The Manager's Guide
Running Composa with your team without breaking trust · Last updated 2026-07-09 · Ousios LLC
Composa gives a team a shared vocabulary for how each person works — and any shared vocabulary about people can be used well or badly. This guide is the difference. It's short on purpose: read it before you send the assessment link, and keep the debrief script handy.
The rules that are never optional
- Never for hiring, firing, promotion, compensation, or performance review. Not as input, not as tiebreaker, not as "context." This is in our terms, and it's the fastest way to destroy the trust that makes the tool useful.
- Never covertly. Everyone knows the team is doing this, why, and what will be discussed. No taking the quiz "on behalf of" someone, no profiling absent colleagues.
- Results belong to the person. Each teammate decides whether and how much of their result to share. Invite sharing; never require it.
- Archetypes describe, they don't confine. A result is how someone tends to work right now — role, season, and context shift it. Treat it as a snapshot, not a diagnosis.
Before: how to introduce it (2 minutes, verbatim if you like)
"I'd like us each to take a 90-second working-style assessment. It's not a personality test and it will never touch anyone's review — I can't see your answers, and you choose what to share. The point is to figure out, together, which AI workflows actually fit how each of us works, instead of adopting tools one-size-fits-all. We'll compare notes as a team on [day]."
Three things that sentence gets right: it's voluntary framing (not surveillance), it names the concrete purpose (AI workflow fit, not judgment), and it sets a shared moment so results become a conversation, not a file.
During: the 45-minute team debrief
| 0–5 | Re-frame. Repeat the rules above out loud — especially "this never touches reviews." Saying it in the room is what makes it true for people. |
| 5–15 | Voluntary reveals. Go around; each person shares as much of their result as they want — archetype, family, or just "one thing it got right, one thing it got wrong." The disagreements are as useful as the confirmations. |
| 15–25 | The team map. Build the team in Composa's team view from what people chose to share. Look at the composition together: where the team is deep, where it's thin. Let people react — "no wonder our retros die" is the sound of it working. |
| 25–40 | Pick 2–3 workflows. From the recommended plays, choose a small number that fit real pain the team already feels. Assign each a volunteer owner — ideally the archetype it's matched to, but interest beats typecasting. |
| 40–45 | Set the check-in. Book a 15-minute slot in ~30 days: which workflows stuck, which didn't, what to swap. The follow-through is the whole game. |
After: retakes and drift
- Retake on real change: new role, new team, or every 6–12 months. Working styles drift; stale labels calcify.
- Near-ties are honest: if someone's result showed a close call between two archetypes, read both. The family layer is the reliable signal (see how Composa works).
- Re-map when the roster changes: the team strategy recomputes when composition changes — a new hire is a reason to revisit the plays, not to hand them the old plan.
Anti-patterns — what using it badly looks like
| If you hear (or say) this… | …do this instead |
| "Classic Maverick. That's why he's like that." | Typecasting closes the conversation the tool exists to open. Styles explain tendencies, not incidents — ask what happened, don't file it under a label. |
| "Give it to the Closer, Closers ship." | Assign by interest and growth, informed by style — never by label alone. The archetype suggests how someone might approach work, not which work they're allowed. |
| "I'll mention her archetype in the review as context." | Hard no — see the rules. If a style-related behavior matters for feedback, describe the behavior itself. |
| "Let's see who got the 'best' archetype." | There is no best; the model is compositional. If your team reads a ranking into it, re-run the 0–5 minute re-frame. |
| "You said you were an Anchor six months ago." | People update. That's the point of retakes — treat the newest result as current and the old one as history. |
When someone disagrees with their result
Perfect — that's a feature. The person is the authority on themselves; the result is a conversation starter, not a verdict. "What did it get wrong?" is often the most productive question of the whole debrief: the answer usually names exactly the context the 12 questions couldn't see. Nobody should ever feel they must defend themselves against a quiz.
What you're actually buying with all this care
Handled this way, the assessment produces the thing AI rollouts usually lack: a team that opted in. People adopt workflows matched to how they already work, chosen in the open, owned by volunteers, checked at day 30. Handled carelessly, the same tool produces labels, suspicion, and one more dashboard nobody opens. The 45 minutes above is the difference.