Seats and pairing
A delivery team works because everyone knows their job. aayudha gives its agents the same clarity: fourteen seats, each a role with its own responsibilities, such as Product Strategist or Engineering Manager. Every verb belongs to exactly one seat, so there is never a question of who is accountable for a step.
Every seat has an agent twin. A paired human holds the gates.
Twins and humans
Each seat has an agent twin: the agent half that does the seat's work in your host, such as interviewing, drafting, building, or checking. A seat runs in one of two modes:
- Solo: the twin works alone, and anything that needs a person goes to the founder.
- Paired: a person is attached to the seat. They answer the twin's questions, review its work in their layers, and give the approvals that only a person may give.
You pair a person with a seat in one command, and you can see every seat's mode at any time:
aayu seat add product-strategist --human <github-login>
aayu seat list
The fourteen seats
| Seat | In plain words | Some of its verbs |
|---|---|---|
| Founder | Final authority; sets up repositories and accepts decisions | init, decide |
| Product Strategist | Turns ideas into clear, checked work | grill, slice, ready, triage |
| Code Architect | Plans how each story is built, layer by layer | plan |
| Data & AI Architect | Designs the data and evidence behind numbers and AI answers, and measures them | chart, measure |
| Engineering Manager | Reviews, ships, and keeps delivery healthy | inspect, ship, status, retro |
| AI Native Engineer | Builds the software and runs the checks | build, verify, evolve |
| Production Engineer | Runs the software in production: incidents, SLOs, runbooks, rollbacks | respond |
| Quality Architect | Owns the floors and the proof packs | proof |
| UX Researcher | Designs and reviews the screens | research, wireframe, review, polish |
| DevOps Tinkerer | Releases, installs, and keeps the tooling working | release, install, doctor |
| Revenue Architect | Sets the targets and runs experiments | experiment, readout |
| Mentalist | Explains work to any audience | brief |
| Marketer and Documenter | Writes docs and launch copy from proven work | tell |
| Auditor | Checks the record and reports on it | audit |
In the story: in benefits-qa, the Product Strategist's twin interviewed the founder for Goal 125, the AI Native Engineer's twin built the stories, and the Engineering Manager's twin, running in CI, reviewed each pull request. Every seat ran solo, because the repository has one person.
Why seats help you
- No gaps. Every step has an owner, so nothing falls between two agents.
- Different eyes. The seat that reviews is not the seat that built, and its model comes from a different vendor.
- Room to grow. Start solo. Pair a teammate with the seats they care about when they join, and their reviews become required for their layers.
On a small team you do not manage fourteen seats. They all run solo by default; you pair a person only where you want their review.
Every twin, its layers, and its model lanes are on your agentic team. Full details are in the RACI table. Terms are defined in the glossary.
Under the hood: seats, verbs, and their RACI are declared once in the verb registry; see the verb registry and code generation.
What's next
Seats own parts of the code, called layers. See the nine layers.