Edition #29: The Gap Nobody Is Measuring

AI capability is on a hockey stick. Human capability is crawling. And the bill is about to come due.

By Mykel Salomon · 2026-07-12

<h3>AI capability is on a hockey stick. Human capability is crawling. And the bill is about to come due.</h3><hr><h3>🔓 Opening Reflection</h3>There's a gap opening up right now, and almost nobody is measuring it.It isn't between companies. It isn't between countries. It's between what AI can do and what humans are ready to understand, adopt, and trust.That's the argument my guest this week, James Garner, brought to the show, and once you see it, you can't unsee it.His framing starts with speed. Every technology before this one gave us time. The printing press took generations to reshape society. The internet took roughly seventeen years from invention to mass adoption. Time to argue, adjust, build institutions, retrain.Generative AI did it in months.So picture two lines on a graph. AI capability and adoption: a hockey stick. Human capability, our actual ability to use this well: a slow crawl. And the space between those two lines is widening every single week.Here's why that matters more than it sounds. When chatbots arrived, the gap was small. You typed in plain English, you got something back. There wasn't much to learn.But we've moved on to agents, orchestration, multi-step workflows , and a huge number of people never made it past the first rung. The ladder is getting taller while they're still standing at the bottom of it.James's fear, and mine: this hardens into a permanent divide between the haves and the have-nots.And his frustration, which I share completely: it doesn't have to be this way. The tools are free. The information is everywhere. Nobody is locked out by anything except attention and attitude.That's what makes the gap so maddening.It's not a technology problem.It's a human one.<hr><h3>🔦 Signal of the Week</h3><h3>The bill just arrived, and even Microsoft flinched</h3>Here's the story that should reframe how every leader thinks about AI budgets this year.In late 2025, <a target="_blank" rel="noopener noreferrer nofollow" class="text-[#4FD1C5] underline hover:opacity-80 tBFjffeGIQFLpveCPjltPbGBTBmBprvDA " href="https://www.linkedin.com/company/microsoft/">Microsoft</a> gave thousands of its engineers, designers, and PMs access to a premium AI coding tool. It worked. It spread fast. Then the token bills started landing.By June 2026, Microsoft had cancelled most of those licenses across the division that builds Windows, Microsoft 365, Outlook, and Teams — moving engineers to a cheaper in-house alternative.Read that again. The company that has invested roughly $13 billion in OpenAI, that writes up to 30% of its own code with AI, pulled the tool back because it cost too much.And they're not alone. <a target="_blank" rel="noopener noreferrer nofollow" class="text-[#4FD1C5] underline hover:opacity-80 tBFjffeGIQFLpveCPjltPbGBTBmBprvDA " href="https://www.linkedin.com/company/uber-com/">Uber</a> 's CTO said the company burned through its entire 2026 AI coding budget in four months. Individual engineers were running $500–$2,000 a month. Adoption hit 84%.Here's the cruel irony, and it's the whole point: the budgets blew up because the tools worked. Nobody misused anything. People just found it genuinely useful and used it constantly.Then there's the line from <a target="_blank" rel="noopener noreferrer nofollow" class="text-[#4FD1C5] underline hover:opacity-80 tBFjffeGIQFLpveCPjltPbGBTBmBprvDA " href="https://www.linkedin.com/company/nvidia/">NVIDIA</a> 's VP of Applied Deep Learning Research that should stop every executive cold: for his team, the cost of compute now far exceeds the cost of the employees.We have been living in what James calls a false sense of security. The frontier labs are running at a loss, competing for market dominance, and pricing accordingly. It's the <a target="_blank" rel="noopener noreferrer nofollow" class="text-[#4FD1C5] underline hover:opacity-80 tBFjffeGIQFLpveCPjltPbGBTBmBprvDA " href="https://www.linkedin.com/company/netflix/">Netflix</a> playbook: cheap until you're hooked.An this part is one we should all debate and think more about: when you're swimming in oil, you don't ration it. Right now AI feels effectively free, so we prompt carelessly, lazily, endlessly. When the tap tightens , and it will, we're going to discover just how wasteful we've been.The forecast that makes this concrete: <a target="_blank" rel="noopener noreferrer nofollow" class="text-[#4FD1C5] underline hover:opacity-80 tBFjffeGIQFLpveCPjltPbGBTBmBprvDA " href="https://www.linkedin.com/company/goldman-sachs/">Goldman Sachs</a> projects agentic AI could drive a 24-fold increase in token consumption by 2030. And <a target="_blank" rel="noopener noreferrer nofollow" class="text-[#4FD1C5] underline hover:opacity-80 tBFjffeGIQFLpveCPjltPbGBTBmBprvDA " href="https://www.linkedin.com/company/gartner/">Gartner</a> 's warning is the important nuance, even though the price per token is expected to fall roughly 90% by 2030, enterprise bills won't fall with it, because agents burn far more tokens per task. Cheaper units, vastly more of them.