In partnership with
The prompt that revealed the real problem
I sat with a founder last spring while she ran an AI tool through a strategic planning exercise. Her prompts were good-specific, structured, well-framed.
The outputs were coherent. But forty minutes in, she leaned back and said: I don't actually know if this is answering the right question.
She was not bad at prompting. She was losing track of what she was trying to think through in the first place.
The tool was giving her answers faster than she could evaluate what she had asked. And because the outputs looked reasonable, she kept going.
The capability gap I see most often with AI is not about prompt technique. It is about something older and less discussed: the ability to think about your own thinking while you are doing it.
How owning AI deployment expands your career
Across product, ops, and CX teams, a new kind of role is taking shape: the person responsible for making AI actually work, day to day.
On July 16, three people living this shift join a live roundtable: Simone Santiago Broad (Yoco), Yelva Espinoza (Zumba Fitness), and Fin's Dave Lynch. You'll hear what the job really looks like across industries, how they carved out these roles, the skills they'd hire for, and the challenges they're tackling now. Bring your questions, since the best moments happen live.
Register for the roundtable to save your spot.
What metacognition actually means here
Metacognition-a term from cognitive psychology, first formalised by John Flavell in 1979-is the capacity to monitor your own mental process as it happens.
Not after. During. It is the difference between knowing an answer and knowing how confident you should be in that answer.
Between producing an output and knowing whether the question that generated it was the right one.
With AI, metacognition shows up in a specific way: the ability to hold two levels of awareness simultaneously.
Level one-what is the tool producing? Level two-what was I actually trying to figure out, and is this helping?
Most AI training focuses entirely on level one. Level two is almost never discussed.
The monitor and the driver
Think of it as two roles running in parallel. The driver operates the tool-writes prompts, evaluates outputs, iterates.
The monitor sits slightly above the process, asking: are we going in the right direction? Is this question still the right question?
The driver works fast. The monitor works slow. Both are needed. What AI does, without deliberate effort, is accelerate the driver and starve the monitor.
What shifted for Marcus
Marcus, a strategy consultant, was using AI to synthesise research for a client engagement. He was productive-five or six solid summaries a day, well-sourced.
After three weeks, his project lead asked him to present his overall read on the client situation. He found he could not. He had summaries. He did not have a view.
The tool had been answering questions. He had stopped asking whether they were the right ones.
He introduced one habit: before each AI session, he wrote a single sentence-the actual question he was trying to resolve, not the task he was about to run.
It slowed him down by perhaps ten minutes a day. His project lead noticed a difference within a fortnight.
What learning science shows about self-monitoring
A meta-analysis by Hattie and Timperley (2007), published in Review of Educational Research, examined over 800 studies on feedback and learning.
Metacognitive strategies-specifically self-questioning and monitoring comprehension during a task, not after-produced among the largest effect sizes observed, with an average d of 0.69.
More relevant to knowledge work: the benefit was largest in complex, ambiguous tasks where correct answers were not immediately verifiable.
That is precisely the profile of most AI-assisted work. The skill that compounds most in uncertain environments is not better tool use. It is better self-monitoring.
What is an EOR—and why are companies using it?
Opening entities in every country can be slow, expensive, and hard to scale.
That's why more companies are using EOR to hire globally faster.
See how Oyster helps teams hire, pay, and support talent in 180+ countries while staying compliant along the way.
Before the next session, one sentence
You do not need to prompt differently. You need to know, before you open the tool, what you are actually trying to resolve.
Not the task. The question underneath the task. Those are often not the same thing. The task is 'summarise this report.'
The question underneath might be 'do I trust this source enough to build on it' or 'what is the one thing I need to know before the meeting.'
The tool can help with all of those-but only if you have named which one you are working on.
This week: before your next AI session, write the underlying question in one sentence. Not the prompt. The question the prompt is meant to answer.
I have written more on how this applies across different kinds of knowledge work at promptnproductive.com.
Reply and tell us: when did you last lose track of what you were actually trying to figure out?