Show your work
One hope for AI at work was that it would level the field. A new study of sales teams at Stripe and Checkr found the biggest gains going to the people who were already the best, because they learn new tools the way autodidacts learn anything: by trying things and comparing notes. That's also the opening for everyone else. The teams that pulled ahead ran weekly AI show-and-tells and let people build their own small tools, while a central team kept the shared data and rules clean. I just asked to try this at my day job, assuming there's interest, and time. This one is about why showing your work is how a team spreads what its best people figure out, and what that borrows from the way paramedics practice.
In partnership with
Yes, that Arnold Schwarzenegger.
He has a newsletter on beehiiv. So do Codie Sanchez, Jay Shetty, David Begnaud, Colin & Samir, and Joanna Stern.
They could publish just about anywhere. They chose beehiiv to build a direct connection with the people who want to hear from them most. And you can do the same, whether you’re world-famous or just at the start of your plan for world domination.
Assuming there's interest, and time
At my day job I'm a project manager, and Microsoft's Copilot is the only AI we're allowed to use. A few weeks ago I started building a small knowledge base and some agents for my team with it, in my own OneDrive, so nothing I broke would touch a document that mattered.
This week I read an HBR piece on what top performers do differently with AI, and sent my manager a note. I'll have to look into the full capabilities of Copilot and the integrations we have access to, I wrote, but it makes me want to set up an internal AI group to mess around building with it. Assuming there's interest, and time.
On my team, a PM's week goes to managing hundreds of small projects and dozens of larger, multimillion-dollar ones, so anything new has to fit in whatever's left over. That's why "and time" is the real ask. If it's structured and company approved, it might be an okay one. And the hour may already be on the calendar. Anyone who's sat through a lunch-and-learn has seen half the room working on their laptops, clearly not paying attention. Might as well spend that hour on something people are actually interested in, where they learn by building and the format earns their attention.
It reminded me of how we learned codes. As paramedics we practiced CPR in fake calls, as a team, over and over, for a reason: when the real one came, it was seamless and smooth. After the first couple of real ones, it got uninteresting, because it was the same every time. Good for the patient. Intellectually stimulating it is not. But that's what practice buys you. The routine part runs itself, and your attention goes to whatever is different about this patient.
AI is starting to do the same thing to the routine parts of office work. The question is where the freed-up attention goes.
Where the freed time goes
Christopher Stanton, Justin Shriber, and Simon Rudat studied sales because it mixes routine work with judgment calls. AI helped broadly. But after adopting it, managers gave higher-value work to 44 percent of their top-quartile reps and only 12 percent of their bottom quartile. The freed time went to the people who already knew what to do with it.
What set them apart was AI literacy: using the tools, learning new ones, checking the output, and knowing when a person should stay in control. The top performers didn't wait for a training to get it. They built it by trying things and comparing notes with each other.
That's the opening. If the advantage comes from trying things and swapping what worked, a team can make the swapping deliberate. The top-performing teams did: weekly show-and-tells, prompts shared between peers, and small tools built locally that a central team could later harden and roll out. At Brex, one employee built an AI tool over a weekend because it seemed useful, and the company later rolled it out more broadly.
The authors' name for the model is the best line in the piece: centralize context, decentralize discovery. Keep the data, permissions, and rules in one place, and let everyone else find the uses nobody planned for.
Brandon Sammut's team at Mind Robotics wrote the same balance into the AI principles they open-sourced, which I wrote about on LinkedIn when the flat-org debate came up. Their first principle: "You are responsible for anything you put your name on." Their sixth: "Work with AI in public where you can." The first is the center that holds. The sixth is how discovery spreads.
The strongest case against a show-and-tell is that it becomes one more meeting, or a demo contest. There's a quieter cost too. In Power Tripping, I wrote that a lot of people stay quiet so they don't look incompetent, and showing a half-working prompt asks for exactly that risk. So the claim narrows: a show-and-tell earns its hour when the room is safe enough to show the unglamorous stuff. That's what Mind asks for too, when it tells people to share the prompts and workflows that worked, "including the unglamorous ones."
The Vitals
The study. Christopher Stanton (Harvard Business School), Justin Shriber (Terret), and Simon Rudat, published in Harvard Business Review, October 8, 2026.
Who and what. Interviews with eight C-level revenue executives, plus a survey of 236 sales professionals at Stripe and Checkr.
The finding. After adopting AI, managers gave higher-value tasks to 44 percent of top-quartile reps and 12 percent of bottom-quartile reps. Each one-point rise in self-rated AI literacy, on a five-point scale, came with a 0.71-point rise in reported benefit.
How to read it. A self-reported survey at two AI-forward companies, not yet published elsewhere, and one co-author runs a company that deploys AI for sales teams. Read it as a direction to watch. It gives no rate for other teams.
The Handoff
The one conversation this week: ask one person on your team to show you the AI session behind their last piece of work, the prompts and the dead ends along with the output. One leader in the study asks job candidates the same thing: "share with me the AI session you used to prepare for this interview." If what they show you is useful, ask them to show the rest of the team.
This is my fifth newsletter. Reply and tell me one thing someone on your team figured out with AI that the rest of the team hasn't seen yet. I'll read every response.
With love,
Jim Sullivan
