Appearance
Track AI in your workflows
See where AI actually sits in the work, not only what it costs: which project each AI session served and what kind of work it was, which agents do work beside your people, and how much of what your teams ship was AI-assisted or written by an agent.
AI shows up in your workflows in two ways, and this checklist covers both:
- AI in people's work. Someone uses an AI tool, such as an engineer in Claude Code or a support lead in ChatGPT, and each session is tied to the project it served.
- Agents doing the work. An AI agent you plan in Flowstate, such as an assistant answering customers or a bot reviewing code, with its own owner, projects and cost.
When you've finished, Insights → Agent insights → Agent sessions lists each session with its person, project and category, the Dashboard splits spend by the kind of work, and Resourcing → Agents & copilots lists the agents you run.
An engineering or operations lead owns this checklist, with IT and security, your project tool admin and a Flowstate admin.
Workflow means the AI tool on Agent sessions
On Agent sessions, the Workflow column is the AI tool a session ran in, such as Claude Code. The project and category say which part of your work it served.
Before you start
- Track AI adoption
- Set up your projects
- Connect your project tool, so sessions can be tied to projects.
- Roll out the Cloud Proxy
1. Switch on the modules
What: Ask for AI attribution and Agents & copilots, if they aren't on yet. Who: Implementation lead. Where: Your Flowstate contact. Done when: Insights → Agent insights shows an Agent sessions tab, and Resourcing → Agents & copilots opens.
2. Check sessions land against the right people
What: Confirm sessions arrive from people across your teams, not only engineering. Who: Engineering or operations lead. Where: Insights → Agent insights → Agent sessions. See Review AI sessions. Done when: Sessions show under people's names as an Employee or Contractor, and none you expected shows the person as Unknown.
3. Tie sessions to projects
What: Look at which sessions show No project, and fix the cause. It's usually projects that aren't in Flowstate or aren't linked to your project tool. Who: Engineering or operations lead. Where: Insights → Agent insights → Agent sessions. See Tying sessions to work. Done when: Most sessions show a project rather than No project.
- A session's cost goes to its project, and to that project's initiative.
- People can pick a project for their own sessions under My AI → My sessions → Needs a project.
- To move someone else's session, open it and select Session actions → Reassign project. You need AI governance admin access.
4. See what kind of work AI does
What: Check sessions carry a Category, the kind of work such as Feature, Bug Fix or Investigation / Spike, and that the Dashboard splits spend by it. Who: Engineering or operations lead. Where: Insights → Agent insights → Agent sessions, then Dashboard. Done when: Spend by task on the Dashboard is filled in.
You can't change a session's category. Where a session's cost counts is decided by its project.
5. Add the agents you run
What: Add each AI agent that does work on its own, give it an owner, and put it on the projects it works on. Who: Engineering or operations lead. Where: Resourcing → Agents & copilots → Agents, then on each project Resourcing plan → Agents. See Agents. Done when: Every agent you run is listed with an Owner, and shows on the projects it works on.
Don't add the AI tools people use, such as Claude Code or Cursor, as agents. Those are copilots, and their cost belongs to the person using them. See The AI cost model.
6. Optional: give your own AI services an account
What: Give each of your own services that calls AI providers, such as a support chatbot, its own AI service account, so its AI use is its own line. Who: Engineering lead. Where: Settings → AI → AI Service Accounts. See AI service accounts. Done when: Each service's account shows a Last seen date rather than Never.
7. Optional: see AI in what your engineers ship
What: Connect GitHub and ask for Flow and AI impact, so you can see how much merged work was AI-assisted, agent-authored or human, and what AI cost per merged pull request. Who: Engineering lead, with a GitHub organisation owner. Where: Connect GitHub, then your Flowstate contact. The full checklist is Measure what delivery returns. Done when: Insights → Engineering insights → AI impact shows your own Delivery mix, rather than sample data.
8. Optional: be told when sensitive data goes to AI
What: Name the people who get a daily alert when high-severity AI signals need review, such as sensitive data sent to AI. Then add alerts for sensitive data and unapproved AI tools. Who: Security lead, with a Flowstate admin. Where: Settings → AI → Agent Policy → Alert recipients, then Insights → Agent insights → Alerts → New alert. See Alerts. Done when: At least one person is listed under Alert recipients, and a Data loss prevention alert is Active with someone under Recipients.
While nobody is listed under Alert recipients, the page says "No one is listed — AI policy alerts are not being sent."
You're set up when
- Sessions from your people appear under Agent sessions, against the right names.
- Most sessions show a project rather than No project.
- Spend by task on the Dashboard is filled in.
- Every agent you run has an owner and sits on its projects.
If something's not right
Most sessions show No project. The projects people work on aren't in Flowstate, or aren't linked to your project tool. See Link synced projects to Flowstate projects.
Sessions only come from engineers. The Cloud Proxy only covers company Macs. Check which teams' Macs are in scope. See Roll out the Cloud Proxy.
AI impact shows agent-written pull requests but no AI-assisted ones. Your engineers' sessions aren't reaching Flowstate, or they aren't linked to pull requests yet. See AI impact.
Someone asks whether their use of Claude Code makes it an agent. It doesn't. A tool a person prompts is a copilot, and its cost is theirs. An agent is a worker you plan in Agents & copilots.