<h3>And underneath the bill: the physical world</h3>This is where the data centers come in — because tokens aren't abstract. They're electricity, water, land, and concrete.<ul><li>U.S. data centers consumed about 183 TWh in 2024 — over 4% of all U.S. electricity, roughly the annual power demand of Pakistan. Lawrence Berkeley National Lab projects that reaching 6.7% to 12% by 2028.</li><li>Globally, the IEA expects data center electricity use to roughly double, from 485 TWh in 2025 to about 950 TWh by 2030. AI-focused data centers grew 50% in 2025 alone.</li><li>The capital numbers are staggering: the largest tech companies spent over $400 billion on data center capex in 2025 — and the IEA notes that the capital expenditure of just five tech companies now exceeds global investment in oil and gas production.</li><li>It lands on real people. In some regions, communities near major data center clusters are already seeing electricity rate increases, and grid operators are warning about capacity.</li></ul>James also floated the fix, and I think he's right: a shift toward local and open models handling perhaps 80% of routine queries, with expensive frontier models reserved for genuinely hard problems. Efficiency is about to become a competitive advantage rather than an afterthought.Why this belongs in a human-centered newsletter: because the budget line is quietly changing from "cost of people" to "cost of people + cost of intelligence." Almost no leader has been taught how to price the second one. And when that number gets uncomfortable, it will start driving decisions about the first one.The capability gap isn't only about skills anymore.It's about to become a gap about who can afford intelligence at all.<hr><h3>🌍 In the Wild</h3><h3>1. The gap has no demographic pattern, and that breaks every assumption</h3>You'd expect a clean story: older workers resist, digital natives embrace. It isn't true. He's watched people who should be textbook technophobes build remarkable things with AI, and watched university students boo speakers at graduation ceremonies at the mere mention of it.Young people aren't rejecting AI because they don't understand it. They're rejecting it because they see a threat to their future.And James doesn't blame the students. He blames the institutions sending them impossibly mixed signals. His own son is at university in the UK, where AI has become something close to a dirty word, while everyone quietly uses it anyway.<blockquote>"It's like a sin. You don't speak about it. Everyone just knows."</blockquote>Students use it. Professors likely use it to mark the work. Nobody says so out loud.That silence is the real damage. You cannot teach people to use a tool responsibly while pretending the tool doesn't exist.The alternative he gave his own son: don't use AI to produce the assignment. Write it yourself, then feed it back into the AI and ask it to quiz you, challenge you, find your weak points. Use it to prove you understand,not to avoid understanding.Same tool. Opposite outcome. The only variable is intent.<h3>2. The MIT finding every leader should sit with</h3>James pointed to the MIT study on enterprise AI, and it's worth being precise about what it actually found.MIT's NANDA initiative reported that roughly 95% of enterprise generative AI pilots delivered no measurable P&L impact — despite $30–40 billion in enterprise spending.But the reason matters more than the number. The researchers were explicit: the failure isn't the models. It's what they call the learning gap — organizations unable to integrate, adapt, and retain context.Most companies are running AI as a technology rollout when it is fundamentally a change management problem. You're not asking people to learn software. You're asking people who have done a job the same way for twenty or thirty years to change how they work, quickly, and without a map.Humans hate that. Predictably. Reasonably.If your AI plan doesn't have a human change plan inside it, you're already in the 95%.<h3>3. Doomers, gloomers, bloomers, zoomers — and why your boardroom is stuck</h3>James leans on <a target="_blank" rel="noopener noreferrer nofollow" class="text-[#4FD1C5] underline hover:opacity-80 ember-view" href="https://www.linkedin.com/in/reidhoffman/">Reid Hoffman</a> 's four AI personas, and it's the most useful boardroom tool I've heard in a while:<ul><li>Doomers — AI is an existential threat to humanity</li><li>Gloomers — AI is coming for my job and my purpose</li><li>Bloomers — cautious optimism; want to embrace it, worried about doing it badly</li><li>Zoomers — full speed ahead (and, as James admits about people like us, "incredibly annoying")</li></ul>Here's the insight: all four are sitting at your table right now. And most leadership teams treat that as a problem to resolve by force, the loudest persona wins, or the conflict produces paralysis dressed up as "strategy."Every one of those positions comes from somewhere real. The gloomer's fear isn't noise; it's data. If you don't address it, you don't get alignment, you get people who feel unheard, quietly withholding effort from a rollout they never bought into.