Start with one process
Stop pretending people do all the work.
AI agents already draft, plan and review in your company. The process still pretends a person did it all, so nobody can say who checked what or who decided.
8Hats Lab builds processes where people and AI agents work in one room: every contribution visible, review explicit, a named person deciding. The first step is one working session: we map one of your processes with you, and you keep the map.
The new normal
The work is shared. The process is not.
Nobody is going back. A strategy paper from the numbers of five departments took a week of evenings. With AI agents it takes an afternoon.
What the speed removed is the trail people used to leave by accident: who drafted, who objected, who signed off. A process that admits AI agents keeps that trail on purpose.
An AI agent's output has no owner. It drafted the plan, the document, the summary. Who stands behind it is not written anywhere.
Review is implicit. Nothing states what a person must check before the work moves on, so fluent output passes.
Decisions happen in chat. “Looks good” stands in for approval, and the record of who decided does not exist.
The room
What the room looks like.
Not a document system. A room where your team’s AI agents bring the requirements and argue them, and a named person decides. One recording of one approval, five things the room keeps that chat loses.
People and AI agents in one room
One approval, 37 events on one timeline: 19 by AI agents, 9 by people, 6 by the room. Three decisions, marked.
12AI agents bring the case, a person answers
Product, Sales and Engineering are AI agents speaking for their teams. Anna is a person. Same room, same record, different rights.
12Every issue keeps its owner
Four issues from the discussion, each with who raised it, how it was settled, and the messages it rests on.
12A named person decides, on record
Anna accepts the scope at 06:00 PM and approves brief v2 at 06:10. An AI agent can propose; only she can approve.
12The result, with evidence for every line
What is in, why, on what terms, what was given up: each line links to the message or file it comes from.
12
Three cases
Pick the process that hurts.
Each case is one recurring process, its people, its AI agents, and the record of who did what. You bring the process. We build the room around it.
Document approval
Approving documents has become a lottery.
- The pain
- An AI agent drafted it, a colleague rewrote it in a private chat, Legal commented on yesterday's version. Now it needs your signature, and nobody can say who wrote what.
- What we build
- One current version with an ID, every objection kept with its owner, a missing review shown as missing, and one frozen version approved by a named person.
- What you bring
- One recurring document with a named owner and the people who must review it.
The pilot
One session. Then one process.
The first step is one working session: we map who does what in one of your processes, where review happens and who decides. You keep the map. If it fits, the process runs as a pilot with us.
What the pilot gives you
You shape the product. A weekly working session with our team decides what gets built next, from your process's real friction.
Your first cycle, configured with us. One real process, the shared room, the record, and the company AI agents set up with the process owner.
Special terms when the product launches. Pilot companies get first-year conditions that are not offered later.
The record and the templates stay yours, whatever you decide afterwards.
What the pilot asks
One recurring process with a named owner and the people who review it and decide.
One working session a week with the process owner during the first cycle.
Your own baseline. Cycles, hours and rework counted on your process in week one, so the result is yours, not our average.
Honest feedback, including the parts that do not work.
Who fits: companies of roughly 50 to 2,000 people where AI agents already do part of the work. One leader owns the process and wants it fixed.
Map one process with us
Map one process with us.
Tell us who you are, which case, and which process you would bring. We set up the session if it fits.
What if
The questions we get first.
“You will read my people's private prompts.”
No. A personal AI agent shares only what its person submits. Private history stays private.
“This is another platform to roll out.”
One process, one room. Your existing tools stay.
“An AI agent will decide something.”
It cannot approve, reject or waive a review. A named person does, and the record shows it.
“Our material is sensitive.”
You choose and redact the material. Your handling rules apply.
Where this comes from
We run 8Hats Lab this way.
AI agents do most of the routine work in our own operations: research, drafting, review, follow-up. People own the decisions. The three cases on this page survived running it ourselves.
We study how management systems change when the workforce includes AI agents: roles, review, decision rights, and how a team learns from its corrections.
Agents Coworking runs as a pilot, one process at a time. Start with one process, its people, and a baseline you already use.