When Not to Use AI
The judgment layer. Keep AI off the calls it should never make.
The free, open-source Copilot prompt library that sits under this work was cited by Jared Spataro (Microsoft, AI at Work) in his 25 February 2026 note. He was citing that free repo when it held 300+ prompts. This is a separate, paid product and he did not review or endorse it.
The situation
AI is being added to every workflow, every tool, every team. The conversations are all about adoption. Nobody is having the other conversation: when do you say no.
Not because AI is not useful, it is. But using AI in the wrong places does not just produce bad outputs there. It erodes trust in every output you produce, including the ones where AI is genuinely helping.
What this is
A practical decision framework for anyone who has to make the call on whether a workflow gets AI. The decision lines come from real rollouts, including the calls that went wrong.
What is inside
- A 5-question pre-flight check that takes 10 minutes
- A use-case scoring matrix with a worked example
- A red flags checklist: 12 observable signals in 3 categories
- Conversation scripts for 3 situations where you need to say no
- A decision tree, and verdicts on 10 real use cases with full reasoning
- A copy-pasteable AI use policy clause
Answer these questions before deploying AI to any new workflow. They take 10 minutes. If you answer YES to two or more, proceed with significant caution or don't proceed at all. If you answer YES to three or more, do not deploy until the conditions change.
Question 2: Is the cost of a confident wrong answer higher than the cost of a slow right answer?
Why it matters: AI optimises for producing an answer. It does not optimise for knowing when it doesn't know. A slow process that surfaces uncertainty is often better than a fast process that masks it.
Every question carries a "YES means" and a "NO means" block telling you what the answer commits you to. Four more questions follow.
Who it is for
- IT operations, project managers and team leads
- AI champions handed a mandate to "add AI" who have to work out where it actually belongs
Who it is not for
- Anyone looking for an argument against AI. This is not an anti-AI book. It is what makes the AI you do use trustworthy.
Questions people actually ask
Is this Copilot-specific?
No. The framework is about which workflows should get AI at all, so it reads the same whether you are deploying Copilot, a Claude or GPT deployment, or something you built yourself. Copilot rollouts are where most of the examples came from, because that is where most of the real decisions have been made.
We already have an AI policy. Does this duplicate it?
Policies say what is allowed. This says which workflows are worth it, which is a different question and usually the unanswered one. There is a copy-pasteable policy clause at the end for the gap it does tend to leave: the use cases nobody wrote down because nobody wanted to be the person saying no.
Does this cover safety-critical work?
It covers how to keep AI out of it. The line this material holds throughout is that AI prepares and humans decide. Nothing in it advises using AI for safety authorisation, and the framework is explicit about why those workflows fail the pre-flight check.
Buying this with a company card?
Your receipt email carries a Generate link that opens an invoice generator: add your billing address, put your legal company name and any reference your finance team needs in the Additional notes box, and download the PDF. If you have an EU VAT number, enter it at checkout and no VAT is charged. Here is the approval email, so you do not have to write one.
Subject: Approval request: When Not to Use AI ($29, one-time) Hi [Manager], I would like to expense a one-time $29 purchase: When Not to Use AI by Kesslernity. What it is: The judgment layer. Keep AI off the calls it should never make. What we get: - A 5-question pre-flight check that takes 10 minutes - A use-case scoring matrix with a worked example - A red flags checklist: 12 observable signals in 3 categories - Conversation scripts for 3 situations where you need to say no Why it is worth it: it is a one-time purchase with no subscription, and at $29 there is no statement of work, no discovery call and nothing to negotiate. The author publishes the open-source Copilot prompt library that Microsoft's CMO for AI at Work cited as "battle-tested in production environments". A downloadable invoice is available from the purchase receipt for expenses. [Your name]
When Not to Use AI
The judgment layer. Keep AI off the calls it should never make.
Get it for $29Licensed, not sold. The full License & Terms apply. Kesslernity is an independent publisher: this is independent analysis, not affiliated with, sponsored by, or endorsed by Microsoft. It is practitioner guidance, not professional, legal, or financial advice.
What this pairs with
Picked by hand, not by sales data.
The AI Champion's Playbook
The operator's manual for the accidental AI lead. Second edition, August 2026.
$59→Critical Density
AI governance read through the Kessler Syndrome, with a five-artifact toolkit.
$59→When Not to Use AI is also part of From Decision to Deployment: $39 for both instead of $48 separately.