Who does what so that AI with Microsoft stays in operation.
A business team keeps the test set, one person owns every agent, a directive says what is allowed, and your IT keeps identities, data and cost in view.
The Frontier Firm designs workflows for people and agents from the start.
In the sales team's inbox, prepared quotes, assigned enquiries and a list of the cases that need a person's decision are waiting in the morning; the agents worked through the night. That is how Microsoft describes the Frontier Firm, and that is how we understand AI-native work: a workflow is designed for people and agents from the start, with a person leading a team of agents, and with the prospect that over time one agent carries the largest part of a workflow. The road there starts small, with a first case, a business team that owns it, and a sheet that says who decides what. The strategy before it, from the goals of the management to the order of the areas, is worked out with you by KA Partners; the implementation on Copilot, Copilot Studio and Microsoft Foundry is ours, and both sides work from the same plan.
02
Owner, business team, data protection and platform carry every application.
Every agent and every application has an owner, who is the sponsor of the identity in Entra, decides on changes, answers approvals, reads the reports and hands over to a named person on leaving; the agent list in the Microsoft 365 admin center shows where that is missing. The business team keeps the test set: it collects real questions and tasks, notes the expected answer and the source, rates the agent's answers and adds the questions it could not solve. The data protection office countersigns the risk sheet, demands the impact assessment where the risk is high, and keeps the application in the record of processing activities. And the platform owner in IT keeps tenant settings, licences, Conditional Access, guardrail policies, budgets and the message center in view. These names stand on a sheet per application, and where one is missing, the agent does not start.
03
The directive says what is allowed and offers the approved route.
At the centre of the AI directive is a table of the approved tools, their contracts and the highest sensitivity level that may be entered there; added to that are rules on when AI content is labelled, for example when an agent writes to customers, and who reviews the result of Cowork before it is passed on. To shadow AI the directive answers with offer and boundary: Copilot Chat with web results is open to everyone in eligible subscriptions, the approved apps and agents are in the table, and on managed devices a data loss prevention rule warns when confidential content is pasted into AI websites and blocks it for personal data; the notice leads straight to the relevant passage of the directive. Data Security Posture Management shows which AI apps and agents touch sensitive data, and the Purview browser extension records the use of AI websites without a company contract. Whoever makes the approved route convenient rarely needs the ban.
04
Training happens on real emails and documents, in small groups.
A training session on someone else's examples stays theory; on their own files a group learns in one session what it needs in the afternoon: writing a good brief, with goal, context, sources and the desired form; opening the reference before adopting an answer; spotting an invented figure and an outdated version; and knowing when the person takes over. In each area a champion starts ahead of the others and answers questions in their own team, the communication calendar says who sends what to whom and when, quick guides and videos from every session land on the learning site, and in the prompt-a-thon the areas practise on tasks from their own work. Microsoft provides the Success Kit, the Launch Day kit, the scenario library and instructor-led QuickStart sessions for this. The directive belongs in every session, with the cases in which it applies.
05
The test set is the release gate, versions and handover follow it.
From pilot to operation lead the stages the Agent Explorer shows in chapter 3: the task, the test set, the pilot, the release, the operation and the evolution, and every stage ends with your decision. The test set runs against the agent before every new version, in Copilot Studio with test sets from CSV, in Foundry with evaluations and the Red Teaming Agent, and continuously in operation; a version goes live only when it scores no worse than the previous one and the owner approves. Every change to model, instructions, sources or tools is a new version and takes the same route. The handover into operation is a sheet: data, actions, approvals, owner and deputy, test set, cost per case, the model's retirement date and the date of the next review. Agents without owners are counted by the agent list in a tile of its own; they are handed over or blocked.
06
Support after handover and a fixed review rhythm keep the operation going.
