AI fluency spreads when you redesign the workflow first

Two colleagues at a standing desk in an open office study a laptop edged in jade, a picture of AI fluency built through shared work.

I’ve been reviewing Wharton’s latest AI adoption report, and I’m curious about one finding in particular. Managers and above were far more likely than entry and mid-level employees to say employees receive adequate AI training. It made me wonder how many AI fluency plans are, in practice, just course catalogs.

Courses matter, of course. What I’m less sure of is how much a course can do for someone who goes back to a team where nobody has time to try things and a good idea has no clear pathway to change formal ways of working.

What the study says about AI training and its effectiveness

The report, The Adaptive Advantage: Who’s Winning With AI and How from Wharton Human-AI Research and GBK Collective, surveyed about 1,400 people at US enterprises with 1,000 or more employees and more than $50 million in revenue. Most of the leaders in it like what they’re seeing so far. Eighty percent of managers and above report a positive return from AI, and 28% call it strongly positive. Yet only 7% of organizations clear all three of Wharton’s markers for success: proving value, transforming workflows, and aligning the organization.

On training, 78% of managers and above agree that employees receive adequate AI training. Among entry and mid-level employees, the figure is 56%, the largest gap of any readiness measure in the study. And 65% of managers and above agree that what holds their department back with AI is mostly people and processes.

I want to be careful with this because it is self-reported survey data, so it tells us what is associated with success, not what causes it. Adequate training is one of Wharton’s five alignment factors, so the report does not say training is a waste. It counts training as one factor among several. Wharton calls the organizations pulling ahead adaptive, and defines an adaptive organization as “one that rebuilds how its work gets done around AI and brings its people along with it.”

Where AI enablement grinds to a halt

In my experience, individual skill with AI is rarely the bottleneck. Most teams already have a few people who are very good with these tools. The trouble is that what they know stays with them as individual intellectual capital and has a hard time spreading to others.

Then there’s the route from a working prototype tool to everyday use. A vendor-published customer story on LangChain’s blog describes how things used to go at Workday, where “building a prototype agent took 4 to 5 months, with no guarantee it could ever reach production.” Workday then built an internal platform that handles authentication, security, compliance and infrastructure, so teams could focus on what the agent needed to do, not necessarily how to do it. According to the story, a team can now have a prototype running in a staging environment in under a day.

In my experience, building a platform like this is out of reach for most teams. The more important part of Workday’s story is why it got faster. Someone treated the path from idea to production as something worth transforming. When nobody designs that path, a capable person finishes the course, builds something useful, and then watches it sit in review for a quarter.

My read is that AI fluency spreads when people have dedicated time to experiment, a shared practice to draw from, and a clear path from prototype to daily use. A curriculum can’t supply any of that on its own. You can design those conditions, the same way you design a product.

Designing the conditions for AI fluency to shift the org.

If experimenting with AI is something people do after their real work, it mostly won’t happen. It needs protected time on the calendar, ideally with a teammate. Then tie it to an actual task from that week rather than a practice exercise in a vacuum.

Shared practice keeps every team from starting over. On a How I AI episode published on Lenny’s Newsletter, Figma PM Ezra Mechaber describes PM OS, a shared repo of 31 skills that every PM at Figma now uses. (A skill, in this sense, is a saved and reusable set of instructions for an AI tool.) The repo matters less than the habits surrounding it. It started as Mechaber’s personal setup, and now 60-plus PMs start from there. That’s how AI fluency moves past the few people who had it first.

The route out of the sandbox needs designing too. Teams should know how an experiment becomes an approved tool: who reviews it, what the reviewers look for, and roughly how long it should take. Governance also controls how fast a shared library grows. Teams have to decide how broad a shared skill should be. Broad skills cover many scenarios with some flexibility and keep the library small and manageable. Narrow skills fit one use case and can be more accurate, but they pile up into an app store nobody can browse. The backend processes and governance deserve the same research and design attention as anything a customer touches.

What to measure instead of completions

All of this is hard to see if the only thing you track is training completion. Completion rates tell you how many people sat through a module. They tell you nothing about whether a claim gets resolved faster or a monthly report needs fewer handoffs. NN/g’s guidance on establishing baselines makes the point that one of the most common ways UX teams undermine their own value is by measuring too late. Take a before-measurement on the actual workflow, before anything changes. Without it you can’t show that AI did anything. That makes a pilot that worked very hard to defend at budget time.

Where we start at Grand

Most of our AI work starts as research into how people actually do their job or how customers are engaging with a brand. For a quality team at a global manufacturer, that meant interviews and surveys with roughly 2,500 internal users to learn which proposed features would change their work and which only looked useful. We made the case for cutting the second group. For the AI itself, we designed the interface to show the risk and rationale behind what an agent was doing, not only its output, so reliance could match what the system could actually do, which is what we call calibrated trust. We started from how the work actually gets done, and the AI fit inside that.

At Grand, we believe people get fluent with AI inside work they’re allowed to change. So the work and its processes are what we redesign first, transparently and with the people who do it, and so AI has a baseline to prove its value against.


