Hand your skeptics the don't-ship list
Virginia (CPO & CMO)
- ai-transformation
- running-the-initiative
- fluency
- quality-bar
- playbook
Hand your skeptics the don't-ship list
The people who distrust AI output are often the ones who should write your quality bar.
Not as a pep talk. As authors of a concrete artifact: Play 2's "good / better / don't ship" examples, and Play 4's per-channel quality bar for anything that leaves the building.
Skeptics aren't one group
When leaders talk about "AI resistance," they usually mean one bucket. The data doesn't.
In organizations that make AI available, non-users and infrequent users name different reasons for holding back. Among non-users, 39% don't believe AI can help their work, and 43% say they are ethically opposed. Infrequent users cite those less (22% and 25%) and point more to usefulness and privacy. Concern that their role will be replaced sits at only 10% for either group. (Gallup, Feb 2026 fieldwork, n=23,717)
Job-loss fear shows up elsewhere — and later. It rises with use, not with avoidance. Frequent users are more than twice as likely as infrequent users to fear elimination within five years. Managers who treat people as respected, and organizations that show they care about wellbeing, weaken that link. Broadcast strategy decks don't. (Gallup, Sept 2026)
So the person who refuses the tool and the heavy user who worries about it need different conversations. This post is about one type: quality and trust skeptics — the people who have used AI and still don't trust what it produces.
The quality skeptic is experiential
About a quarter of infrequent users and non-users in AI-enabled organizations say they have used AI and do not find it useful in their role (27% and 26%). (Gallup, Feb 2026)
That is not abstract fear. It is usually rework, wrong answers, and "sounds fine until a human reads it." These people are often still using the tools. They are hard on the outputs, not opposed to the idea.
Our Fluency plays already ask for the artifacts they care about:
- Play 2 — shared prompting craft with your own domain examples, including "good / better / don't ship."
- Play 4 — a quality bar for anything external.
What the playbook does not yet say is who writes those.
Give them the pen
Co-design with skeptics is not a feelings exercise. It is a staffing decision for two columns.
In each function's next craft or listening session:
- Name one credible quality skeptic — not the loudest person in the room. Someone whose craft others already trust.
- Give them the pen for 10–20 cases from their tickets, emails, or docs: good / better / don't ship. Real work, not synthetic prompts.
- Promote that set into the Phase 1 example library and into the external quality-bar paragraph for that channel.
- If a craft session concludes "not this workflow," document it in the inventory. Fluency does not need 100% uptake. "We tried; not here yet" is a valid result, not a failure of persuasion.
This continues the same artifact line as our golden-set post: examples first, then change. The don't-ship list is Phase 1's version of that discipline — and quality skeptics are the natural authors.
It also fits the listening sessions from Start by asking, not banning. You already set up the room. This is what one of those rooms produces when the skeptic holds the marker.
What this is not
It is not "convert the skeptics." It is not a usage dashboard. It is not a promise that every role should use AI.
Relevance skeptics — "AI can't help my job" — are sometimes right. Occupational exposure varies; some work is simply less touched by current generative use cases. (Gallup, Feb 2026) Ethical objections are real too. Training does not dissolve them. Respect, clear scope, and legitimate opt-outs where the work allows are the response — not another town hall.
Managers still matter more than messaging. Where the manager actively supports AI, frequent use is far higher (78% vs 44%). (Gallup, Feb 2026) The don't-ship list is something a manager can sponsor in a craft session this week without waiting for a new platform.
What to do Monday
- Pick one function that already ran (or is about to run) a Fluency craft session.
- Name the quality skeptic who should author the don't-ship column.
- Collect 10–20 real cases before the next change to prompts, tools, or policy.
- Publish the set next to the quality-bar paragraph for anything that leaves that function.
If you want a short companion note later: usage quotas and token leaderboards are being walked back in public (Duolingo dropped AI use from reviews; Meta told engineers it would not use adoption dashboards or token counts to evaluate impact). Keep usage as learning telemetry. Judge outcomes and quality. That extends Play 5 — and it can wait for its own row.
Excellence at the bar compounds. Pep talks about "embracing AI" do not.
This is part of Running the Initiative, our series on operating the AI Transformation Playbook one section at a time. Start with the playbook, or take the Journey Assessment — a mirror, not a score.