.png)
September 6, 2026
The same work gets rated 9% less competent when people believe AI helped. Here is what that is quietly costing all of us, and how to change it.
A woman in one of our cohorts described her week to me. She had rebuilt her entire reporting process with AI. Four hours of work turned into forty minutes.
When I asked whether her manager knew, she looked at me like I had asked whether she leaves her front door unlocked.
"If I tell them, they think the work got easier. Not that I got better."
So she keeps it quiet. She delivers the same output on the same timeline, and the hours she saved go into more output that nobody attributes to her judgment.
I have now heard some version of that sentence from an engineer, a general counsel, a head of sales, and a man two years from retirement. The details change every time. The math never does. Saying you used AI feels like handing someone a reason to discount you.
Look at what is actually being calculated there. It is not "can I do this?" She could do it. She did do it, faster and better than the process she inherited. The question she is contemplating is whether the work still counts as hers. Whether effort that took less effort is worth less. Whether the skill she built a career on is the skill that just got automated, and admitting she uses the tool is admitting she knows.
That is not a training problem. That is a worth problem, and it is sitting inside most organizations right now with no line item and no dashboard.
If you recognized her, you are in good company. If you recognized yourself before you recognized her, that is the more useful thing to notice.
Because nearly all of us are standing on both sides of this. We hesitate before saying how the work actually got done. And we also sit in the rooms where somebody else is quietly deciding whether it is safe to tell us.
There is a dicho I grew up hearing.
"El que no llora, no mama." (The one who does not cry, does not get fed.)
It is the oldest career advice there is, in any language. Speak up. Advocate for yourself. Make sure they see you. Most of us got handed some version of it by someone who loved us and wanted us to survive the room.
I want to push on it this week, because I think we have been aiming it at the wrong person.
The gap is no longer between people who use AI and people who do not. It is between people whose AI work gets seen, and people whose AI work stays hidden because showing it costs them something.
1. Saying you used AI can make the same work look worse. Researchers gave 1,026 engineers an identical piece of code to evaluate. The only thing that changed was whether they were told AI helped write it. That one detail dropped the competence rating by 9%, and the drop was steeper when the engineer was a woman. Same work, same quality, different verdict. (Harvard Business Review)
2. The penalty comes from the manager, not the tool. A Duke team ran four experiments with nearly 4,500 people and found the part every leader should sit with. The people handing out that penalty were the ones who rarely used AI themselves. Among evaluators who used AI weekly or daily, the penalty disappeared entirely. Nobody is reading the policy. They are reading whoever is about to judge the work. (Duke Fuqua)
3. So the capability goes underground, and access was never even there to begin with. In a 2025 survey of 4,000 knowledge workers commissioned by The Adaptavist Group, 78% of the highest earners had regular access to new AI tools, against 49% of the lowest earners. (The Adaptavist Group) Meanwhile, in a PagerDuty-commissioned survey of 1,250 professionals at large enterprises, two-thirds said they had used AI tools they believed their own policy did not allow, and 77% said those restrictions were holding their careers back. (PagerDuty) That is not defiance. That is people trying to get better at their jobs, working around a system that has not caught up to them.
Put those together and you get organizations full of people quietly getting better at their work, and almost nobody able to see it happening.
That is why the usual fix doesn't work. You cannot push adoption into people who are already using AI in private. You get there by making it safe to show.
Why we believe it: We were graded on effort long before anyone paid us for outcomes. Longer hours. Visible struggle. The badge of honor that goes to whoever stayed latest. Most of us built a career on being the person who worked hardest in the room, and we were rewarded for it, so we kept the belief.
Reframe: Effort was never the product. It was the price. AI lowered the price, and we mistook that for lowering the value.
This is the belief I ask people to unlearn in every session, because it does not announce itself. It shows up as a feeling. If it only took forty minutes, did it still count? If the tool carried some of the load, what am I being paid for? People feel the break in that equation long before they can name it, and then they go quiet instead of examining it.
The new equation is impact equals success. Better thinking, better decisions, more of them, for the people you serve. Nobody unlearns twenty years of conditioning from a policy update, which is why this has to get said out loud and repeated often until it is normalized.
And notice what the old belief does to the person still holding it. She is not weighing skill when she decides whether to mention her AI workflow. She is weighing risk. Will this read as resourceful or as cutting corners? Does the person reviewing this use these tools, or do they still think of them as a shortcut? Have I built enough credibility here to survive being the first one to say it?
