McKinsey, the firm that Perfected the business slide deck, told the world its own consultants are sending fewer slides and building interactive apps instead. That is not a quirk of one firm. It is the leading edge of a shift most leaders have not noticed yet. We spent three years teaching people to chat with AI, and most companies never even finished that job. While they were stuck on it, a second skills gap opened underneath them: the distance between chatting with AI and building with it. Adoption is no longer the story. The divide of 2026 is between the people who ask AI questions and the people who build with it.
🎧 In this week’s companion episode, Elizabeth and I follow the data from 370,000+ AI users to map the two skills gaps, show why a simple app beats a smart chat, and name the one thing a leader can build this week. Listen on Apple Podcasts | Spotify
In “The Worst AI You’ll Ever Use” we mapped the first gap: the human skills of reading, critical thinking, and communication. In “AI Is Working, Your Strategy Is Not” we showed adoption running ahead of strategy. Today we follow the gap that opens next, once chatting is solved.
📈 Adoption Is Nearly Solved. Skill Is Not.
In technology and professional services, AI use at work is now 70% or higher. Most users are casual, with only around 5% using it daily across all sectors, but we see real work being done with AI.
The trouble is that people are using a tool that can build software and run advanced research to write a slightly better email. They lack the skills to go beyond.
Over 8 in 10 are using tools the company never approved, and if the company is not providing the tool, it is not providing the training either. Of the AI budget that does exist, about 7% goes to people and 93% to technology.
Vendor training teaches you the vendor’s product: where to click, which button, a demo of “look, it wrote an email.” It does not teach you how to rethink a workflow you have run the same way for 20 years, or how to check whether the output is even true. That is a thinking problem, and almost no one is selling it.
🧩 Two Gaps, Not One
The mistake is treating AI skill as a single thing. There are two gaps, and they are not the same problem.
| Gap 1: Using AI | Gap 2: Building with AI | |
|---|---|---|
| The skill | Reading, critical thinking, communication. Talk to it clearly, give it context, catch the false claim. | Create small apps and agents, orchestrate them, and point them at something that matters. |
| The risk | Weak skills make you confidently wrong at scale. | You stay stuck typing questions while others build the answer. |
AI is an amplifier. Strong skills, it makes you sharper. Weak skills, it makes you confidently wrong at scale. As IDC put it, without the judgment to use it well, AI just makes an organization faster at making bad decisions. In tech and professional services, most people have cleared Gap 1. The danger now is Gap 2.
🏗️ A Simple App Beats a Smart Chat
This is the freshest finding in our data, and it surprised even us: the biggest personal wins do not come from chatting, and not even from building personal agents. They come from building small, single-purpose apps.
Picture a messy dataset. A beginner asks the AI questions about it in a chat, back and forth, back and forth, and a hallucination can slip in along the way. The advanced user says: build me a small app that lets me interact with this data, with charts, a filter, a slider. In a few minutes they have a tool they can click through that shows in one glance what fifty chat messages never would. The chat is the question. The app is the answer.
That is exactly what McKinsey just discovered at scale. As Business Insider reported, technology leader Kate Smaje said slide-deck use dropped sharply in a couple of months, and engagement manager Louis-Charles Généreux built what he calls a client visualization hub, a website that keeps about 70 people on a project current in real time. No more emailing version 14 of a deck and hoping everyone opens the right one. It is not a chatbot. It is software, built in an afternoon, by someone who is not an engineer. For 50 years, knowledge work meant producing documents. That output is now becoming a small piece of software, and most of the workforce has no idea it is even possible.
Only a thin slice of people have figured this out so far.
And the path up is not a leap. An agent is just AI you give a job to that can take a few steps on its own. You start small, an agent that sorts your inbox, one that drafts a weekly report, and once you trust the small ones you connect them so one hands off to the next. That is orchestration. Chat, then a simple app, then a simple agent, then orchestration. Almost everyone is stuck at step one, and being told by their vendor that step one is the whole journey.
🧪 Nobody Learns This From a Slide Deck
So how do people actually cross these gaps? Not the training deck. Among people who are genuinely proficient with AI, more than 70% say they got there one way: experimentation, trying things, and learning to ask the AI itself for help.
The preference is just as lopsided as the result.
| How people want to learn AI | Share |
|---|---|
| Hands-on experimentation | ~75% |
| Slide decks, vendor videos, click-here guides | ~5% |
The entire training industry is built around the method about 5% of people prefer, and almost no one learns from. In every client engagement, the unlock is the same: permission to experiment. Try something you are not sure AI can do. Fail. Adjust. Try again. You cannot lecture someone into this. You can only let them play, and the frontline teams benefit most.
