And Fluency is the Game Changer
Everyone uses it now. So what?
Almost everyone on your team uses AI now. I said that in The BS Test Everyone Needs, and it’s only gotten more true. Which means “do they use it” stopped telling you anything about who’s using it effectively, or actually driving impact with it.
I’ve spent the better part of two years leading AI adoption across a large organization with the broad directive of, “We need to be using AI for everything.” The question that matters now is whether they can stand behind what they did with it.
As companies try to answer that question formally, it’s easy to be distracted by how the broader “AI-pilled” crowd talks about things like usage and token burn — as if it means anything.
Usage is a meaningless number number
Usage is the number everyone reaches for first, because it’s the one you can pull out of an admin console, a bill, etc. When you look at the major signals that push people in this direction, it makes sense:
- Shopify’s Tobi Lütke told employees in April 2025 that using AI effectively is a fundamental expectation, and that it would be factored into peer and performance review questionnaires.
- Meta told employees it would tie their performance to “AI-driven impact” starting in 2026, per an internal memo Business Insider reported in November 2025.
- Then the logical end point. In April 2026 The Information reported on an internal Meta dashboard called Claudeonomics that ranked the top 250 token consumers across more than 85,000 staff. Amazon had a beta dashboard, KiroRank, that ranked employees by AI activity. Both came down. (Meta said the employee who built theirs took it down on their own.)
- Countless AI “influencers” buying up Mac Minis, multiple Claude subscriptions, and bragging about token burn and “usage” are taking over all of the feeds.
Two different things get called usage here, and they need to be pulled apart. One is adoption: having the tools set up and actually using them to improve or speed up your work. That should be mandated. I don’t have a problem with any of these companies pushing it.
The other is consumption: tokens burned, allowance used up. Making that a goal is where it goes wrong. Someone who doesn’t know what they’re doing can burn through tokens on an expensive model pretty quickly without making any impact on the business.
A token leaderboard measures exactly one thing: who’s spending the most. And the people I wrote about last time — the twelve confident pages sent after a skim of the headers, the fifteen headlines forwarded as “options” — send a lot. They’d top that leaderboard. Meanwhile it says nothing about the two groups that matter: the people barely using AI who could be driving impact with it, and the people who already are, leveraging subagents instead of burning tokens needlessly. Neither one shows up at the top.
Lattice got closer. When it added AI fluency to its own job competencies on 14 Sep 2026, it said “Success is measured by the impact you create, not by whether or how often you use AI,” and called out “vanity metrics like number of AI tools used, frequency of use, or tokens used.” Anthropic’s free AI Fluency course teaches a 4D framework developed by professors Rick Dakan and Joseph Feller with Anthropic.
I’m not going to remix either one. What I want to give you is the read I actually trust.
Fluency is how well you can put AI to work
Standing behind your work isn’t an AI skill. You don’t stop working on something until it’s ready to put out there, whether you used AI or not. Fluency is how well you can leverage AI to solve different kinds of problems, and it climbs in three steps:
- Offload the busy work. Hand AI the work you don’t like doing, so you can focus on the stuff with more impact.
- Make your best work better. Do the things that fall off when you get busy: cross-checking data sources, finding more research, updating documentation. And do more than you could alone, because it can check several things at once, find more sources, or go through a million customer comments.
- Do what you couldn’t do before. Get a design closer to done. Actually write the fix for a bug. Prototype something usable, or build it on your codebase so it’s easier to ship. Run a financial model when you know nothing about finance, or forecast things that used to be really hard to forecast.
That third step is where I tend to consider someone “fluent”, and it comes down to them being confident enough to go at almost anything with the “I’ll figure it out” mind set — and actually being able to do it.
There are a lot of “lessons” that get you there, and with any skill, you have to practice. You try new things and build upward, and that’s how you learn where things go wrong and where you need to step beyond the basics you already know. Each time you try something new, you learn something.
You’re never done learning. But you do have to know what the AI can do, where it struggles, how to enhance it, what that means for the impact of what you’re doing, and what steps you need to add to your workflow.
Fluency is a repetition thing, but not the same rep over and over. It doesn’t help to be great at having AI write one specific document for one specific purpose. You need a variety of experiences to see where it falls down. That makes it personal: it depends on the way you talk and think, how well you know certain things, and how much you understand the details.
It also takes a real foundation. If you want to use AI to build and launch an app, you could wing it. But if you don’t understand what’s happening behind the scenes, you’re going to hit a wall. Fluency is as much about growing yourself (your understanding, how you think, what you consider your swim lane) as it is about the AI’s capabilities.
The bar I hold myself to isn’t “did this save me time,” but “did I come out of this knowing more than I went in with.” Nobody logs that in an admin console.
