A client bought a second AI to summarise the summaries the first AI was producing

It started, as these things do, with a reasonable problem. Meetings generated notes. Notes generated actions. Nobody read either. So they bought a tool that summarised every meeting into six bullets and pushed them to a channel.

The channel now receives one hundred and forty summaries a day.

The second purchase

By April the summaries had become the thing nobody read. Rather than reduce meetings — which was discussed, in a meeting, which was summarised — they procured a second system to summarise the channel into a daily digest.

The digest is well written. It is also, structurally, a summary of summaries of conversations, and by the time a decision reaches it, the decision has been compressed three times by three systems, none of which were present when it mattered.

“We asked the digest what we had decided about the Munich contract. It said the team had aligned on next steps. It was right. There were no next steps.”

The audit

We ran a reconciliation: forty decisions, traced from digest back to source. Six had survived intact. Twenty-two had lost their owner. Nine had lost their deadline. Three had acquired a deadline that was never proposed by anybody, human or otherwise, and one of those three had been actioned, on time, by a team that had no idea where it came from.

That project shipped. It is the best-performing initiative of the quarter and nobody will admit to authorising it.

Recommendation

We recommended a third layer. It closes the loop, produces a single weekly artefact, and — critically — is priced per seat. They signed in eleven days. The 6/7 score improved to 4.4.