In one sentence: Agentic Maturity is a level from 1 to 5 that tells you how far a person or a team has moved along the AI adoption ladder, from writing every line by hand to agents that open pull requests without anyone asking.
What it measures
AI adoption is not one thing. Using Cursor in the editor, having an agent review your pull requests, and delegating a whole task to Devin are very different stages, and the AI Adoption percentage does not tell them apart. Agentic Maturity does: it looks at every merged pull request and asks how it was built, then summarises the answer as a single level that moves over time.
Like Delivery, it is computed on the artifact, the merged pull request, not on sessions, seats or licences.
The five levels
Every pull request is placed at exactly one level, the highest one it has evidence for.
Level 1, Human-authored
All code is written by hand. Human developers are the sole authors, with no AI in the workflow. This is the traditional baseline. In Pensero: no AI signal is found on the pull request.
Level 2, AI-assisted development
Engineers stay in the driver's seat but accelerate their work with AI copilots directly in the IDE (Claude Code, Cursor, GitHub Copilot, Codex and similar). AI suggests, completes and drafts code in real time while humans decide what to accept. In Pensero: the pull request carries AI-attributed commits, or its author used an AI coding tool in the week it was merged, since these tools report activity per person and day rather than per pull request.
Level 3, AI-augmented review
AI moves beyond authoring into quality assurance. Agents automatically review code, catching bugs, style issues and potential regressions before a human signs off. In Pensero: an AI reviewer or bot (CodeRabbit, Devin Review, Copilot review, Qodo and similar) left a review on the pull request.
Level 4, Orchestrated agents
Engineers delegate entire tasks to AI agents such as Devin that plan and execute multi-step work end to end. A human initiates and supervises, but the agent does the heavy lifting. In Pensero: an agent opened the pull request and a person drove it: they instructed the session, committed on the branch or appear as a co-author. These are the pull requests whose Delivery is shared with the people behind them (see How Pensero credits Devin's pull requests).
Level 5, Autonomous agents
AI operates proactively without human initiation. Agents monitor production signals (Sentry, New Relic and other observability tools), detect issues and open pull requests to fix them; humans shift from directing work to approving outcomes. In Pensero: an agent opened the pull request and no person holds a share of it, because the session was started by an automation, an API call or a code scan and nobody steered it. This is the same work the AI Economics page reports as Agentic delivery.
How the level is calculated
Pensero averages the level of the pull requests in scope, weighted by Delivery points, and rounds the result to the nearest level. A small refactor built with Copilot and a large feature shipped through Devin do not count the same: the bigger pull request pulls the average harder towards its level. Rounding means a team that works almost entirely one way lands on that level, so both 1 and 5 are reachable.
For example, a person whose period contains 1 point of human-authored work, 3 points with AI-augmented review and 4 points delivered through a Devin session they orchestrated has a weighted average of (1×1 + 3×3 + 4×4) ÷ 8 = 3.25, shown as Level 3. Hover the card to see the exact average and the split of Delivery across levels.
For a person, the scope is the pull requests they own plus the agent pull requests they orchestrated, exactly as their Delivery is credited.
For a team or the organisation, the scope is the union of its members' pull requests, each counted once.
For an agent, only its autonomous pull requests count; the ones a person orchestrated belong to that person.
Periods with no merged pull requests show N/A rather than a level.
Where you see it
On the AI Economics page, the Agentic Maturity card shows the level for the people or teams in your current scope, and how it changed against the previous period.
On a person or team report, the same card appears in the AI section, next to the Agentic share card. It is shown once your organisation has connected an AI agent or an AI reviewer.
Hover the card for the weighted average and the share of Delivery at each level; click it to come back to this article.
How to move it up
Connect your AI tools to Pensero. A tool Pensero cannot see leaves its pull requests at Level 1. See Add AI agents to Pensero.
Turn on an AI reviewer for your repositories; every pull request it reviews reaches Level 3 at least.
Delegate well-scoped tasks to an agent and let it open the pull request. Orchestrating the session is Level 4; wiring the agent to your alerts or scheduling it as an automation nobody has to steer is Level 5.
Good to know
Higher is not always better for every team. The level describes how work is built, not how good it is. Read it together with Delivery, quality and cost per delivery point.
Only pull requests count. Reviews, tickets and documents are not placed on the maturity ladder.
AI reviews you run yourself stay at Level 2. Level 3 needs a review that an AI reviewer or bot leaves on the pull request in your git provider. If an engineer runs an AI review in their own tool (for example a Claude Code review skill) and then posts the comments as themselves, Pensero sees a human review. The pull request stays at Level 2. To reach Level 3, connect a review bot that posts its review on the pull request.
Consistent with Delivery. Because the level reuses the same pull request attribution as Delivery, re-attributing a Devin pull request to a person moves it between their levels without changing the team's total.