TL;DR
The short version
Solo agents create private productivity islands. Team agents need intentional ownership, shared knowledge hygiene, and permission design — or they amplify contradictions.
Nathaniel Whittemore and Nufar Gaspar walk four archetypes, three stop signs, and five design decisions. Build the knowledge layer now; do not burn a quarter cloning Claude.tag.
Built on The AI Daily Brief Operators Cut "How to Build Team Agents" (2026-09-29) with Nufar Gaspar.
Why solo agents stall
Most agent wins so far are personal. Companies are not. Sales promised one thing, marketing campaigns another, success discovers the gap at renewal. Multiply that by a dozen private agents with slightly different company pictures and you get an agent pool — overlapping, brittle, useless the week the owner leaves.
If your colleagues keep asking to borrow your private agent, that's probably a sign that you need to make it a team agent.
Nufar Gaspar
Four shapes, three stops
Expert agents hold pricing, compliance, or product truth — interview the holders and make them maintainers. Common-work agents need an agreed standard before you merge three "best" versions. Bridge agents sit between departments. Chief-of-staff agents own cadence, commitments, and onboarding memory — define what they may learn.
- Stop if taste is the point (voice stays private)
- Stop if nobody will own the knowledge
- Stop if it would not simplify work
Sensitive data is not an automatic stop — it is a care requirement.
Five decisions before the tool
Who it serves (roles + don't-list). Where it lives (simplest surface two people will use this week). What it knows (collect → keep out → approve → maintain). What it can touch (asker's access vs agent account vs borrowed login). Who owns conflicts, standing instructions, and evolution.
Platforms are shipping native multiplayer. Invest in shared knowledge, ownership, and permissions. Do not reinvent the harness vendors will commodity.
