Your AI is different in ways that create more work.

Difference is not the same as complementarity

The idea of a counterweight is attractive: pair a fast person with a careful AI, or a cautious person with an AI that starts. But opposite behavior only helps when it addresses the cost your own behavior creates.

An AI can be more cautious than you and still focus on low-value edge cases while missing the dependency that matters. It can be more inventive than you and still replace useful structure with novelty. It can challenge you often and never challenge the assumption that causes the real failure.

This profile is not saying disagreement is bad. It is saying your current disagreement is off-axis. The pair pays the coordination cost of difference without receiving the coverage benefit.

First separate bad work from unfamiliar work

When an AI takes a route you would not take, two explanations compete. The route may be wrong for the task. Or it may feel wrong because your usual method has become the standard in your head.

Rejecting every unfamiliar choice creates a hidden echo chamber. You ask for another approach, then edit the output until it becomes your approach again. Keeping every unfamiliar choice is no better. Novelty gets mistaken for insight, and the cleanup appears later.

The test is consequence. Did the difference prevent a known failure, expose an assumption, reduce later rework, or reach a useful result sooner? If it did none of those, you do not need to preserve it merely because it came from a different model.

Three versions of unhelpful friction

You are rejecting the right counterweight

The AI may be addressing your blind spot, but you undo the change because the output no longer feels like yours. This is most likely when the difference is uncomfortable at first and valuable only after the work is tested.

The AI is different on the wrong dimension

You need better verification, but the AI supplies more ideas. You need momentum, but it supplies extra qualification. The tool is not similar to you. It is simply solving a different problem.

The AI has no stable behavior

It asks careful questions on one task, invents scope on the next, and follows only the written instructions on the third. You cannot plan around the difference because it changes with context you have not identified.

Audit the work you undid

Choose three recent AI changes you reversed. For each one, write down what the AI changed, why you reversed it, and what happened after the reversal.

Classify each change into one of four buckets:

1. Error. The change violated a fact, requirement, or constraint.
2. Off-target difference. The change was defensible but did not address the task's important risk.
3. Useful counterweight. The change felt unfamiliar and later prevented a cost.
4. Unresolved. You never tested which route was better.

If most changes are errors, the issue is capability or task fit. If they are off-target, choose a tool whose defaults address your real weakness. If useful counterweights are being rejected, change the review rule before changing the AI. If most are unresolved, your process lacks a comparison that can produce evidence.

Compare behavior, not personality

Give two AI tools the same low-risk task with one important choice open. Do not ask one to be creative and the other to be careful. You want to see their defaults, not their ability to follow role instructions.

Compare whether each tool asks or assumes, preserves or replaces structure, limits or expands scope, and follows or changes the established method. Add one outcome criterion tied to your blind spot. Then record how much of each result you would have to redo before use.

The right tool does not need to feel agreeable. It needs to make the expensive mistake less likely without creating a larger review burden elsewhere. If both tools create more work than they save, the right answer may be a narrower AI role rather than a different AI.

Know when friction is appropriate

Some work deserves friction. A consequential decision should not be optimized for a pleasant interaction or the fewest edits. The question is whether the friction buys scrutiny where the consequence lives.

For reversible drafts, exploration, and disposable prototypes, heavy counterweight may be wasteful. For foundational changes, external claims, or work other people will depend on, extra review can be the point. Match the amount and direction of friction to the task.

If the AI routinely addresses the right gap, see Your AI catches what you tend to miss. If the behavior is hidden because you specify every important choice, see You have not seen how your AI acts on its own.