The Meeting an Agent Can't Attend

A meaningful share of engineering work happens in rooms, and what happens there isn't a decision being announced — it's a decision being made, out of inputs that exist nowhere before the conversation starts and mostly aren't recorded after it ends.

That's a category of work that isn't automated, and describing it precisely is more useful than gesturing at "soft skills."

What actually happens in those rooms

Preferences get discovered, not stated. Someone says they want X. Three questions in, it emerges they want X because of Y, and Y is better served by Z, which nobody had considered. The requirement that goes into the ticket didn't exist when the meeting began — it was constructed during it.

Constraints surface that nobody would have written down. "We can't do that this quarter because the platform team is mid-migration." "That customer is in a renewal negotiation." True, load-bearing, and absent from every artifact.

Disagreement gets resolved by people with standing. Two teams want incompatible things. The resolution isn't determined by which is technically better — it's negotiated, and it depends on relationships, history, and who has capital to spend.

Commitment gets made. People agreeing to something in a room, in front of each other, is what makes it happen afterward. A decision document doesn't produce the same follow-through, which is a fact about people and not about documentation quality.

Trust gets built or spent. Much of what makes a senior engineer effective is that other teams believe them. That's built over many interactions and can't be delegated.

Why recording it doesn't substitute

Transcripts and summaries help — genuinely. They don't convert the meeting into an artifact an agent can act on, for three reasons.

The important part is often what wasn't said. The pause before someone agreed. The topic nobody raised. The proposal that was dropped without discussion because everyone understood why.

Participation changes the outcome. Being in the room means influencing the decision. Reading about it afterward means receiving it. Those are different roles, and only one of them is the work.

The context needed to interpret it is also unrecorded. A transcript of a meeting between people with fifteen years of shared history is not self-describing.

✅ What this means practically

Don't treat meeting-time as overhead to be minimized. ⚠️ The efficiency instinct — with implementation faster, cut the coordination too — gets this backwards. As the mechanical work compresses, the proportion of the job that's this kind of work rises. Cutting it removes the part that isn't being automated.

Convert what's convertible, deliberately. After a decision is made, write the decision, the reasoning, the alternatives, and the constraints that drove it. That's the artifact that feeds both future humans and agents, and it's a small effort immediately after while the reasoning is fresh — and effectively impossible to reconstruct a month later.

Be in the rooms. For an engineer positioning for the work that isn't being automated, this is the concrete move. The rooms where scope is set, where trade-offs are negotiated, where priorities are argued. Not because meetings are inherently valuable, but because that's where the unautomatable half of the job happens.

Bring what agents can't. The value of being there isn't availability. It's holding the context nobody wrote down — what's fragile, what was tried, what this will cost — and applying it while the decision is still open.

💡 The reframe worth making

Engineers often experience meetings as the thing preventing real work. Under the old distribution, that was defensible: implementation was the constraint and the scarce hours were the ones spent building.

That distribution is inverting. The scarce, valuable, unautomatable hours are increasingly the ones spent deciding what to build and negotiating what's possible — which is exactly what those rooms are for.

The reframe isn't that meetings became good. It's that the ratio changed, and the activity that used to be the interruption is now closer to the substance.

The takeaway

Requirements get constructed rather than transmitted, constraints surface that were never written, disagreements get negotiated by people with standing, and commitment happens because people agreed in front of each other. None of that is available to an agent, and recording it doesn't convert it. Write down the decisions and the reasoning while they're fresh, be present where scope is set, and stop treating that time as the overhead around the real work — the ratio has changed.