If the agent answers wrongly on Tuesday, on Tuesday someone knows what to do: the champion in the area takes the report, the owner decides, and our Support & Operations team checks trace, sources and version, keeps cost alerts and failed runs in view and installs updates. The review rhythm is in the sheet: the message center announces major changes at least 30 days in advance and is read and assessed; the retirement dates of the models from chapter 2 are in the data sheet; facts, prices and rights we check every quarter; and usage per area comes from the Copilot Dashboard, which needs no extra licence to view, and from the usage reports, which conceal names by default. How finely usage is evaluated and who sees the figures is set by a data rule that the data protection office and the staff committee sign. What the quarterly review puts on the table:
Test set and evaluation: which questions the agent solves, where it hands over, which sources are out of date and what belongs in the next release.
Cost per case: the consumption from chapter 7 against the cap, and whether the agent gets more people.
Rights and rules: owner, app roles, approval levels and Conditional Access, checked against the risk sheet.
Facts and prices: retirements, licence changes and new requirements from the message center, with the date of the review.
What you decide.
Who owns the agent after the pilot?
Options
1IT, because the agent runs on its platform
2An owner in the business team, the platform owner in IT, and a second line for support
3The project team that built it
Our recommendation: An owner in the business team, as sponsor of the identity in Entra, because they decide what the agent may do and whether its answers hold up. The platform owner in IT keeps tenant, licences, policies and budgets; the second line, with you or in our Support & Operations team, checks trace and version. Both names and a deputy stand in the handover sheet before the pilot ends.
What we advise against: The project team or nobody. A project team dissolves, and an agent without an owner is blocked.
Ban shadow AI or offer an approved route?
Options
1Block AI websites on managed devices
2Allow everything and trust in good sense
3Offer and boundary: approved tools in the directive, a warning for confidential content, a block for personal data
Our recommendation: Offer and boundary. Copilot Chat for everyone, the approved apps and agents in the directive's table, training on real tasks, and on managed devices the rule that warns for confidential content and blocks for personal data, with the pointer to the directive. Data Security Posture Management shows whether the route is taken, and the table grows with every reviewed tool.
What we advise against: A pure ban, because the work then moves to the private phone, and equally having no rule at all.
When does a new version go live?
Options
1When the developers say it works
2When the test set passes with no regression against the previous version and the owner approves
3On fixed dates
Our recommendation: When the test set passes and the owner approves. Every change to model, instructions, sources or tools counts as a version; agents that read input from outside additionally go through the Red Teaming Agent. The result is recorded in the sheet, and the predecessor stays ready until the new version has shown in operation what the test set promised.
What we advise against: Rolling out a version because the vendor released a new model. A new model answers differently, and that shows first in the test set.
How often do we review facts, prices and rights?
Options
1When something stands out
2Per quarter per application, plus the message center continuously and the cost monthly
3Once a year
Our recommendation: Per quarter per application, with a fixed date in the sheet: test set and evaluation, cost per case, rights and rules, facts and prices. The platform owner reads the message center continuously, the cost alerts come monthly or earlier, and the retirement dates of the models are in the data sheet. The record of processing activities is reviewed by the data protection office once a year.
What we advise against: Once a year only. Microsoft announces major changes with at least 30 days' notice, and whoever reads them only in the annual review has experienced them as a disruption.
Status and sources.
As of: October 2026
As of 29 September 2026
The Copilot Dashboard shows adoption and impact per area, with no extra licence to view.
The Microsoft Copilot Dashboard in Viva Insights is available to Microsoft 365 customers with a business or enterprise subscription, with no paid Viva Insights or Copilot licence needed to view it. From one Copilot licence in the tenant it shows adoption and impact, including Copilot Chat; from 50 Copilot licences it adds agent insights, benchmarks, surveys and views for managers. Groups below the minimum group size are not shown.
Usage reports cover Copilot, Copilot Chat and agents, with names concealed.
