An inside sales manager at a hypothetical distributor has worked out how to use AI to prepare quotes: it pulls the customer's order history, flags lines where the margin looks thin, and drafts the quote for her review. Her quotes go out faster and with fewer pricing mistakes. Her colleagues see the finished quotes, but they never see the requests she typed, the flags the system raised, or the corrections she made before anything went to a customer. When her manager asks her to explain her method at a team meeting, people nod, but almost nobody tries it.
The real question underneath that scene is one many leadership teams face: how does one person's discovery become the whole team's capability? I argue that leaders should make AI learning transparent because people learn AI fastest by watching colleagues use it in real work (corrections included), so the company should go ahead and design for that visibility in a way that spreads good practice without making everyone think alike.
What One Company Learned by Working in the Open
Every, a small company that publishes about AI and builds AI tools, launched a shared AI agent that works inside Slack on October 6, 2026. In the launch essay, CEO Dan Shipper describes a problem that will sound familiar. Even at a company whose job is to stay ahead on AI, new workflows were hard to turn into team habits. He turned 30,000 of his editor in chief's past copyedits into a reusable skill and found that getting colleagues to adopt it "feels like shouting into a hurricane" (Every: Introducing the Every Agent).
Shipper's explanation is that most AI use is invisible. People work alone in a chat window, and colleagues see the finished report but never the request that produced it or the corrections made along the way. Every's initial approach (March 2026) was to give each employee a personal agent. An engineer on the project wrote on October 8 that those personal agents had a ceiling: each one knew only its owner's work, and when someone figured out a good workflow, it stayed with their agent (Every: Why We Handed Our Agent’s Infrastructure to Anthropic). Every's own summary adds that most of the personal agents faded because they were hard to maintain and no one could remember which one did what. The company replaced them with one shared agent after Shipper noticed that "people learned the most from watching someone else's agent at work" (Every: One Agent for the Whole Company, October 11, 2026).
This evidence has a specific scope. Every is a small, highly AI-literate company whose people are comfortable working in public, and these accounts come from the company describing a product it now sells. They show what one team learned over several months; they don't show what will happen in a 500-person company with a sales force, a finance team, and customer data to protect. I find the pattern worth taking seriously anyway, because it matches something most leaders already know about how people learn a craft.
Why Private AI Learning Stalls
Many companies handle AI adoption in roughly the same way: buy licenses, run training, and leave employees to experiment on their own. Training can explain what a tool does but it rarely shows how a colleague in your role handled a real task, including what went wrong and how they fixed it, which is where most of the learning sits.
The result is that discoveries stay private. One person finds something that saves hours a week, a few people hear about it secondhand, someone else rediscovers it months later, and much of the team never learns it exists. The company pays for the tools across the board while the capability builds up in a handful of individual setups.
What Transparency Changes
When AI work happens out in the open where colleagues can see it and collaborate about it, the learning sequence changes. People watch a teammate make a request, see what comes back, see the correction, try the same thing on their own work, and adjust it accordingly. When a workflow proves itself, it can be saved so anyone on the team can run it, along with the corrections that made it work.
Every's launch essay offers a small example. Its chief operating officer asked the agent to prepare a speaker agreement and the email to accompany it. The email came back as one collapsed paragraph, so he sent it back. The agent fixed the draft and updated the saved workflow so it would check the formatting next time, and now anyone at the company can send a properly formatted agreement.
Back at our hypothetical distributor, the same approach would put the sales manager's quoting work in a shared team space. Her colleagues would see her request, the margin flag, and her instruction not to apply the volume discount to a special-order line. Once the team agreed the workflow was sound, it would become something any rep could run, and her correction would travel with it.
Transparency at Scale Needs Design
A company with $50 million or $500 million in revenue can't simply put all of its AI work in one public channel. Customer data, pricing, personnel matters, and regulated information all have rules about who can see them. Different functions need different context. Visibility can also feel like surveillance, and people won't show their corrections if they believe mistakes will count against them.
Even Every's design reflects some of this. Its agent works in public threads and in private messages, it won't show people information they don't have access to without approval from an authorized user, and anyone can ask to change a saved skill, but nothing changes until the skill's owner approves (Every: Why We Handed Our Agent’s Infrastructure to Anthropic). I suspect larger companies will end up with shared spaces organized by team or function, each with its own permissions, rather than one agent for everyone.
So making AI learning transparent in your company comes with a set of decisions: which work gets shown, in which spaces, to whom, who owns a workflow once others depend on it, and how people are credited when their method becomes the team's method. Those decisions belong to leaders, and they sit as much in workforce development and culture as in technology.
Smarter, or Just More Uniform?
There is a risk in getting this right. If your best salesperson's method, your best writer's style, and your best manager's habits all become reusable AI workflows, the company gains a great deal of leverage but it may also drift toward sameness (or “group think”), with everyone quoting like the sales manager and every decision filtered through the accumulated house view.
A related experiment at Every suggests why that matters. Mike Taylor, the company's head of evals, built AI versions of colleagues from their past messages and writing, and found they learned each person's rules but not when to set them aside. In one comparison across 201 Slack threads, the real Dan Shipper went along with the way a question was framed 34% of the time; the model built to imitate him went along 68% of the time. The real Shipper more often pushed back, redirected, or added an idea of his own (Every: You Already Signed This Off (In the Simulation), October 9, 2026). That is one team's internal test of one person, so I'd treat it as an illustration. It does point to what a library of saved workflows can't supply on its own: someone willing to say, "That's how we do it, but I think it's wrong here."
The Whole Enterprise concept that “Human Value Moves Upstream™” describes the shift I expect: as AI absorbs more of a company's repeatable know-how, more of people's contribution moves toward judging when the standard way still fits. Leaders can protect that deliberately by letting anyone propose an alternative to a saved workflow, giving each workflow an owner who reviews changes, checking the most-used workflows against actual results on a regular schedule, and recognizing the people who catch a standard method going wrong.
Where to Start
Start with one team where one person's AI use is clearly ahead of everyone else's. Give that work a shared space where teammates can see the requests and corrections, within the access rules you already have. Agree on what gets saved as a reusable workflow, who owns it, and how someone can challenge it. After two months, look at who adopted it, what changed in the work, and whether anyone improved on the original.
Here are a few questions to bring to your next leadership conversation:
Where in our company is the most useful AI work happening out of sight?
Who owns a workflow once other people depend on it, and how does it get challenged?
Would our people feel safe showing the corrections behind their results?
The goal is a company where AI makes it easier for more people to think, and where anyone can still say the standard way is wrong.