If you’re trying to get AI fluency to spread past the few people who already have it, we’d like to hear what you’re building, and we’d be glad to look at the workflow with you. Bring any in-progress ideas. Let’s connect

A human in the loop is a start, not a design strategy

Woman looking at her phone with her thumb on a toggle switch, circled by an open green hand-drawn loop

I’ve been nerding out on two things from the first week of November. The OCC published a framework for supervising AI agents, and Wells Fargo updated its online banking agreement to cover them. Both put a human in the loop on paper.

That’s a reasonable default. What I’m less sure of is whether anyone’s designing the actual human in the loop. Putting a person in the middle of an agent’s work is a policy call. What that person can actually see, approve and stop is a design call, and most of that is still open.

What the OCC wants from the human in the loop

The OCC’s framework (Bulletin 2026-54) is supervisory guidance for its examiners. It isn’t a new rule, and it’s careful to call its tiers and controls illustrative. There are a few parts in here worth taking in if you’re building digital products in the finance world.

It says human approval has to be “more than a nominal click.” It calls out alert fatigue and asks how many agents one person can credibly monitor. And it lists the controls you’d expect for anything that moves money: circuit breakers, rate limits, least-privilege permissions, kill switches.

Read as a design brief, it’s pretty specific, which is great. A human in the loop clicking approve on a steady stream of agent actions without the context to judge them isn’t really overseeing anything.

The framework also says revocation of outside agents “may be calibrated to the action requested.” That’s a control mechanism, but a person still has to know how to use it. At Grand we spend a lot of our time on calibrated trust: getting someone’s reliance on AI to match what the system can actually do.

A nominal click is what over-trust looks like from the inside. You approve because everything before it was fine. And if stopping an agent is hard to do and harder to show you did, a careful customer just won’t turn one on.

What the bank’s contract says about AI agents

Wells Fargo’s updated Online Access Agreement takes effect November 24, and a new Section 15 covers third-party AI agents. Instructions your agent gives the bank are “deemed to have been initiated and authorized directly by you, without further verification.” Using one is “at your own risk,” including when it acts on bad or manipulated inputs.

If you revoke an agent, you have to keep evidence that you did and hand it over if the bank asks. Until you notify the bank, you’re responsible for what the agent does, to the extent the law allows. The bank, for its part, can require you to register agents and can suspend or cut off their access.

Douglas Borthwick, who spent time at Morgan Stanley and Standard Chartered, told American Banker that telling a chatbot to stop “leaves nothing in the customer’s hands that anyone else can check.” In other words, there is no receipt. The contract asks the customer to keep evidence. Whether the product gives them any is still an open design question.

Two documents, one human in the loop

The regulator wants the human in the loop to be able to oversee for real, without drowning in alerts. The bank wants the customer responsible until they say stop, with evidence kept on hand. Both are fair, protective asks, and they picture the same person: someone who knows what the agent’s allowed to do and can shut it off with a record.

I don’t think most of the products I see are built for that type of person yet. The questions are just newer than the products.

I’d start with the concrete ones. An approval should transparently show the payee, the amount and why the agent picked them, so saying no takes as little effort as saying yes. An approval you can read in two seconds is also how one person keeps up with more than one agent. And a customer should be able to say in plain words what they’ve let an agent do, and have that match its real permissions.

Then there’s the receipt. It doesn’t need to be clever. A confirmation that names the agent, says what access ended and when, and matches the record the bank keeps would give a customer something to hand over. If stopping an agent means typing “stop” into a chat window, there’s nothing to hand over, which is Borthwick’s point.

The bank has screens to design here too. Registering an agent, reviewing it and cutting off its access are all moments someone on the bank’s side has to get right. A clean revocation record helps them as well, since if a dispute comes up, both sides start from the same set of facts.

Designing the loop

One thing I’ve noticed in our financial services work is that AI wins trust a little bit at a time, as long as the person retains ownership of the decision. On a trading-desk tool, the easy version would have handed traders a pricing window and told them to use it. We showed the reasoning in layers, with the boundaries behind each recommendation, and protected the trader’s own read of a fast-moving market. Most of the work was deciding which parts of the decision-making process belonged to the system and which stayed with the person.

On a wealth management engagement, we went in expecting to improve a self-service tool. What we found was that advisors and their clients didn’t want self-service at all. They wanted the relationship between them made stronger, more collaborative.

Agents raise the same question, except now the money moves without anyone touching it: which parts of a decision does the person hand over, and how do they take them back? In my experience you answer that by testing with real customers before the permissions model gets built, because it’s expensive to change once it’s in code.

The approve button, the scope screen (where the customer can review what an agent is allowed to do) and the off switch are what a customer remembers when something goes wrong. If you leave them to a disclaimer, you’ve pushed the risk onto your customers and called it consent. I think the answer is showing people the right thing at the right moment.

That trust gets decided by one person at one screen, choosing how much to let an agent do with their money. The regulator and the bank have both put that person in the loop. What’s left is designing agentic experiences they can actually steer with confidence.


If you’re working on how customers approve and revoke an agent, we’d like to hear what you’re building. Bring any in-progress ideas, and we’ll go through what we’d test before the permissions model gets built. Let’s connect.