The usage reports in the Microsoft 365 admin center show use of Teams, SharePoint, OneDrive, Copilot, Copilot Chat and agents over 7, 30, 90 or 180 days. User, group and site names are concealed by default; showing them means changing an organisation setting, and that change is recorded in the Microsoft Purview audit log.
Success Kit, Launch Day kit and scenario library for the Copilot rollout.
Microsoft's Copilot adoption and onboarding guide points to the Microsoft Adoption site: a Success Kit for user enablement and technical readiness, a Launch Day kit for the start, a scenario library with success measures by industry and role, and instructor-led QuickStart training sessions.
The message center announces major changes at least 30 days in advance.
Microsoft announces changes to Microsoft 365 in the admin center's message center, grouped into plan for change, stay informed and prevent or fix issues. Major updates come with at least 30 days' notice. Anyone who reads and assesses the message center does not experience retirements and new defaults as a surprise.
In the Microsoft 365 admin center, the registry lists all agents available to your organisation: from Microsoft, from partners, published by your organisation and shared by individual people. A tile counts the agents without owners, for example after the person who created them has left; these can be blocked or deleted. Risk signals from Purview, Entra and Defender require a Microsoft 365 E7 or Agent 365 licence.
Cowork takes a described task from start to finish: it gathers context from emails, meetings, chats and files, works across several tools and delivers a finished result. Who reviews the result before it is passed on is set by your directive.
This is how we put it into practice: an adoption plan per area, champions, training on your files and usage per area, with KA Partners for the strategy before it. Adoption
Who writes the AI directive, and how long does it get?
The platform owner in IT and the data protection office write it together, the management signs. Its core is a table: approved tools, their contract, the highest sensitivity level that may go in, and who reviews results before they are passed on. Added to that is the labelling rule for agents in customer contact. It rarely needs more than two pages, and it grows with every reviewed tool.
What do we do about employees who use private AI accounts?
First make the approved route convenient: Copilot Chat is open to everyone, the approved apps are in the directive, and the training shows the route on real tasks. Then the boundary: on managed devices a data loss prevention rule warns when confidential content is pasted into AI websites and blocks it for personal data, with the pointer to the directive. Data Security Posture Management shows whether the route is taken.
How big does the test set have to be?
Big enough to cover the cases of daily work: correct, incomplete and forbidden requests, documents with hidden instructions, outages of the sources. The business team starts with the questions from the pilot and adds every question the agent could not solve; real personal data does not belong in it, and for customer questions approved test data is used. A test set that never grows is the first sign that nobody maintains it.
How do we hand an agent over into operation?
With a sheet that names data, actions and approval levels, owner and deputy, test set and evaluation, cost per case, the model's retirement date and the date of the next review, and with the sponsor in Entra and the entry in the agent list. The second line for support knows trace, sources and version, and the champion in the area knows where a report goes.
What does KA Partners do, and what do we do?
KA Partners works out the AI strategy with your management: goals, the order of the areas, roles and the directive. We implement: the first case on Copilot, Copilot Studio or Microsoft Foundry, test set and approvals, training on your files and the operation afterwards. Both work from the same plan, and in the first conversation we establish where your initiative stands.
How do we notice that the rollout is holding?
By the sign in daily work that every area head set in the adoption plan, for example that receipts arrive without mail attachments, and by the figures per area from the Copilot Dashboard and the usage reports, with names concealed and evaluated under the data rule. The usage tracker shows where the rollout is holding and where more support helps, and the test set shows whether the agent is getting better.
How often do we have to re-check prices and facts?
Per application every quarter, with a date in the sheet: facts, prices, licences, rights and test set. In between, the platform owner reads the message center, which announces major changes at least 30 days in advance, and the cost alerts report on their own. The dated articles of this guide carry the date of their last review, and the chronicle shows what has changed since.
To the guide
Understanding Microsoft AI/SI, from Copilot to Foundry.
The guide shows what Copilot, Copilot Studio and Foundry take on, where your data runs and what your IT decides, with date and source.