and a failure shows up loudly instead of quietly passing as success.

Reading order

  1. 1
    The Project That Shipped Nothing Twice

    building-the-agentic-factory

    Twenty-five days of beautiful documentation, zero source files, and a sibling lab that had quietly done the exact same thing. The fix wasn't more planning. It was an honest demotion.

  2. 2
    Status Is Derived, Never Stored

    agentic-sdlc-patterns

    The original design had a status field you edited. The rewrite has no status field at all, just an append-only event log and a pure function that replays it. Concurrency safety and text-editor repair both fall out for free.

  3. 3
    Giving the Factory a Spine in a Day

    building-the-agentic-factory

    A session tracker became a task-DAG orchestrator by adding optional fields and one new self-transition event to what already existed, instead of forking a second system, proven the same day against a real 12-task build.

  4. 4
    When Exit 0 Lies

    governance-without-killing-the-joy

    A dispatcher trusted exit code alone to mean success, until a neighboring lab's unrelated refactor made exit 0 and 'nothing happened, on purpose' the same integer. The bug was caught before it ever reached a real build.

  5. 5
    The Lego-Block Contract

    from-prompt-to-platform

    A tool is only genuinely modular if it's fully useful with zero of its integration partners present. The way to prove that instead of asserting it: a runner-agnostic contract plus a checked-in reference implementation that needs none of them.

  6. 6
    Build the Fake Project First

    agentic-sdlc-patterns

    Every remaining feature got proven against a checked-in, disposable demo project before touching anything real, until an ARB review found the demo's own success report could go green without ever actually checking the thing it claimed to prove.

  7. 7
    The Rework Path That Never Existed

    gremlins-at-work

    The CLI's own error message told you to block a failed task for rework. That advice was an illegal state transition every time it printed, since the day the vocabulary was written, and nobody noticed until a review actually read the transition table.

  8. 8
    Verifying the Wrong Universe

    agentic-sdlc-patterns

    Worktree isolation was correct. Post-task verification was correct. Composed together with no validation at the seam, a relative path silently pointed the verify step at the wrong directory, and produced a confident, meaningless result.

  9. 9
    Two Gates, Two Claims

    governance-without-killing-the-joy

    V1 and go-live sound like two names for the same milestone. Treating them as one claim would have let a comprehensive test suite stand in for real-world proof, so the gates were deliberately split, and the first honest evidence for the second one arrived not yet met.