How fluency builds
It’s not unlike building any other skill. Drumming, boxing, martial arts: for me it’s all the same. You start by doing, and by noticing what feels natural and where you need to work. Then you learn the fundamentals. It can seem boring, but knowing the basic concepts, prompting best practices and security makes a big difference.
Here’s an example. A lot of people assume that if an MCP server or connector isn’t in Claude’s library, it doesn’t exist. Several great ones are available as custom connectors, where you just add the URL endpoint. Many more run locally and need to be configured in Claude Desktop’s MCP config. Knowing it’s possible is what gives you the confidence to try. Once you take that step, several other doors open.
I use Claude for a lot of this, but this isn’t a Claude post. Most agents and platforms now support what’s been put out in the ecosystem, like agent skills and MCP, and many of them use OAuth 2.0 to manage permissions. Gemini, Antigravity, Codex, GPT, Grok: every tool has these built in one way or another. I’m not going to walk through every one. You’re smart enough to look up the equivalent for your own tool.
Underneath the three steps above is a ladder of skill, and that’s the part you can actually build:
- Foundations. What do you know about the core models and tools: GPT, Codex, Claude, Claude Desktop?
- Beyond chat. Can you connect AI to your systems to get information or post it, and use that to keep your team aligned?
- Across systems. Can you use skills and advanced prompting to have it work fluently across multiple systems, to create better deliverables or execute day-to-day tasks?
- Advanced combinations. Can you put those together to fill skill gaps and add capabilities you didn’t have?
I’ll say this confidently: you’ll never be as effective or fluent with AI if you only operate at the surface, through chat. You might figure a lot out from sheer willpower, but your understanding will be stifled. That means you’ll be slower at solving problems and getting complex things done.
It doesn’t really matter how you climb it. It can be a team doc, a shared goal, or a challenge everybody does together. What gets people through is the way you structure the skill building.
The bar I hold myself to isn’t “did this save me time,” but “did I come out of this knowing more than I went in with.”
Defining it is Subjective
You can mandate adoption. Fluency you have to build: getting comfortable enough with AI to go at any problem or challenge and actually get it done, or make a significant leap in progress you couldn’t have made on your own before. Here’s how I frame it for the people who haven’t made it past basic chat: if AI lets you move something forward you couldn’t have done on your own, that’s opportunity, not risk.
At David’s Bridal it happens much more organically. We have different problems across the business and different people using AI on them. People come to me and ask how to use it for something more complex than they’re used to, or what tooling would help them get there. A lot of the time, they’re figuring it out on their own: product setup for drop shippers, migrations to different platforms, new tooling we want to roll out. That’s work we can’t always prioritize resources for, the kind the CTO or I make time for because it’ll make a difference. Even when the team doesn’t have the bandwidth to pick up the whole thing, we can get it closer to done. There’s more on the broader picture in how David’s Bridal is rebuilding a 76-year-old business around AI.
What comes next
For me, it’s getting more confident in code reviews, security, DevOps and infrastructure, so I can move more end-to-end without feeling like I need a full review. Design is starting to go further too, which makes design systems more important: built out and enforced. And memory has always been a passion of mine. I think it’ll become more important as the structure under everything, so there aren’t misleading or conflicting facts floating around.
There’s really no limit to where it goes. At some point I’ll probably be doing the same thing with a robot in my house that does the dishes, or trying to solve some other kind of problem.
So what comes next is a growth mindset: assume you’ve never reached the top. Fluency is really the complexity of the problem you can take on with AI, and how much help you need to get it across the line. Everybody can start approaching a problem and share the output. Not very many people can implement the different pieces, connect them to the right tools, and check, test and validate the result. That’s what we’re building toward.
So here’s the one I’d actually like an answer to: how does your company decide who’s good at AI right now — a usage number, a review question, or nothing yet? And if you manage people, what’s the tell you actually trust? I read every reply.
Mike Bal is Head of Product and AI at David’s Bridal and sits on its executive leadership team. He’s building Pearl Planner — an AI-native wedding planning platform. He writes about product management, AI implementation, and building your command center at mikebal.com.
Resources
The thing I built
- bs-detector — a cold second reader for anything you’re about to put your name on. Useful before the conversation above, not instead of it.
Definitions worth reading
- Lattice: Why Lattice Is Adding AI Fluency to Job Competencies — ten competencies, measured by impact rather than usage or tokens.
- Anthropic: AI Fluency course — free, and the source of the 4D framework.
Related Leading Product posts
- The BS Test Everyone Needs — what happens when people hand over the thinking.
- I Overlooked Agent Skills — how I build the tools I actually use every day.
- How David’s Bridal is rebuilding a 76-year-old business around AI — the bigger picture behind the adoption work.