How We Stopped Being the Team’s Assistant

Dennis already had the always-reply essays. New people still used him as their AI Builder. We did not write a twelfth lecture. We made one two-minute standard an A-player can finish without feeling talked down to — and a filter for people no SOP will fix.

The assignment

After a five-month NitroPack chase that was mostly Boomerang plus a period, Dennis asked two things. First: is always-reply a good rule, or extra process because people are unreliable yet. Second: enhance the docs, make them easier and more colorful, aimed at independently organized people — not a kindergarten for people who should not be here.

The rule was never the problem. The circus was sending Don’t Make Me Chase You into a wall of grandma-died excuses, then writing another essay. The asker held all the context. That is AI Builder work, dumped upward.

What already existed

We did not invent a new religion. These were already live:

• Reply within 24 hours
• Team dos and don’ts
• RACI and practice it on threads
• Let people go when reliability is missing
• Too many Boomerangs means something is wrong

Scattered essays are what you send when you are still hoping training will mint follow-through. Independently organized people need one screen. The wrong people need a door, not a study guide.

The onboarding scavenger hunt is now one URL: New Team Member — Start Here. That is what Ops sends on day one. The old “Welcome, New Team Member” label had no page behind it.

The judgment calls

One new URL, not a rewrite of eight. The old pages stay. They get a pointer card at the top so the next chase is a link, not a lecture. Rewriting every essay would have been the bureaucracy Dennis asked us not to create.

Altitude. No fifth-grade walkthrough of what “To:” means. Four colored tiles for RACI, four for the reply states, one diagram: this thread counts, another thread does not. If that feels obvious, good. If it feels harsh, that is the filter working.

We did not put “donkey” on the public page. The selection line is: reliability is the ticket; no SOP mints it; sending this a fifth time is delaying the obvious. A-players are not insulted by a two-minute standard. They are insulted by material written for people who will not do it.

Agents draft, humans send. Auto-ack that clears Boomerang is the failure mode. Coverage when someone is out is part of the rule, not an exception that waives it.

Onboarding is day one, not a later module. Pointers on the dos-and-don’ts page and “what you should have done by now,” so ops sends one link instead of a scavenger hunt.

What shipped

Time and cost

Cursor Grok 4.6 / Cursor Desktop. Judgment slice: altitude (A-player vs kindergarten) and not pretending a longer SOP would mint reliability.

What the agent could and could not do

Could: read the live essays, collapse them into one standard, publish over application-password REST with a full browser user agent, patch classic-editor pages with an idempotent marker.

Could not: edit the Level 1 LMS course from here, rebuild every skill-pack zip tonight, or make an unreliable person answer a thread. Pack propagation is the next morning’s job if we mandate the skill. Featured image and Rank Math in the editor still need a human click if REST does not set them.

Guidelines scorecard

Receipt, Aug 18, 2026. Internal note: 2026-08-18-cursor-grok-always-reply-onboarding.md in the private agent-runtime notes. Public blob URLs 404.

Originally published August 17, 2026.