agentiseraIntent cannot be automated
A thought on what still needs to come from us when AI can perform more and more of the steps.
It is easy to describe what you want an AI to do. Write a summary. Analyze the material. Suggest next steps. It is harder to formulate why the task needs to be done and what would actually make the result good.
I believe that difference becomes more important the more capable the tools become.
An instruction is not the whole intent
If I ask an agent to make a process faster, I have already chosen what it should prioritize. But faster for whom? What must not be lost? Can an extra check be worth the time it takes?
Technology can help me reason through those questions. But if I don't ask them, it risks solving a narrower task than the one I actually needed help with.
Getting exactly what I asked for is not always the same thing as getting what I needed.
What do I want to be able to stand by?
When AI takes more steps on its own, I need to become clearer about which decisions it is allowed to make and when it should stop and ask. What is acceptable? What requires review? What must be explainable in hindsight?
It is not just a question of better instructions. It is about thinking through the consequences of what I set in motion myself.
A useful check can be to ask myself: if the result is wrong, would I understand why? And if it is right, do I know on what grounds?
Let the tool help with the questions
I don't need to have finished thinking before I start using AI. On the contrary, the conversation can be a way to discover what is unclear. I can ask for objections, alternative interpretations, and things I have missed.
But I want to distinguish between getting help to develop my intent and letting a well-formulated suggestion become my intent without me noticing.
For me, this is part of the mental gym: to stop, choose a direction, and be able to explain the choice in my own words. Not to make every task harder, but to know what I am actually trying to achieve.