Human-AI Collaboration

I built agents that could research a prospect, draft questions, prepare an email, and assemble a pitch deck. Then I had to read and check all of it. The preparation grew faster than my capacity to use it.

I’m redesigning that workflow around a short brief with suggested next actions. The question is whether it helps me enter the conversation with a clearer sense of what matters and what I can contribute. I want the recommendation, the evidence behind it, and what remains unknown, with detail available when I need it.

For a first experiment, choose a familiar meeting and a small set of notes. Familiarity gives you a way to judge the result. Check the facts that could change the conversation, use the brief, and notice what you still had to figure out yourself.

There is craft in explaining what is off. Sometimes I can name the correction. Sometimes I need the model to interview me, explore alternatives, or help me articulate a standard I recognize but haven’t written down.

An example gives that conversation something concrete to work with. Compare a useful result with one that missed the point. Explain why the difference matters: the intended reader, the decision, the missing context, or the tone. When a correction keeps recurring, capture it in reusable guidance and examples.

The standard still needs a human owner. An instruction can become stale, and an elegant response can be wrong. I want to improve both the result and my ability to tell whether it deserves to be used.

I can now make significant progress on several hard problems concurrently. That is exhilarating, and it changes the job. I still have to remember what each stream is doing, which decision is waiting on me, and how the pieces fit together.

Separate work where the dependencies allow it. One agent might explore alternatives while another challenges the assumptions. Give each a clear question, relevant context, and a result you can review. If one task needs a decision from another, that dependency belongs in the plan.

Leave room to bring the work back together. Comparing answers, resolving contradictions, and deciding what to use are work. When the result goes to a colleague, make the recommendation, unresolved questions, and next owner clear. A faster handoff only helps if the other person can act on it.