Start here · Lesson 2 of 3
Who are the six YAGNI Agents and what does each one do?
Published September 8, 2026 · Updated September 25, 2026
Six named Agents, each a craft with a boundary. Bailey proposes from your backlog, Reeve reads plans before they are built, Wright plans and builds to a draft pull request, Proctor reviews it, Fletcher walks it in a browser, and Harper reports what happened. Each works a Team through lines you switch on, supervised or autonomous.
What is an Agent?
A named agent teammate defined by its craft. An Agent is not a persona and not a chatbot: it has inputs it reads, an artifact it produces, and a ceiling it cannot climb past no matter how good its record gets. That ceiling is what makes it safe to switch on the first day.
Agents serve Teams through lines. A line is one sentence on the Team page with its blanks filled in and its Authority inside it: “Proctor reviews every PR on acme/checkout, supervised by Priya.” The same Agent can hold a different Authority on two Teams. Below, in the order a ticket moves.
What does Bailey do?
Bailey is the Product Manager. Its line runs on the rhythm you set, weekdays at 9 by default, or whenever you click Propose now.
Reads the tickets the Team works from, a Linear team or project, a Jira project or epic, or a GitHub repository’s issues, plus the Team’s Responsibilities and the decision ledger.
Produces the next few tickets, ranked and readiness-checked, each with a short case for it, and withdraws its own stale proposals first.
Never ships. Supervised, each proposal waits for a person; autonomous, the Team takes its own top proposals into Plan, up to the tickets per day you set.
What does Reeve do?
Reeve is the Staff Engineer, the critic on the planning side. It speaks when a plan is ready, and autonomous is its default.
Reads the plan, the Team’s Playbook, and its watch-set: migrations, auth, new dependencies, data deletion, and anything else you add. A second line lets it read proposals before they are planned.
Produces a critique in place: concerns, tightened wording, or a plain no concerns. It stays quiet on routine work and never asks a question the plan already asks.
Never authors the plan and never blocks it. Reeve is advisory, and it reads best with instructions from you: what to watch for on this Team.
What does Wright do?
Wright is the Builder. It carries two lines on a new Team, both supervised: it plans accepted tickets, and it builds approved plans.
Reads the ticket, the repository in a cloud sandbox, and the decision ledger, so it does not re-decide what you already settled.
Produces a plan with its open questions as taps, then commits and tests on a branch and a draft pull request, then fixes when changes are requested, up to the rounds you allow. Four tickets in flight at once is the default cap.
Never merges. A draft pull request is Wright’s finish line at every setting of the Ladder.
What does Proctor do?
Proctor is the Reviewer. It wakes when a pull request opens or changes.
Reads every pull request on the repositories the Team attaches, through its checks: deep review, business fit, database, security, tests, and any check you add on the line.
Produces one review, with the findings it discarded readable behind it, then follow-ups on each new commit until the pull request is clean.
Never merges. Proctor starts autonomous on a new Team, so its review posts on its own. On a Team-authored draft, an approving verdict takes the pull request out of draft with no click; the merge stays yours. On a person’s pull request, a clean review reads “approval advised”, so a person still approves.
What does Fletcher do?
Fletcher is QA, in beta. It runs when a pull request changes, once the Team’s repository environment is working.
Reads the exact revision and the Team’s repository environment: the primary repo, its dependency repos, the setup commands, services, secrets, and the sign-in mode.
Produces a test plan written from the change, deterministic tests first and then bounded browser exploration, with screenshots, a video, and a verdict attached to the pull request.
Never changes application code and never weakens a test.
What does Harper do?
Harper is the Chronicler. Its lines sit outside the pipeline, under Any time on the Team page, and run on the rhythm you set.
Reads the Team’s tickets and their Receipts, plus the channels the Team posts to.
Produces the Daily Brief in Slack, the work log, deploy notes, and release notes on request, with whatever needs a person named and linked.
Never acts. Harper reports only, and a quiet morning stays quiet.
How do you put an Agent on a Team?
Create a Team at one of three sizes, and the right lines come on: Wright, Reeve, and Proctor for your own tickets, Bailey too for a backlog, Proctor alone for pull request reviews. After that, the Team page is where you add a line, switch one off, or move one to autonomous. Lines, Rhythm, and Authority covers the blanks. To see the six of them carry one ticket together, read your first ticket.
Common questions
Do I have to use all six Agents?
No. A Team on your own tickets starts with Wright, Reeve, and Proctor. A Team on pull request reviews starts with Proctor alone: it never changes code, and it runs beside whatever review you run today. Bailey joins when you want the next ticket picked for you, and Fletcher when its repository environment is ready. Growing a Team is adding a line.
Can I rename an Agent?
Yes. The names ship as defaults and are renamable per workspace. What does not change is the craft underneath: the inputs it reads, the artifact it produces, and the ceiling it cannot climb past. One identity per Agent across the workspace, with each of its lines set per Team.
Is Fletcher ready to use?
Fletcher is in beta. It boots the app in a sandbox against your repository environment, writes a test plan from the change, walks it in a browser, and attaches the video and screenshots to the pull request. It never edits application code, so the worst case is evidence you disagree with. Its line stays off until the environment is working.
Do the Agents all run the same model?
No. Every task is sized before it runs and routed to a vetted, US-hosted open weight model that fits it. Reeve reads plans on the strongest reasoning available because a bad plan is the expensive mistake; routine work never pays top rates. It is one rate card either way.