Context

You have been in the meeting before. The senior person says let's revisit this next quarter and everybody in the room knows the project is dead. The new hire doesn't know. AI doesn't know. Someone has to take the new hire aside afterward and explain what just happened.

Context is the thing that's not on the page. It's the room's history with the topic. It's the unsaid agreement that A doesn't speak when B is in the room. It's the awareness that the question you were asked is not the question you were really asked, and that answering the literal version will get you politely thanked and quietly demoted.

Every AI system in 2026 is structurally context-poor. You can fill the context window with documents — and the tools genuinely help with that — but the documents don't contain the unwritten parts. They don't contain who is angry with whom, what the last reorg taught everyone, why the team will reject any proposal that mentions a particular vendor, or what the CEO means by agility this quarter as opposed to last quarter.

The people who run real organizations spend most of their day operating in this context. The decisions look small — who do I cc on this email, do I walk into the room or send the message, do I bring this up before the all-hands or after — and they're often the difference between something landing and dying. None of it is data the model has access to.

There's a class of work that's almost entirely context: management, diplomacy, negotiation, fundraising, anything involving multiple stakeholders with conflicting interests. The AI tools can summarize the meeting and draft the email, and they can do that quite well. They cannot tell you whether the email should be sent.

If your day looks like a series of small political calls — who to talk to first, what tone, what to leave unsaid — you are doing the work that doesn't compress. The summary will keep getting better. The judgment will keep being yours.