Anyone who has ever had their competence questioned before they opened their mouth runs that calculation faster than the rest of us. Not because they are less confident. Because they have more evidence.
Now go back to the saying. The one who does not cry, does not get fed. We aim that advice at the person staying quiet, as though the fix is for her to speak up louder.
But look at what she actually did. She weighed whether it was safe to tell her manager, and she decided it was not. Given everything above, she was probably right.
That is not a confidence problem you can coach her out of. She is reading the room accurately. So change the room, and she will tell you herself.
What to do: Two moves, and most of us need both.
When you are the one being read, stop describing what the tool did and start describing what you decided. "I used AI to draft it" and "I gave it the last three quarters, caught where it flattened the regional story, and rewrote that section myself" are the same work and completely different sentences.
When you are the one doing the reading, keep pushing adoption and put attribution beside it. Ask who has changed how their work gets done this quarter, and whether anyone besides them knows. Adoption counts how many people opened the tool. Attribution tells you whether it was safe to say so.
Most AI capability right now lives in one person's private chat history. Somebody built the prompt. They tuned it over six weeks. It saves them hours every month. Nobody else can see it, use it, or credit them for it.
Shared workspaces fix both problems at once. The context, the instructions, and the source material go in one place that other people can open. The work becomes reusable and attributable.
That second part is why it belongs in this issue. A shared project turns private craft into something with a name on it.
Run this 30-minute test:
Step four is the intervention. Steps one through three are setup.
Prompt to steal:
I am building a shared workspace for a task my team repeats. The task is [TASK]. The audience for the output is [AUDIENCE], and a good result lets them [DECISION OR ACTION]. Ask me for the context, examples, and standards you need to do this reliably for someone who has never done it before. Then write the project instructions, list the files I should add, and tell me the three places a new person is most likely to get a bad result so I can put guardrails there.
1. Run the audit at whatever scale you have. If you lead a team, one row per person and five columns: who has paid tool access, who has had more than five hours of real training, who gets the AI-related stretch assignments, who has permission to experiment during work hours, and whose AI-assisted work you have actually seen. If you do not lead a team, run the same five columns on yourself and the three people you work with most. Fill it in from memory first, then go check. The gap between what you assumed and what you find is the finding.
2. Go first, in whatever room you are in. Name one thing you did with AI this week and what you had to fix in it. Not the polished win. The messy version, including where it was wrong. Everyone in that room is running the same calculation about whether it is safe here, and one real example moves it further than anything you could say about psychological safety.
3. Change one attribution. Someone near you has done AI work that nobody has credited. Say so where it counts: in their manager's hearing, in the staff meeting, in the promotion conversation. Describe it as judgment and design, because that is what it was. You do not need authority to do this. You need to be in the room.
When someone shows you AI-assisted work, resist the first instinct, which is to ask whether AI wrote it.
Ask what they had to decide.
What did they give it? What did it get wrong? What did they throw out? What would a person without their experience have shipped by accident? Those answers are where the expertise lives, and they are completely invisible in the finished document.
This is a change to one question, and it does more for adoption than a policy memo. It tells everyone in earshot that using AI well is a skill you can see and name, which is the only thing that makes it safe to show. It works the same whether you are their manager or the person sitting next to them.
We have spent decades telling people that the answer is to speak louder. Advocate for yourself. Do not let the work speak for itself, because it will not.
I have given that advice. I have lived it. And I have come to think we hand it to the one person in the room with the least power to change anything.
AI is going to run that same pattern again, faster and at a larger scale. The people with the tools, the training, and the cover will compound. The people building capability quietly, on their own time, in a tool they are not certain they are allowed to use, will keep producing results that get credited to the software instead of to them.
Which of those happens around you is not decided by a policy. It is decided by who gets named, what gets counted as skill, and who is willing to go first. That last one does not require a title. It requires being in the room and saying it out loud.
Adoption doesn't come from pressure, and it doesn't come from another training invitation. It comes from people deciding it is safe to be seen learning.
Make it safe in whatever room you have. Then find out how much capability has been sitting there the whole time.
♻️ Share this with someone who is better at their job than anyone around them realizes.
¡Hasta la próxima, un abrazo fuerte! (Until next week, a big hug!)
