Skip to content

Boss Fight 0B: Automate one workflow that saves 4+ hours/week

Status: drafted · Time: 120 min · Audience: pm-designer-ops Outcome: Run one safe automation for two weeks, prove it saves time, and document the recipe for a teammate.

🏁 Boss fight · ⏱ 2 weeks · 🎯 Win condition: one real workflow automated end-to-end, measured for two weeks, and documented so a teammate can fork it.


Ops 101 is not complete because you read eight chapters. It is complete when your week is materially lighter.

The boss fight asks you to pick the highest-leverage entry from your agent diary, configure the workflow, run it for two weeks, and show the evidence. The target is simple: 4+ hours saved per week.

If that sounds high, good. A boss fight should not be a worksheet. It should change your calendar.


Use the diary from Quest 0B-2. The right workflow usually has five traits:

  • it happens at least weekly;
  • it takes 30+ minutes each time, or creates delay for multiple people;
  • the inputs are accessible through an approved surface or connector;
  • the output has a repeatable shape;
  • a human can review the output before it affects anyone else.

Do not pick a workflow whose first version requires sensitive data transmission, money movement, irreversible changes, or judgement you cannot explain.


Your automation can be a configured Claude project, a Cowork workflow, a Slash pattern, a connector-backed recipe, a lightweight agent, or a documented prompt-and-review process. The surface matters less than the repeatability.

It must have:

  • a named trigger;
  • named inputs;
  • a clear output format;
  • a review step;
  • a failure path;
  • a place where the recipe is documented.

The review step is required. The first version should make you faster, not invisible.


For each run, capture:

  • date;
  • time spent before automation would have run;
  • time spent with automation;
  • manual review time;
  • quality issues found;
  • whether you trusted the output enough to use it.

Two weeks is long enough to expose whether the workflow actually holds up. One impressive demo does not count.


Submit a short boss-fight note with:

  • the workflow name;
  • the old process;
  • the new process;
  • the AI surface or agent used;
  • two weeks of timing evidence;
  • before/after examples with sensitive details removed;
  • the review checklist;
  • the documented recipe for a teammate;
  • the weekly hours saved.

If you saved less than four hours, document the result anyway. You may still have built a useful recipe, but the boss fight is not complete until the impact clears the bar or you pick a better workflow.


You pass Boss Fight 0B when:

  • the automation ran for two weeks;
  • it saved 4+ hours per week or removed equivalent team delay;
  • the output was reviewed by a human before use;
  • the recipe is safe under the Prologue safety brief;
  • a teammate could fork the recipe without a private walkthrough.

“I can identify routine ops work, turn it into a safe AI-assisted workflow, measure the time saved, and leave behind a recipe my team can reuse.”

That is the Ops 101 outcome. Not curiosity. Not prompt fluency. A lighter week, proven.


Previous: ← Quest 0B-2 The agent diary · Up to: ↑ Ops 101 README