Why AI-Assisted Decisions Feel Less Like Yours by Sign-Off

A head of revenue I spoke with last week walked out of a Tuesday planning meeting saying she had approved the new territory cuts. Then she paused at the door. “Or I let the model approve them and signed it.” She laughed at herself. It was the same decision either way. The bonuses were going to be the same. But the felt sense of having decided it had quietly moved somewhere she could not find.

What the research shows

A paper that came out in February in Frontiers in Psychology, in the cognitive science section, is starting to give that small confusion a name. Gaiqing Kong, at Wuhan Sports University, argues that the sense of agency people have at work has been split for a long time into the action level (the click, the signature, the words said out loud) and the outcome level (what happened next). What is missing, Kong proposes, is a third level upstream of both. Decision-level Agency. The experience of originating and committing to a decision in the first place.

The reason it matters now is that the new tools work upstream of the action. The author puts it plainly. AI-driven nudging, automated option framing, and the design of choice architecture can subtly diminish genuine agency, even when the final, overt actions remain voluntary. The choice still gets clicked. It just no longer forms where it used to form. This is the same agency question that shows up on the design side, when builders are choosing whether agentic AI defaults preserve user agency or quietly take it.

A second piece of in-window research, published in April in Electronic Markets, gives the pattern an empirical edge. Lars Andraschko and Pauline Weritz surveyed 232 platform workers in the UK and found something useful for any leader noticing the same thing. The workers in the study were still the ones signing the work. They just felt less like the ones who decided it.

What is interesting, in the same study, is that the same workers also showed a deliberate reassertion of agency, when they noticed the slippage and pulled the decision back to themselves. It did not happen on its own. It was something they did. This is what sits underneath the accountability gap boards have been talking about when they describe themselves as confident while the CEO board AI risk gap stays open. The formal authority is still there. The felt responsibility for the decision is what has moved.

What it does not tell us yet

Two careful caveats. The Kong paper is a framework piece, not an experiment, and the specific predictions it lays out have not been tested. The Andraschko and Weritz study is one survey of 232 platform workers in one country, so the transfer to a working leader in a board meeting is an editorial extension, not a settled finding. There is also a 2025 paper, out of window for this read, that found the opposite direction in marketing managers. The honest version is that the direction of the felt-responsibility shift in AI-assisted decisions is not yet settled. The leader’s own noticing is what tells them which direction it has gone this week.

One thing to notice in your work today

One small thing to watch before the next decision you sign with AI in the loop. Not at the moment of sign-off. Earlier. The moment between you having the thing you are trying to decide and the screen having a draft of the answer. That gap used to be longer. It is where the decision used to form, and it is also where the deliberate reassertion of agency, when it happens, starts. The work this week is not to fight the tool. The work is to sit in that small gap for a beat before clicking approve. It is the upstream version of the agentic AI leader reflection time the research keeps pointing at, and it costs nothing to try once this week.