The blank-box problem
Most AI products begin with an empty box. That is a clean interface, but it is rarely a clean representation of work. Before anyone can ask for help, they have to gather the thread, explain the project, name the people involved, and describe what a useful result looks like.
Team chat already holds much of that context. The request sits beside the decision that prompted it, the person who owns it, and the channel where the answer needs to return. Starting there removes a round of translation.
A request should feel like delegation
The useful unit is not a prompt. It is an outcome: compare this week with last week, find what moved, explain why, and bring back the decisions that need a person. That is how teammates hand work to one another.
An AI coworker should be able to gather information from approved tools, show the steps it plans to take, and keep the team updated in the same thread. The interface becomes the work itself, not a separate place to visit.
Return the result to the room
Finished work should come back with its sources and any assumptions made along the way. Everyone in the channel can inspect it, correct it, or continue from it without asking one person to copy the result across tools.
That is the standard we use for OpenTag: begin where the request is made, make the work visible, and return the artifact to the people who need it.
One line to keep
“The closer an AI coworker is to the conversation, the less context the team has to rebuild.”
Published August 28, 2026
Back to the blog