A team quality standard for AI-assisted work

Frontier managers set a quality standard for AI work. They do not set a usage quota. Your team already knows the difference between "we used Copilot" and "this is good enough to send." Write that down so it does not live only in your head.

What “good” means, who reviews, how handoffs work.

Three lines are enough: what good looks like for AI-assisted work on this team, who reviews output that came from an agent or a long Chat, and how a handoff is documented so the next person is not reverse-engineering a thread.

Core principles

  1. Good: the human can explain what they kept, what they changed, and what they verified. Fluent is not the bar.
  2. Review: name a role, not a vibe. Agent output and client-facing drafts get a named reviewer. Low-stakes internal notes can be self-reviewed.
  3. Handoffs: if an agent or Chat produced a draft, the file or email says so in one line, plus the sources used. The next person should not have to guess.
  4. Jobs: the standard protects judgment. It is not a performance score for how often people open Copilot.
  5. Publish it. A standard nobody can find is a speech.

Related: Lesson 5 — stories beat usage counts

Check yourself

What belongs on a team quality standard for AI-assisted work?

Do this in Copilot

Write the one-page standard this week. Share it in a team meeting. Invite one edit from someone who is not your AI champion.

Paste into Copilot Chat. Then edit until you would actually pin this.

Draft the one-page team standard

Draft a one-page team quality standard for AI-assisted work. Team: [NAME]. Work we send: [INTERNAL / CLIENT / MIX]. Include: (1) what “good” means in one paragraph, (2) who reviews agent or Copilot output, by work type, (3) how we document a handoff, (4) one sentence on jobs and judgment staying human, (5) what we will not measure (prompt counts, logins). Tone: plain, MillerKnoll, no slogans. Do not invent a maturity ladder. Do not recommend Copilot Studio.

Techniques: Constraints · Negative constraints

Open Copilot

Did you run this in Copilot? Mark complete when you have tried it.

AI

Part of the MillerKnoll collective

Studio · 2026