Skip to main content
// series

Ship With Agents

Agents do not ship software because you bought a seat. This series is the working setup that makes them ship: runs that last for days, a stack staffed like a team, a phone that counts as a dev machine, the right model for each job, and clear ownership when agents run in parallel.

5 posts · oldest first

  1. 2026.097 min

    Let It Run for Days

    Letting an agent chase a UI bug across days is easy. Making that run bounded, reversible, and inspectable is the actual skill.

  2. 2026.099 min

    Your Stack Is a Team, Not a Subscription

    A paid AI stack only becomes leverage when each tool has a job, a handoff, and an owner. Receipts without roles just multiply tabs.

  3. 2026.109 min

    The Phone Is a Valid Dev Machine

    The phone is not a tiny laptop. It is a valid control surface for agents, as long as the seam to the desk is a contract, not a vibe.

  4. 2026.108 min

    Pick the Model for the Job

    Models are instruments, not personalities. Match the seat to the phase (UI, plan, refactor, review), write the stop condition, and demand the proof that seat is for.

  5. 2026.108 min

    Parallel Agents Need Ownership Too

    Parallel agents only become throughput when each lane has a named owner, a blast radius, and a kill switch. Tabs do not dilute accountability.

back to /blog/