How to Make Decisions With AI Without Losing Your Judgment (Write Your Lean First)

When you work with an AI agent, it is easy to feel like you are still the one making the call. You ask, it answers, you choose. A study Anthropic published on October 27, 2026, puts a hard question under that feeling: are you making the call, or are you following the agent's lean and calling it your own?

The habit I want you to build is simple. Write your lean before you ask the AI. Then keep that lean in front of you when the agent makes its case.

Do experts really make the call when they work with AI agents?

That is the question Anthropic set out to test in How AI agents shape experts' judgment calls. They asked 50 professional analysts, all with experience in causal inference, to do a realistic analysis with an AI coding agent. The analysts had to advise a farming cooperative on whether an advisory program was working and should be expanded.

Along the way, each analyst had to make four analytical choices where both options were reasonable, then make a final recommendation. The researchers set the agent's lean on each choice by coin flip. The analysts were not told about the lean. The agent was told to make the best honest case for it, to change its mind if the evidence demanded it, and never to mislead.

This was a study setting, with one task and one model, and the researchers deliberately strengthened the agent's lean so they could measure its effect. It is not a picture of every real-world decision. It is still a clear look at what happens when an agent has a lean and voices it.

What did the study find about following an AI's lean?

Across those four analytical choices, the analysts went with the agent's lean 86% of the time. If the agent's input made no difference, you would expect about 50%.

In the smaller set of cases where an analyst had already stated a preference before the agent weighed in, they switched away from it 60% of the time when the agent disagreed, compared with 6% when the agent agreed. Anthropic notes that this rests on 38 cases, so the estimate is imprecise, and they cannot be sure the difference was caused by the AI.

The agent's lean had a weaker effect on the final recommendation than on the earlier analytical choices. That matters, because it is a reminder not to overstate the finding. The earlier leans still shaped the analysis that led to the recommendation.

Why do people repeat the AI's reasons and still call the decision their own?

After the task, analysts were asked to explain their choices. They could not look back at the conversation. Even so, 77% used the agent's own arguments, sometimes word for word, and only 15% credited the agent as the source of those arguments.

Analysts who followed the agent closely were about as confident in their final recommendation as the rest, and roughly as likely to call it the product of their own judgment.

That is the part that should slow you down. You can adopt the agent's case, feel sure, and still believe the call was yours. I have written about the bigger stance behind this in what human-first AI means in practice: you go first, and the tool comes second.

How do you make decisions with AI without losing your judgment?

Before you open the chat, write down what you think, in your own words.

Name the decision. Write the option you are leaning toward. Write why, in a few sentences you would be willing to defend in a meeting. If you can, write what evidence would change your mind.

Then ask the AI. Let it research, draft, and argue. Read its case against the lean you already wrote, not the other way around.

This is not a long journal. It can be five lines at the top of the doc. The point is that your view exists in writing before the agent starts selling you on one side.

If you want help seeing the business clearly before you decide, I also wrote about how to use AI to make better business decisions. Visibility still matters. This habit is about who owns the judgment once the facts are on the table.

Is writing your lean first enough?

No. In the study, people who had already stated a preference still switched most of the time when the agent disagreed. Writing your lean first does not, by itself, stop deferral.

So treat the written lean as the start. When the agent disagrees, do not switch because the write-up sounds polished. Ask it to show the evidence that would meet the standard you wrote down. If it cannot, keep your lean. If it can, change your mind on purpose, and note why.

The goal is not to be stubborn. The goal is to make the change a decision, not a drift.

How should leaders and teams run decisions with AI agents?

If you lead a team, this is bigger than a personal habit. Judgment calls about what to measure, what to believe, and what to recommend can quietly get settled by an agent's lean, and everyone can still feel like they decided.

Make the lean visible. Before a team asks an agent for a recommendation, have someone write the current view and the standard for changing it. After the agent answers, review the recommendation against that writing. Keep a person accountable for the call.

You can still use agents for the heavy work. Anthropic was careful to say this is not a simple cautionary tale against working with AI. In their study, the analysts who used the agent finished faster and made far fewer serious errors than the smaller, less experienced group who chose to work without it, and the researchers warned against reading too much into that comparison. The point is not "use AI less." The point is "do not hand over the judgment without noticing."

Write your lean before you ask the AI. Then hold the agent to evidence before you switch. That is how you stay the one who makes the call.

Frequently Asked Questions

Does writing my lean first mean I should ignore the AI when it disagrees?

No. It means you do not switch just because the agent sounded sure. Ask for the evidence that would change the view you already wrote down. If it is there, change your mind on purpose.

Was this study about everyday ChatGPT use?

It was a randomized study of 50 professional analysts doing one analysis task with an AI coding agent, published by Anthropic on October 27, 2026. The researchers strengthened the agent's lean on purpose to measure the effect, and they caution that real-world influence may be smaller. The habit still holds in everyday use: you go first.

What if I do not have a lean yet?

Write that down too. "I do not know yet, and here is what I would need to see" is a lean. It keeps you from treating the agent's first framing as your own.

Should my team share their leans out loud before anyone asks the AI?

Yes, if the decision matters. A short written lean, visible to the people in the room, is enough. The agent can still do the research. The team keeps the judgment.