More than a year of working with AI changed how I plan and think. I’m still working out whether that’s a good thing.
AI summary
While working with AI agents, the author developed a habit of making the outcome, context, and constraints explicit before work begins, and it has started showing up in ordinary parts of his life. Agents exposed weaknesses in his own thinking; the failures he first blamed on the model were usually failures of the brief. He does not know whether the habit has made him think better.
He argues that agents did not teach the lesson directly; they compressed the feedback loop, so a weak brief shows its consequences almost immediately rather than weeks later. Hearing similar experiences from people who had never written code, he came to see the lesson as less about AI than about a broader truth: the quality of many outcomes is determined long before the work itself begins.
The essay also considers the downside. People are not models, and a useful scaffold can become a reflex. The author concludes that in market research, the more capable the tools become, the more experience and judgment seem to matter; the tool comes after the decision it is meant to serve. It ends with what he notices: a question that is really a decision no one has made yet.
Claude Opus 5.5 wrote this summary from the essay. Sergio Patiño checked it against the essay on Sep 23, 2026.
Last week I sat in my car outside the grocery store and wrote a project brief for pozole.
I was trying to decide whether I had enough time to make it that night. Dinner by seven. One store. Under an hour. Pork shoulder if they had a small enough piece. Hominy, chiles, cilantro, radishes, limes. If any of that turned into a scavenger hunt, make something else.
When I looked back at the note, I realized it wasn’t really a shopping list. It was a description of an outcome, a set of constraints, and a fallback plan.
In other words, it looked exactly like the briefs I had been writing for AI agents.
That realization felt strange because I hadn’t intentionally adopted a new way of thinking. I hadn’t read a book about planning. I hadn’t taken a course. Yet somewhere over the last year, a habit I had developed while working with AI had started showing up in ordinary parts of my life.
I have spent a lot of time with AI agents since last December. I started with code and gradually moved into research, writing, planning, and analysis. Unlike a chatbot, an agent is given a goal and expected to figure out much of the path on its own. You don’t ask it a question. You hand it work.
What surprised me wasn’t what the agents produced. It was how quickly they exposed weaknesses in my own thinking.
If I left out context, they immediately asked for it. If I wasn’t clear about the goal, they confidently headed in the wrong direction. Sometimes they delivered something polished and plausible that turned out to be completely unhelpful.
The output often looked finished before I had figured out what I was actually asking for.
For a while I treated those moments as failures of the model. Eventually, I realized they were usually failures of the brief.
The model had not misunderstood the assignment. My question had.
That realization changed the way I approached work.
Over time, I started noticing gaps before handing something off. Context I had assumed was obvious. Decisions I thought had already been made. Constraints that existed only in my head.
The more I worked with agents, the more I found myself making those things explicit before work began.
None of those ideas are new. Good managers, researchers, consultants, and project leaders have always practiced some version of them. What interests me is how I learned the habit.
The agent didn’t teach the lesson directly. It simply compressed the feedback loop.
With people, a weak brief can take days or weeks to reveal itself. Work progresses. Meetings happen. Assumptions pile onto assumptions. Eventually someone discovers that everyone has been solving a slightly different problem.
With an agent, the consequences appear almost immediately. You get an answer, see what’s missing, refine the request, and try again. Failure becomes faster, cheaper, and easier to observe.
The lesson repeats until you begin to internalize it.
But the feedback wasn’t the only thing that changed me.
After months of working with agents, I noticed that I had also spent a great deal of time watching them plan.
Agents often make their process unusually visible. They create task lists, organize work into steps, summarize findings, identify unknowns, and stop when the next decision belongs to the human. Whether those processes resemble actual thought is a different question. What matters is that I spent months observing a consistent pattern of planning and execution.
Eventually I started doing more of it myself.
I wrote things down earlier. I clarified objectives sooner. I became more conscious of the assumptions sitting underneath a request.
At first I thought this was a technology story.
Then I started hearing similar experiences from people who had never written a line of code.
The specifics were different, but the pattern was familiar. They prepared requests before meetings instead of during them. They supplied context before someone had to ask. They recognized when a vague question was actually an undecided business choice hiding behind better wording.
Most noticed the shift before they could fully explain it.
The more I thought about it, the less convinced I became that this was really about AI.
AI simply happened to be the thing exposing a broader truth.
The quality of many outcomes is determined long before the work itself begins.
I see versions of this every day in market research.
A project often arrives disguised as a research question. After enough discussion, it becomes clear that the real challenge is not methodological at all. The challenge is defining the decision the work is supposed to support.
The most valuable conversations frequently happen before a single respondent is recruited.
What are we trying to learn?
What decision will this inform?
What tradeoffs matter?
What would make this effort successful?
The answers shape everything that follows.
A clear brief does not eliminate disagreement, nor should it. What it does is move the important disagreements forward, while there is still time for them to improve the work.
Working with AI has made that easier for me to see.
The difference between a useful answer and an impressive but irrelevant one is often established before the work begins.
There is a limit to this lesson, though.
People are not models.
Some work benefits from structure and explicit objectives. Other work benefits from ambiguity, exploration, and room to discover something unexpected. Some people think through outlines. Others think through conversation. Not every challenge needs a framework before it needs a whiteboard.
That distinction matters.
There is a version of this habit I do not want to develop: one where every uncertain situation looks like a missing field in a form and every human problem appears solvable through better instructions.
Some things need to be clarified.
Other things need to be discovered.
That brings me back to the question I started with.
Has working this way actually made me think better?
Or have I simply become dependent on a process?
I honestly don’t know.
I can still plan without writing anything down. I can still begin with a half-formed idea and work it out along the way.
I just find myself trusting that approach less than I used to.
What I do know is that AI has made me more aware of the invisible preparation beneath good work: the outcome, the context, the constraints, the tradeoffs, and the definition of success.
These are the things that often matter most before anyone begins producing an answer.
Eventually everyone will have access to remarkable AI tools. In many ways, we’re already headed there.
In market research, that realization has stayed with me.
The industry will continue adopting better tools. It should. AI will help teams move faster, explore more possibilities, and spend less time on work that adds little value.
But the quality of the outcome still depends on experience, domain knowledge, methodological judgment, and a clear understanding of the decision being supported. Those things shape the question before the first prompt is ever written.
In market research, that perspective is usually earned over time. Through years of helping organizations navigate decisions, deepening category expertise, learning where methodologies succeed and fail, and seeing similar problems emerge in different forms.
Those qualities do not compete with AI.
They determine how productively it can be used.
AI can accelerate the work. It can expand what we explore and make established expertise more productive. But it still depends on people who understand what the work is meant to accomplish and who remain accountable for what the findings mean.
The more capable the tools become, the more those human contributions seem to matter.
AI does not diminish the value of accumulated experience.
It raises the value of knowing where and how to apply it.
The tool comes after the decision it is meant to serve.
The pozole note was a small version of that realization.
Before I bought a single ingredient, I had already decided what success looked like and when to abandon the plan altogether.
Working with AI has changed what I notice before I begin.
I notice when something that sounds like a question is actually a decision no one has made yet.
Next
Wildfire is a public place for Chicagoans who notice the same problem and want to do more than start over alone. It helps neighbors find one another, turn concern into action, and keep the answer—or silence—on the record.