AI assistants, with human rules attached
The newest workplace pilots are less about replacing staff and more about deciding who signs off.
The pilot team: the assistant drafts, the humans decide — in that order, every time.
The assistant writes the first draft of everything: rotas, reports, replies to complaints. The humans do something the vendors didn't put on the slides — they take responsibility, loudly, in writing, for every word that goes out.
That rule, invented by a council housing team in week one of their pilot, has since been copied by half the organisations we surveyed. The assistant proposes; a named human disposes. No anonymous output, ever.
The sign-off is the product
The result is less dramatic than the predictions and more useful than the demos. Drafting time fell by two-thirds; error rates stayed flat, because the sign-off step catches what the model invents. “It's a bicycle for the mind,” one manager said, “with the brakes still attached.”
What the pilots forbid
The rulebooks share red lines: no automated rejections, no performance assessments written by the model, no resident-facing decision without a human name on it. The organisations that wrote these rules report the highest staff satisfaction with the tools — constraint, it turns out, builds trust.
Key points
- Every AI output needs a named human sign-off
- Drafting time down two-thirds; error rates flat
- Automated rejections and assessments banned in pilots
The housing team's backlog, for the record, is gone — cleared in eleven weeks. The assistant got a thank-you in the team meeting. The humans got the credit, which is exactly how they designed it.
Related stories
Signals, weekly, in plain English.
The technology stories that will matter to public life — without the hype cycle.