Why live events still need human judgement in the age of AI

I want to be specific about this rather than defensive, because the useful cases are real and quite large.

Turning an unstructured brief into a structured draft is genuinely valuable — a client sends a document, and instead of an afternoon of transcription you get sessions, speakers, rooms and durations to correct. Finding things in a large event is valuable: which sessions have no room, which speaker is double-booked, which turnaround is shorter than any other turnaround in the building.

And answering questions about the event you are actually running — what is in Hall 2 after lunch, who is on the panel at 15:20 — where the answer cites the record it came from.

Event Brain proposes. A person decides. Nothing it suggests enters the event until someone approves it.

This is not a hedge and it is not a phased rollout toward eventual autonomy. It is a description of where the information lives. The decisions that matter in a live show depend on inputs that were never in the system and often never written down anywhere.

The model has the programme. The producer has the room.

A suggestion you cannot inspect is a suggestion you cannot responsibly accept, so proposals arrive as changes you can read before they are changes you have made.

What it wants to change, what it would affect downstream, and what it read to get there. Approve it, edit it, or throw it away. The default is that nothing happens.

Here is the part that does not fit in a schema.

A speaker is thirty seconds into a twenty-minute slot and you can hear that they are not going to make it. The audience is warmer than the run-through suggested and the Q&A wants another ten minutes. The client’s chief executive has just walked in at the back. The panel before yours overran and the right answer is to absorb it rather than compress the thing that actually matters.

Every one of those is a decision made from information that exists only in the room, held by someone who has done this before. No amount of context window substitutes for it.

Use it for the transcription, the search and the arithmetic. Use it hardest in the planning weeks, where a wrong suggestion costs nothing because you are going to review it anyway.

Trust it least in the last hour before doors, which is exactly when the pressure to accept a plausible answer is highest and the cost of a wrong one is worst.

Founder, EventBlok. Live, hybrid and broadcast producer, showcaller and technical director, in production since 1996.

About EventBlok