The problem nobody governs until it is too lateYou licensed Copilot Studio. Makers started building agents. Six months later you pull the Agent Registry export and it is four thousand rows of New Agent, cr8a3_copilot1, and Copilot final (2). You cannot tell a read-only personal helper from a finance agent touching confidential data. You cannot sort the export, filter it, or answer a single oversight question from it. And here is the part that hurts: the system name locks the moment the agent is created. No rename fixes it. The convention has to be right the first time.Agent naming looks cosmetic. It is actually a governance control. The name is the one field that surfaces in every inventory, every audit log, and every @mention box in Teams. Junk names make live agents invisible to the people who answer for them.This book is the structured fix, and the toolkit makes it enforceable.What is insideThe book — 10 chapters, governance-first, no filler.Not a style guide. Every rule traces to an operational consequence: inventory legibility, audit-trail integrity, sprawl control, oversight cost. If a rule names no consequence and no enforcement point, it is not in here.It does two jobs and refuses to do only one:- Adopt as is (Chapter 4). A complete, ready-to-use convention you can put into production today, without modification: the two-name model, the canonical system-name regex, a starter department and function dictionary, the binding description format, and a full worked example set on a neutral fictional org (Contoso) so you can map it onto your tenant without translation.- Design your own (Chapter 6). The ten-step framework to derive a different but equally sound convention if your org needs one, with the reasoning behind every segment so you can defend each decision to your CoE.Plus: why two names are never one (the immutable schema name vs the editable display name), how naming wires into zones, environments, and ALM, the four enforcement points that make a name right before the agent exists, how to read an inventory export audit-first, the retrofit playbook for the sprawl you already have, and a one-page anti-pattern rule set.Ships as a print-ready PDF (125 pages, A4) and a reflowable EPUB for any e-reader.The toolkit — five working artifacts, not screenshots.1. Rule card — the one-page convention you print and paste into your CoE wiki.2. Validation checklist — the validator, including the canonical system-name regex, to run at intake, at the gate, and on every inventory pull.3. Intake field spec — the field-by-field spec for the agent intake form, where the two names and the binding description are captured together.4. Abbreviation dictionary — the governed source of truth for department and function codes, with the change process.5. Retrofit worksheet — the per-agent triage and phasing worksheet for cleaning up existing sprawl.Every rule in the book has a matching artifact or worked example. You read the chapter to understand the rule, then use the file to do the work.Who this is for- Copilot Studio Center of Excellence leads who own the registry export- Power Platform and Microsoft 365 administrators who answer for sprawl- AI champions standing up an intake or governance gate- Platform and governance teams enforcing a convention at scaleIt assumes you already know what a managed environment and a DLP policy are. It skips the throat-clearing and hands you the operational answer plus the gate that enforces it.What makes this differentIt treats naming as a control, not decoration. Most naming advice is a list of preferences. This is a convention with an enforcement point for every rule: where the regex runs, where the CoE gate fires, where the dictionary check rejects a junk code. A reviewer cites the rule, not an opinion.Two complete paths ship in one book. You are not forced to adopt someone else's taxonomy. Adopt Chapter 4 wholesale, or use Chapter 6 to build your own and know exactly why each segment exists. Both paths are complete. Both ship.It survives the platform changing under you. One number ages fastest: the agent display-name character limit. No durable count is hard-coded anywhere in the book. The length rule is expressed as a percentage of whatever the live cap is, so it still holds when the number moves. The book tells you where to confirm the current limit before you rely on it.The artifacts are usable today. The regex is byte-identical everywhere it appears. The validator runs. The intake spec maps to real form fields. The retrofit worksheet triages a live estate. This is the difference between reading about a convention and shipping one.Frequently asked questionsIs this tied to one company or sector?No. Every worked example uses the fictional org Contoso and universal business departments (Finance, HR, Procurement, Legal, and so on). Nothing is specific to one industry or tenant. You map it onto your own org without translation.Do I need Copilot Studio specifically, or does this work for Agent Builder too?The convention governs the two names and the binding description that both surfaces carry. It applies wherever agents are created and inventoried. The enforcement points reference the intake form and CoE gate, which you wire to your own maker surface.I already have hundreds of badly named agents. Is it too late?No. Chapter 9 and the retrofit worksheet are built for exactly that. They sort what is locked (the schema name) from what is still editable (display name, description, metadata), then give you a five-step phased playbook so you fix the highest-risk agents first without trying to boil the ocean.What format are the files?A ZIP containing the book as a print-ready PDF (A4) and a reflowable EPUB, plus the five toolkit artifacts as clean Markdown you can drop into your internal docs or wiki.How current is it?Version 1.0, 2026. Platform facts are sourced and dated. The one volatile number, the display-name length limit, is deferred to the live source by design, so the rule survives the change. Free updates re-deliver when anything material shifts.Kesslernity builds governed AI agents and the controls that keep them auditable at enterprise scale.store.kesslernity.com · kesslernity.com · Version 1.0, 2026Licensed, not sold. Full License & Terms apply (ref KESS-LIC-2026-001). By purchasing you agree to the version in force on your purchase date.Kesslernity is an independent publisher: this product is independent analysis, not affiliated with, sponsored by, or endorsed by Microsoft. Microsoft 365 and Copilot are trademarks of the Microsoft group of companies. It is practitioner guidance, not professional, legal, or financial advice.Questions before you buy, or support after? Contact mathieu@kesslernity.com.Terms of Service · Privacy Policy