Signal vs. Noise: We gave everyone a place to share AI wins. That was a problem.

A place to post about AI is not a place where people learn together. Ours has every feature you’d want, and people still defaulted to highlight reels. What changed my mind was an internal challenge, and the posts that won. The mandate template that starts the whole thing is at the bottom. Copy it.

I built my team a place to get smarter about AI together.

Then I spent a week reading what they posted there, and I had to admit most of it wasn’t making anyone smarter. Including me..

The place is called Valere Evolve. It has everything you’d want. A Q&A feed, learning pathways, topic tags people follow, a system for leveling up as you contribute. People use it every day.

And here’s what I kept seeing. Someone posts what worked. A clean output, a tool they liked. It collects a few likes and sits there. What I wasn’t seeing enough of was the other thing. Someone picking up where a colleague left off and pushing it somewhere new.

That’s show-and-tell. From a distance, it looks like a learning culture. But when you look up close, it’s just a reel. Everyone is posting their wins, but nobody in the messy middle where the learning lives.

Then we ran an internal challenge, and the posts that won rearranged how I think about this. The winners weren’t the cleanest outputs. They were the people who let you watch the work break.

One of our product managers wrote up a workflow for turning a feature spec into wireframes with Figma Make. The post wasn’t “look what I built.” It was about what she got wrong first. She’d been generating fast and cleaning up after, and the rework kept exploding, until she realized the order was the problem. Do the slow work up front, make the prompt exhaustive before anything shows up on screen, and the first pass comes back nearly done. Skip it, and the gaps spread across every screen at once. Her phrase for it stuck with me: AI amplifies upstream ambiguity. The whole team can use that. A finished wireframe, they can only admire. You can sign up for Valere Evovle and see more about the workflow. Here’s the link to Paula Lebed’s post.

Then there was Micaela. She tried to automate the transfer of meeting notes to a shared client folder, but it failed. The tool stalled on the shared folder, a failure she reported without the polish. She’d keep doing it by hand until she found a way through.

The post could have ended there. Instead, Domagoj saw it. He’d hit the same wall automating file transfers, and he put a full workaround in the replies. Log each file’s location in a sheet, have Gemini write a script to move them, and run it on a schedule. Then the part that turns a tip into real help: the access settings he’d already learned you need before any of it works.

Nobody asked him to. He reached into a problem someone else hadn’t finished solving. That is the entire idea, and it is the thing I had been waiting to see. That exchange can be found here.

Now, here’s the difference, and it’s small enough to miss. A post that shares a finished thing ends a conversation. A post that shares an unfinished thing starts one. A win is closed. All you can do is clap. But the open question, the thing that didn’t quite work, the half-formed idea, that has a handle on it. Someone can grab it.

So I’m changing what I ask people to post. Not “show us what you made.” At least one of these in every post:

• What did you try that didn’t work, and what do you now think the problem was?
• What did the tool do that you didn’t expect?
• What’s the question you’re stuck on right now?

The first time someone answers that last one for somebody else, you have a team. Until then, you have a feed of AI wins people can like and forget.

But none of it works if there’s nothing to post about. Which is why the posts come from experiments, and the experiments come from a mandate. Here’s the one I sent around ten weeks ago, stripped down so you can adapt it and send it today.

Copy, fill in the brackets, send to your team

Over the next [4 weeks], I’m asking each of you to run one AI experiment on work you already own, not a task invented for the exercise.

1. Find a microfriction point. Something in your week that costs you time or that you dread. Be specific. Not “do emails faster,” but “the two hours every Monday I lose writing a client update from raw notes.”
2. Pick one tool that might address that friction. (Starting points by role below.)
3. Write a one-line hypothesis before you start: “I believe [tool] will [specific outcome] on [specific task].” It gives you something to measure against.
4. Use it on live work, about [2 to 3 hours a week]. The task with stakes, not a sandbox.
5. Keep a running log: what improved, what broke, what surprised you, and any prompt worth reusing.
6. On [date], we share back, and the failures count double. “It didn’t work, here’s why” is the most useful thing you can bring.

Two ground rules:

1. You’re not graded on output, only on how you worked the tool and what you learned. And protect anything sensitive.
2. Agree to [your safe-use policy] and the tool’s terms before you start.

Starting points by role: Sales, Gong, or Fireflies for call analysis. Marketing, ChatGPT, Claude, or Jasper for copy and campaigns. Engineering, Claude Code, or Cursor. Ops or PM, Notion AI or Otter for notes into action items. Support, Intercom Fin, or Zendesk AI. New hire or anyone, NotebookLM to search internal docs.

The hardest part for me was resisting the urge to wrap it in a curriculum + modules and tie it to a completion badge. Don’t. A friction point, one tool, protected time. That beats any course I could build.

And one caveat, because a clean ending would repeat the mistake my last edition was about. A platform and a mandate still won’t guarantee people learn together. Dr. Edmondson at Harvard spent decades proving the idea underneath this. Teams that learn while working beat teams that don’t. So, the platform is only where it happens. The part you can’t install is people reaching into each other’s unfinished work.

I know that because the best thing that happened on Evolve this month wasn’t a feature. It was Micaela admitting her automation flopped, and Domagoj turning up in the replies with the way around it. I had nothing to do with it, other than preparing the place for the conversation to take place. They did it on their own, and I’m pretty proud, I should add. That’s the whole idea in one exchange.

So here’s the homework, and it’s two parts.

Send that mandate to your team this week.

Then come see the other half of it. The argument I keep having on Evolve is one I can’t settle alone. What’s one task you still refuse to hand to AI, even though it probably could do it? Mine’s on the list. Here’s a look inside. Come find out what everyone else won’t give up.

And let me know what surprised you most in the comments on this article. I read every single one of them.

Guy

P.S. One of these goes out every week. Ideas I’ve run through the wringer before they reach you, including the ones that don’t survive. Subscribe, and I’ll keep them coming.

Discover why leading companies trust Valere