Dr. Jonah Tebaa's Seat Sort: Where Should an AI Worker Sit?
To decide where an AI worker sits, Dr. Jonah Tebaa introduces the Seat Sort, a five-question method evaluating judgment specificity, rule change frequency, sharing across teams, exception handling, and failure impact. A majority tally determines whether an e-mployee is centralised, embedded in a business team, or divided under a written split model. In Dr. Jonah Tebaa's composite case of a distributor with 600 staff, this framework placed shared order entry centrally while relocating quote drafting directly into sales.
An e-mployee is Dr. Jonah Tebaa's term for an AI worker given a defined role inside a team. Where that worker sits on the org chart carries a real cost when the choice is wrong, and Dr. Tebaa has built a short method for getting it right.
Start with the cost. In a scenario he uses, a distributor changed a pricing rule on a Monday. Its quote-drafting e-mployee kept applying the old rule, because changes to it went through a central queue and the answer arrived the following Tuesday. Eight days of quotes went out on the wrong rule. Nobody had been careless. In Dr. Tebaa's reading, the worker had simply been given the wrong seat.
Why does the usual placement go wrong?
Dr. Tebaa observes that most organisations seat an AI worker according to who built it. A central AI group builds the worker, so the worker stays central. A business team builds one, so it stays with that team. Both habits fix the seat before the work has been examined.
Each habit has a known failure. Full centralisation produces consistency and a clean audit trail, but it places the owners of the worker far from the people who notice errors, so fixes queue up. Full embedding produces speed, but several teams end up maintaining their own versions of the same worker, such as five invoice-matching workers running on five sets of rules.
What are the five questions in the Seat Sort?
The Seat Sort asks five questions about a single e-mployee. Each answer points toward the business team, toward the centre or, less often, toward a split.
- How specific to one team is the judgment the worker applies? Narrow, team-owned judgment points to the team.
- How often do its inputs and rules change? Frequent change points to the team, while a quarterly review points to the centre.
- How many teams share it? A single user team points to the team, and several teams point to the centre.
- Where do the exception-handling skills live? With the business team points to the team. With a shared desk or a technical group points to the centre.
- What would a failure cost if it leaked across teams? A contained error points to the team. An error that spreads into finance or other teams' customers points to the centre.
The tally decides the seat. A majority of team answers means the worker is embedded in the team. A majority of centre answers means it is centralised. A mixed result means a split, and Dr. Tebaa singles out one pattern: when the second and fifth questions pull in opposite directions, with fast-changing rules on one side and wide-reaching failures on the other, a split is usually the right answer.
What does a split between centre and team involve?
A split divides responsibility for one worker between two groups. The centre builds the e-mployee, tests it, versions it and keeps the change log. The business team runs it daily, resolves its exceptions and proposes changes as it sees them.
Dr. Tebaa insists that the arrangement be written down. The handoff states who may change what, and how quickly the centre must respond to a request. A team might be allowed to adjust prices and thresholds inside agreed limits, while anything that alters the worker's structure goes to the centre for approval. In his view, a split without that document is a central model with a friendlier name.
How did the distributor in the composite case apply it?
Dr. Tebaa states plainly that this is a composite, built from patterns he sees, not one named company, and that the numbers are illustrative. The distributor has about 600 staff and four regional desks, and it runs three e-mployees that were all central by default.
Order entry is shared by four desks, has one rule review a quarter, relies on a shared order desk for exceptions and sends its failures to finance. All five answers pointed to the centre, so it stayed. Invoice matching feeds finance and scored four centre answers and one split, so it stayed as well.
Quote drafting told a different story. Only sales uses it. Its rules or prices change about nine times a month, sales ops resolves its exceptions, and a wrong quote stays inside sales. Four answers pointed to the team and one to a split. The central turnaround on a change ran about six working days against a same-day need, and the worker's exception rate was 14%, compared with about 3% for the other two.
The company moved quote drafting to Sales. The centre kept the build and test pipeline and committed to answering change requests within 24 hours. Of the three e-mployees, one had been misplaced, and it was the one nearest the customer.
What should a company do after the sort?
Dr. Tebaa offers a six-step placement scorecard. A company names the e-mployee and the team that lives with its output, counts the teams that share it, and counts its rule changes over the last 90 days. It then names the three people who resolve exceptions and notes where they sit, records where a wrong output goes next, and tallies the answers. If the result is a split, the one-page handoff is written before launch.
He adds that placement is not fixed. Any worker whose answers have changed since launch should be re-seated, and he suggests running the sort on one e-mployee in the coming week. The wider design of such teams is set out in his book, The E-mployee Operating Model.
Dr. Tebaa tells the same story in his own words in Where Should an AI Worker Sit: In the Team or in the Centre?
Related evidence: Conway's law holds that organisations design systems that copy their own communication structures, so where a system is owned shapes what it becomes. (Conway's law, which says systems end up mirroring the communication structure of the organisation that builds them)
Team Topologies treats the stream-aligned team, responsible for a single business capability, as the primary team type, with a platform team building the shared internal services those teams consume. (Martin Fowler's summary of Team Topologies, which separates stream-aligned teams from the platform team that serves them)