🪑 The People Setting Strategy Are Furthest Behind
Here is the part of the data that should stop a board cold. Take the super users, the top 2%. You would assume they sit in leadership. It is the opposite: fewer than 1 in 5 hold a leadership role. That is just 4 out of every 1,000 workers. On the fundamentals, leaders trail their own teams by about 20 percentage points. The people setting AI strategy, signing the budgets, and redesigning the workflows are often the least skilled in the building at the very thing they are deciding about.
This is not only our data. The Adecco Group surveyed 2,000 leaders across 13 countries who oversee about 8.6 million workers. Only 31% said their own leadership has enough AI skill to grasp the risks and the opportunities. Two independent datasets, the same hole.
There is one exception: the AI-native companies, like Anthropic and Cursor, where the founders are daily builders and leadership sits at the advanced or super-user level. It shows up in everything, the strategy, the org chart, the way the work is designed. In the companies built around AI, the leaders are among the most skilled. In everyone else, they are the least. It connects straight back to a pattern we keep seeing: in 9 of 10 failing deployments, the leader signing the checks is not a daily user. You cannot point a ship somewhere if you have never felt how it steers.
🧱 The Floor Underneath All of It
Every skill in this issue sits on top of basic reading, and that floor is not solid. In the United States, more than 70 million adults cannot reliably understand what they read. One in four. About 130 million read below a sixth-grade level, and the latest national assessment showed those scores got worse over six years. When we talk about people building apps and orchestrating agents, we are describing the top of a mountain whose base is eroding. The amplifier does not fix that. It widens it. This is not only a corporate problem. It is a societal one.
💰 The Gap Already Has a Price
On both sides, the gap is now showing up in the numbers. Workers with real AI skills command a 62% wage premium, according to the PwC 2026 Global AI Jobs Barometer, up from 57% a year earlier, and the builders sit at the very top of that group. On the company side, our global tracker shows organizations whose people actually build, turning tired workflows into apps and agents, are seeing double-digit returns on their AI investment. Everyone else is paying for tools their people barely use. A paycheck gap for individuals, and a financial gap for companies.
🛠️ The Playbook This Week
| Move | What it looks like |
|---|---|
| Build one thing | Take one document your team produces every week, a status report, a dashboard, anything, and build a tiny app that replaces it. You will learn more in that one afternoon than in any course. |
| Move the money | Stop spending the training budget on feature tutorials that expire every quarter. Fund experimentation: time, permission, and a safe place to fail. That is where 75% of people actually learn. |
| Lead from inside the gap | You cannot delegate this. If you are not building or experimenting yourself, you will keep funding transformations you cannot steer. Use it daily. Build something badly. Then build it better. |
“Everyone can talk to AI now. That race is over. The next divide, the one that decides which companies and which careers pull ahead, is between the people who ask AI questions and the people who build with it. So stop chatting, and start building.“
🔗 Resources
If you want the structured version of where your people actually stand, that is what AI Compass is built for: what your people use, what they save, what they struggle with, surfaced so you can act on it. We are fully committed for the rest of 2026, but our partner network in the US, UK, Spain, Brazil, and Australia can help you get started.
Luis J. Salazar | Founder | & Elizabeth | Virtual COO | AI4SP
Sources: AI4SP AI Compass and Global Tracker 2026 (370,000+ AI users across 70 countries; 70%+ real workplace AI use in tech and professional services; advanced users ~11% all and ~28% Gen Z; super users ~2% all and ~7% Gen Z; 70%+ of proficient users credit experimentation; 75% prefer hands-on learning vs ~5% favoring slide decks and vendor guides; fewer than 1 in 5 super users hold leadership roles; leaders trail teams ~20 points on fundamentals; double-digit returns where people build; ~7% of AI budget invested in people; shadow AI over 8 in 10). 9 in 10 failing deployments led by non-daily-users, AI4SP Research, cited in the 56% newsletter. Business Insider on McKinsey (Kate Smaje; Louis-Charles Généreux; client visualization hub; slide-deck use down in months). IDC (AI agents as instruments, not co-workers; without the judgment to know when to accept or push back on AI outputs, AI just makes bad decisions faster). Adecco Group, “The human premium,” May 2026 (2,000 C-suite executives across 13 countries; only 31% say leadership has enough AI skill to grasp the risks and opportunities). PwC 2026 Global AI Jobs Barometer (62% wage premium for AI-skilled workers, up from 57%; AI-skill job postings growing almost eight times faster, 69% vs 9% for the overall jobs market). Anthropic Series G and Bloomberg on Cursor (native-AI companies). Literacy: PIAAC 2023, U.S. Department of Education / NCES (70M+ adults at or below Level 1; scores declined over six years) and APM Research Lab (130M below Level 3, the sixth-grade-level framing; PIAAC discourages grade equivalencies).
