Manager's AI Working Agreement Kit

The situation you are in

Some number of your people are using these tools right now, on work you are accountable for, and they have not told you. Not because they are hiding anything — because nobody asked, there is no rule, and volunteering it feels like volunteering for a conversation they do not want to have.

So you are already downstream of adoption. What you do not have is a frame. Nobody on your team knows where you stand on whether the analysis they generated and edited counts as their analysis, whether the self-assessment they wrote with help is honest, or whether saying "I drafted this with an assistant" will cost them something at review time. They are each guessing, they are guessing differently, and the guesses will collide the first time two people's work sits side by side.

Meanwhile the signals you used to manage by have gone quiet. The weakest writer on your team now produces competent prose, which is genuinely good and also means writing quality has stopped telling you anything about thinking quality. Volume is up. Your reading load is up. And the thing you actually manage — judgment — has not improved at all; it is just harder to see.

You do not have a quarter to work this out. You have a full week, a meeting slot you could take, and the knowledge that the first real incident is coming and you would rather have set the frame before it than after.

What this kit is

One meeting, one page, and a thirty-day check.

Ten chapters, about 17,000 words, and twelve templates that are finished documents rather than outlines. It is written for a manager with three to fifteen direct reports who wants a defensible, specific answer to one question — what part of this work is supposed to be your own thinking? — that every person on the team would answer the same way.

It is opinionated where opinions help. It says the ban and the mandate both fail, and why. It says disclosure is a sentence, not a ceremony, and that if disclosure costs anyone anything with you it will stop within a month. It says the tool gets the questions and never the read on a person, and it names the four prompts that break that rule while feeling like diligence. It says do not record one-on-ones, ever, even where it is legal and one click. It says never ask a tool to compare two people, and it explains the calibration-season trap that produces nine reviews with a family resemblance and no differentiation you can defend.

What is inside

• The guide. Ten chapters: what actually changes for a manager, the production-preparation-presence split the whole agreement hangs on, your operating context and your company's, the four decisions, one-on-ones and coaching, the three hard messages, meetings and transcripts, privacy and fairness, the norms meeting run beat by beat, and the thirty-day review.
• The agreement. The one-page team working agreement, fill-in, with the four decisions, the human-only list, and your own obligations on the same page — because a one-sided document is a policy, and people notice.
• Prep templates. A manager context block, one-on-one prep, coaching conversation prep, and the weekly update structure.
• Hard-message drafts. Performance, change, and a no, each with every judgment call bracketed so the parts that have to be yours are impossible to skip.
• Meeting machinery. The three-tier notes protocol, the four-section note that replaces the raw summary, and a decision log with a reopen condition on every line.
• Rollout. The team meeting scripted with timings and what to say in the open, the 30-day rollout, the never-goes-in list, and the thirty-day review sheet.
• START-HERE. The map: what to do if you have an hour, a day, or a week, a read order for the ten chapters, and a table that names which file to bring to which meeting or moment — the norms meeting, the first one-on-one after it, review season, day thirty, the day someone new joins.

Every template uses [BRACKETED PLACEHOLDERS] and carries a "How to adapt" note. They are plain markdown, so they paste into a document, a wiki, or a chat without fighting formatting.

Who wrote it and why he can

Wayne Bridges, a public-sector solutions engineer who builds and runs AI agent systems daily.

That second half is the part that matters here. The guidance on what these tools do with data, where their output quietly distorts, and which prompts produce confident nonsense comes from operating them every day, not from reading about them. The management half comes from having sat on both sides of the one-on-one — from being coached, and from being responsible for other people's work and growth.

He is not a lawyer and not an HR professional. That is stated at the top of every chapter that touches law, policy, HR, or procurement, and it is the reason the kit is built to get you to a good conversation with your HR partner rather than to replace one.

What this is not

• Not HR or legal advice. Employment law and company policy vary, and both are moving. The chapters are written so you can ask precise questions of the people who own the answers.
• Not a company policy. It is a team working agreement. It sits underneath your company's policy, never over it, and the page says so in one line.
• Not a product recommendation. No tool is named. Tools change monthly; the four decisions do not.
• Not a prompt library. There are prompts in it, as illustrations of a line being crossed or held. Craft tips belong somewhere else, and mixing them into a norms document makes the norms feel optional.
• Not an adoption program. If you have been handed a usage metric, the kit will not help you hit it, and chapter 9 tells you to keep it out of the norms meeting entirely.

Who should not buy this

• Individual contributors. This is a kit about managing other people's work. Much of it would not apply to you.
• Managers whose company has already published detailed practice guidance — not just a policy, but team-level norms on disclosure, coaching, and documentation — that HR has reviewed and managers actually follow. You may still want the hard-message drafts and the notes protocol.
• Anyone who wants a list of approved tools, a prompt pack, or a way to make their team use AI more. None of those are in here.
• Anyone hoping to shortcut the part where a manager writes the four sentences about a person's work themselves. The kit's whole position is that those sentences are the job.

If you manage a team that is already using these tools and you have not had the conversation, this is for you. Read START-HERE, draft the one page, and put the meeting on the calendar for next week.