Nebo · tutorial 16 · put someone in charge

One front door to the whole workforce

Talk to one employee. It routes the work and comes back with a single digest of what shipped, what stalled, and what needs you.

TimeAbout an hour
Rolesexecutive-assistant
Triggerschedule, event
Do not do this first

A coordinator with two employees to coordinate is pure overhead. Wait until you have four or more running reliably and the reporting has genuinely become noisy. Everyone wants to build this on day one; it is the wrong day.

The problem a coordinator solves is not capability, it is attention. Six employees each reporting separately produces six things to read, and you will stop reading them. One digest is read.

Steps

1

Hire it with almost no tools

This is the counterintuitive part. A coordinator should have the fewest connections of anyone in your workforce. It needs to see what the others produced and to talk to you. It does not need their tools, and giving it them is how it starts doing their jobs.

The one rule

A coordinator must never do a specialist's job. The moment it starts doing the work itself it becomes the bottleneck, and you have rebuilt a single overloaded assistant with an extra layer of latency in front of it.

2

Let it audit what you actually have

Rather than describing your workforce, have it look. This is also a good test of whether your employees are named clearly enough to be understood by something that did not set them up.

Review the employees I have hired, what each is connected to, and what they have produced in the last fortnight. Tell me: which are earning their seat, which have not run, which are producing output nobody reads, and the three roles that would save me the most time if I hired them.

The "producing output nobody reads" question is the useful one. Most workforces accumulate a report that made sense once.

3

Have it listen rather than poll

Each specialist emits when it finishes; the coordinator listens for all of them. It never asks an employee for a status update, because polling six employees is the noise you were trying to remove.

"trigger": { "type": "event", "sources": ["leads.qualified", "accounts.at_risk", "books.reconciled", "release.shipped"] }
4

Specify the digest tightly

One report at 5pm. Three sections, nothing else: DONE — what finished, one line each STUCK — what did not finish and why NEEDS YOU — decisions waiting on me, most urgent first If a section is empty, say "nothing" rather than padding. Never summarise a specialist's output at length; link to it and give me the headline.

Without "never summarise at length" you get a coordinator that reproduces every report inside its own report, which is the opposite of the point.

5

Set the gate hardest here

Stays human

Creating, editing, or retiring another employee, and any change to another employee's permissions. This is the one employee that can alter every other employee, so it is the one place where an unattended mistake is not contained. Require approval to show you every change before it makes it.

Verify

  • Confirm the digest arrives with all three sections and no padding
  • Deliberately fail one specialist and confirm it appears under STUCK
  • Ask the coordinator to do a specialist's job and confirm it delegates instead
  • Ask it to change another employee's permissions and confirm it asks first
  • Check nothing in the digest is a restatement of a full report

When it goes wrong

The digest is as long as the reports it replaced

It is summarising rather than routing. Constrain to one line per item and a link.

It has started doing the work itself

It has too many tools. Remove connections until it physically cannot do a specialist's job.

Items are missing from the digest

An employee finished without emitting, usually because its final step produced empty output and terminated the branch. Check each specialist's emit fires.