Appearance
AI impact
The AI impact tab shows how much of what your teams merge was written with AI, what that AI cost, and how both changed since the previous period. Use it to see whether AI is changing how your teams deliver, as one part of what delivery returns. Go to Insights → Engineering insights → AI impact.
You need access to effort reporting to open it, and access to financial figures to see AI cost. Ask your Flowstate admin.
Before you start
- Engineering insights is switched on. See Switching it on.
- GitHub is connected. See Connect GitHub.
- To see AI-assisted pull requests and AI cost, your engineers use the Cloud Proxy. Without it, only agent-written pull requests show.
Human, AI-assisted and agent pull requests
Every merged pull request is counted once, in one of three groups:
| Group | Means |
|---|---|
| Agent-authored | An AI agent wrote the pull request. |
| AI-assisted | A person wrote the pull request, using AI while they did. |
| Human | Neither. |
Delivery mix shows the split for the period, and Delivery mix over time shows it month by month.
The figures
| Figure | What it tells you |
|---|---|
| AI-assisted PRs % | The share of merged pull requests where a person used AI. |
| AI-assisted PRs | How many there were. |
| Agent-authored % | The share of merged pull requests an agent wrote. |
| AI cost / merged PR | AI cost in the period, per merged pull request. |
| Total AI cost | The cost of the AI sessions linked to pull requests in the period. |
Each figure shows its change against the period of the same length just before the one you chose. To compare with a different stretch of time, such as the months before an AI rollout, see Compare delivery with and without AI.
A pull request counts as AI-assisted when AI sessions are linked to it. Bot pull requests never count.
Line authorship
Line authorship shows who wrote the lines in merged pull requests:
| Segment | Means |
|---|---|
| AI-written | Lines AI wrote — every line of an agent's pull request, plus AI's share of AI-assisted ones where it's known. |
| Human-written | Lines people wrote. |
| AI-assisted (unsplit) | Lines in AI-assisted pull requests where the split between AI and people isn't known. |
| AI-modified | Lines AI changed. |
The split in an AI-assisted pull request is only known when the AI tool on the engineer's device reports what it wrote. Otherwise its lines stay AI-assisted (unsplit) — Flowstate never guesses.
AI cost
AI cost is the cost of the AI sessions linked to each pull request. An agent-written pull request also carries the cost of the time of the person who ran the agent. See Agent pull requests and operators.
For AI spend by person, team and tool, see Agent insights. That includes tools not tied to pull requests, such as Copilot seats.
A lower AI cost isn't a saving on its own
More AI-assisted pull requests means your teams have more capacity. It only becomes money saved when headcount comes down or the team takes on more work. Read it next to what the work delivered on each project's Value tab.
Next
- Compare delivery with and without AI: read the months either side of a change, compare teams, and compare cost per pull request by delivery kind.
- Show what a project returns: what a project cost against what it moved.
- Measure what delivery returns: the set-up checklist.
If something's not right
I see a sample dashboard with Talk to us to turn on Engineering insights. Engineering insights isn't switched on. Ask your Flowstate contact.
AI-assisted PRs % is zero. Your engineers' AI sessions aren't reaching Flowstate. See Cloud Proxy. Agent-written pull requests still count.
"No pull requests in this scope". Nothing merged in the period, teams or repositories you've chosen. Widen the period or clear the filters.
A pull request shows Operator unresolved. Flowstate can't tell who ran the agent, so only its AI cost counts. See Agent pull requests and operators.