The context we assume

Context can feel shared without actually being shared.

We rarely say everything we mean. We rely on the conversation so far, the place where it happens, and what we think the other person already knows.

Most of the time, that makes communication easier. We can say “it”, “the issue”, or “can you check?” without explaining the whole situation again.

But it also leaves room for a quiet misunderstanding: we assume that the context we are using is the context the other person intended.

I ran into this in a project-specific Slack channel. Someone asked for help, and I interpreted the question through the subject of the channel. They were asking about something else. We exchanged answers before realising we were talking about different things.

The channel had supplied context that the message itself did not contain. I had accepted it without checking.

A reasonable assumption can still be wrong

What interests me is how easily this can happen without either person saying anything obviously confusing.

A question can make sense in more than one context. So can the answer. Each person fills in the missing details differently, and the conversation continues because neither interpretation seems unreasonable.

The difficult assumption to catch is the one that feels too obvious to mention.

We need these shortcuts. Restating everything in every message would make collaboration exhausting. But the context of a conversation can change before everyone notices.

Choosing the context we give AI

The same lesson suggests a useful habit with AI: when moving to a separate task, consider starting a fresh conversation. State the goal, include the relevant facts and constraints, and carry over only the decisions that still apply.

If we are refining the same piece of work, the existing history helps. If we are solving a different problem, a new conversation lets us choose its starting context more deliberately. OpenAI’s guidance for organising agent conversations makes a similar distinction: keep a chat focused on one coherent unit of work.

The point is to give the task a clear starting place. A fresh conversation still needs a clear request.

Make the starting point explicit

Before adding more explanation, check whether you are talking about the same thing.

When asking for help, name the subject and signal when it changes. When helping, check an assumption if a different interpretation would change your answer. A short question can be enough:

Are we still talking about the same task, or is this a different one?

With people, that might mean naming the subject or opening a separate thread. With AI, it might mean starting a fresh conversation with the context the new task needs.

We do not need to remove every assumption from a conversation. We need to make the important ones visible before building on them.