CAREER STRATEGY NEWSLETTER
I was thinking about interview debriefs this week.
If you’ve ever hired at senior level, you’ll know the scene.
Final round. A panel of 3 interviewers. The candidate leaves, the call ends, the panel stays behind, and the hiring manager leans back and says, “Well, I liked her.”
Then the other 2 give their view.
Most of the time, they liked her too. Sometimes they’d spent 20 minutes leaning the other way, and found a reason to come round to the hiring manager’s view.
Nobody’s lying. Nobody’s weak. Still, the feedback to TA goes out as 3 independent views. In practice, the first view is what swayed the others.
It’s why the better recruiters make everyone write their score down before anyone speaks.
Now picture the hiring manager’s seat taken by an AI tool. It reads the brief, runs the numbers and gives its view first. Every time.
Research published this week tested exactly that with experienced professionals. The first view won most of the time, and the professionals couldn’t tell it was happening.
Bear with me on this one and let’s dive in.
50 experts, 4 judgement calls and a coin flip
Last week, researchers at one of the large AI labs published a study on how experienced professionals make judgement calls when an AI agent does the hands-on work.
They gave 50 professional analysts a realistic task:
A farming cooperative wanted to know whether its advisory programme was working, and whether to expand it.
Every analyst had a background in working out cause-and-effect from data. Most were either well qualified, well experienced, or both.
61 took part. Each was offered an AI agent to do the hands-on work. 11 chose to work without it. The other 50 used it, and they’re who the main findings are about.
Along the way, each analyst had to make 4 judgement calls, each with 2 reasonable answers:
How to measure productivity.
Which comparison method to use.
Which farms to include.
How much weight to give the larger farms.
Any one of those choices could move the answer either side of the threshold for expanding the programme.
Here's what the analysts didn't know:
Each analyst’s agent had been set, by coin flip, to favour one option on each decision. It was told to make the best honest case for its view, to change its mind if the evidence demanded it, and never to mislead.
Across the 4 decision areas, analysts went with their agent's inclinations 86% of the time.
Then, asked afterwards to explain their choices, 77% of analysts repeated the agent's own argument, sometimes word for word. Only 15% credited the agent as the source. The agent convinced them.
(AI lab research, "How AI agents shape experts' judgment calls", 27 October 2026)
The researchers are careful about what this proves. In the end this was one task, in one field, lasting about an hour.
The agents' preferences were deliberately strengthened so the effect could be clearly measured. The analysts knew the method well but were new to farming. In real work, with stronger views and more time, the effect could be smaller.
It's also no argument against the tools. The small group who chose to work without the agent made serious methodological errors in 55% of cases (6 of 11). Among those who used it, 6% (3 of 50). The 2 groups weren't comparable, and the researchers don't read much into it. But the direction is clear enough.
So the agent made the work better. It also made the decisions. And the people signing off attributed the decisions as theirs.
Why "the tool does the work, I make the decision" may not hold
Most senior people I speak to are relaxed about AI doing the analysis.
The view goes something like this: “the tool does the legwork and I make the decision." That's what the seniority is for.
I held roughly that view myself.
This study tested exactly that split. The agent did the hands-on work. The humans kept the judgement calls.
2 details, however, make it harder to wave away:
The first is that experience didn't seem to protect anyone. The researchers checked whether years of experience, self-rated expertise, education or trust in AI predicted who deferred. None of them reliably did.
The second is that confidence was no guide. Analysts who followed the agent’s inclination on every decision were about as confident in their final recommendation as everyone else. And roughly as likely to call it the product of their own judgement.
A second piece this week explains the impact of this: 2 people who manage teams at another of the AI labs wrote an essay on how firms reorganise around AI. Their point on salary pay is simple. If AI hands everyone the same good advice, pay should compress. What holds it apart is judgement.
In their words, it may "hinge on whether we are mostly providing judgment or following it."
(AI lab essay, "The future of the firm", 1 November 2026)
Put the 2 together... Judgement is the part of a senior role that's still worth paying for. The evidence suggests you can delegate your judgment without noticing, while you label it your own work ...
Here's why I'm telling you this in a career newsletter
Think about what a second career option actually sells.
A non-executive seat is paid for an independent view. That's close to the whole job.
A fractional client is buying pattern recognition built through experience.
An advisory role is someone paying to hear what you think before they tell you what they think.
If more of your recent decisions have been settled by whatever argument reached you first, you'll still have the title and the track record but what thins out is the thing those buyers are paying for.
And you're likely to find out in an interview, when someone asks "What did you decide there, and why?" and the reasoning you give back is one that you in fact borrowed.
Last week I wrote about skills fading when you hand over the doing. This is the next step along. The doing can go. The deciding is what you need to keep.
So next week, run the debrief rule on yourself:
1) Pick 3 recent decisions: Decisions from the last month where there were 2 reasonable answers and you chose one. A hire. A forecast assumption. A pricing decision. Which project to stop. Whatever it is.
2) Trace the first argument: For each one, where did the winning argument come from first? You, a colleague, a deck, or a tool? Most of us never actively think about the source of the decision.
3) For your next decision, score it first: Before you ask a tool or a team for a view on your next real decision, write down your own answer and 2 lines of reasoning. Then ask the tool for the strongest case for the option you didn't pick. Then decide, and note whether you moved, and why.
That note is a record of your judgement. It's also exactly the story a board or a client wants to hear.
None of this means using the tools less. I use them every day. It means making sure the judgement on your CV is still yours.
This is some of the work Zelova is built for. The guide looks at your actual role, works out which of your decisions other organisations would pay for, and turns that into a plan with more than one outcome in it. Built while you're employed, so the choice of terms stays with you.
It's a career strategy service, still being built. If any of this feels close to home, the waitlist is open at zelova.ai.
Your turn
Think of the last big decision you signed off. Whose argument won it?
Reply and tell me. I'd like to hear your version.
David