The quiet machine.

Halden’s problem was arithmetic. Town halls, training, partner announcements, client explainers — the firm produced footage faster than any two humans could cut it. Requests entered a queue; the queue entered legend. By the time a film shipped, the moment it was made for had often passed.

The obvious fix was headcount. But headcount scales cost, not judgment — and judgment was the only part of the pipeline actually working.

We audited the pipeline the way we’d audit a rough cut: where does time go that nobody would miss? Ingest, transcription, selects, versioning, captions, delivery — hours of it, none of it requiring an opinion. That’s what the machine took. Models transcribe, propose selects against a house grammar, assemble first passes, conform every deliverable to every aspect ratio, and file it all where it belongs.

What the machine never touches: the cut that decides what the film means. Every first pass lands on an editor’s timeline, not a publish button. The two humans stayed exactly where humans belong — at the point of judgment, with ten times the room to exercise it.

AI in the back office, never in the taste. The signal is a human decision; everything around it is logistics.

The backlog cleared in six weeks. Output settled at ninety films a month — five times the old ceiling — with the same two editors, now doing only the part of the job that was ever worth their time. Publish latency fell from three weeks to two days.

Nobody at Halden talks about the machine. That was the point. Good automation is like a good edit — you only notice it when it’s missing.