<blockquote>It shouldn't be an AI strategy. It should be a business strategy that is powered by AI.</blockquote>Don't confuse speed with haste. He's watching a lot of rushed, fear-driven decisions right now, in tech, in business, in people's personal lives. Sometimes the strongest move is to stop, take stock, and stay calm while everyone else reacts.<h3>4. "I hate the phrase 'human in the loop'"</h3>This one stopped me in the interview, and I've thought about it every day since.Everyone says "human in the loop" as if it's the answer to AI safety. James pushes back, because the phrase itself smuggles in an assumption. In the loop implies the loop belongs to the machine, and the human is a checkpoint inside someone else's process. It quietly makes us subservient.His alternative: bookending.Humans own the front, the intent, the brief, the definition of what good looks like. And humans own the back, the judgment, the accountability, the decision to ship or not.The machine works in the middle.Same safeguard. Completely different power structure. The human isn't a speed bump in the AI's workflow; the AI is a tool inside the human's.Words shape how organizations behave. This one is worth changing.<h3>5. Nobody taught us to manage agents</h3>Here's the skill gap almost no one is talking about.Agents need orchestration. Orchestration is management. And as James puts it with characteristic bluntness: if you're a lousy manager of humans, you're going to be a lousy manager of agents.Worse, actually. A human employee interprets. They hear a half-formed brief and think I know what she means. An agent does exactly what you said, not what you meant. James is candid that when he gets a bad result from an agent, it always traces back to the same root cause: he briefed it poorly. Not enough context. Not enough clarity about the goal.We are collectively terrible at this. We always have been. AI just made it expensive.And now look at what's happening to the pipeline where people used to learn to manage. The Big Four have cut UK graduate intake sharply, KPMG's graduate scheme dropped from about 1,399 to 942, with graduate job listings in accounting down 44% year-on-year. Law firms are compressing junior associate classes as AI takes over first-pass research and drafting.<a target="_blank" rel="noopener noreferrer nofollow" class="text-[#4FD1C5] underline hover:opacity-80 tBFjffeGIQFLpveCPjltPbGBTBmBprvDA " href="https://www.linkedin.com/company/axios-media/">Axios</a> framed the problem in a way I can't shake: junior work always did two jobs. It billed hours, and it trained people. Automate the billing part, and you've quietly destroyed the training part.We're removing the bottom rung of the ladder while telling people to climb.So the skill to build now is not prompting. Prompting is a feature. The durable skill is the one we skipped: knowing how to define a goal, give context, set constraints, and judge the output. Management, of humans and machines alike.<hr><h3>💬 The Big Question</h3>The real risk isn't what AI can do. It's what we're not ready for.So, honestly:Are you closing the gap , or quietly widening it?And for the leaders reading this: is your organization building capability, or just buying licenses?Because one of those closes the distance. The other just moves the finish line further away.<hr><h3>🧠 A Small Exercise</h3>Ten minutes. Five questions. Answer them honestly.<ol><li>Which persona am I really? Doomer, gloomer, bloomer, or zoomer, and when did I last genuinely listen to someone in a different camp?</li><li>Where am I on the ladder? Am I still on the chatbot rung, or have I actually tried building an agent or a multi-step workflow?</li><li>Am I bookending — or being managed by the machine? Do I own the intent and the accountability, or am I just clicking approve?</li><li>Do I know what my AI costs? Not the subscription. The real usage. If it tripled next year, would my plan survive?</li><li>Am I a good manager? Because that's the skill that's about to matter most, and agents will expose the answer faster than any employee ever did.</li></ol><hr><h3>🎙️ This Week's Episode</h3>The Capability Gap: What AI Can Do vs. What Humans Are Ready For Guest: <a target="_blank" rel="noopener noreferrer nofollow" class="text-[#4FD1C5] underline hover:opacity-80 ember-view" href="https://www.linkedin.com/in/jamesgarner1976/">James Garner</a> — Head of AI & Data, working at the frontier of agent orchestration and real-world AI systemsJames spent two decades as a quantity surveyor delivering major construction projects across Europe before moving into AI, and that unusual path is exactly why he's worth listening to. He knows the domain, the blockers, and the frustrations that tech-first people miss entirely.We get into:<ul><li>The capability gap — and why it has no demographic pattern</li><li>Why students are booing AI, and why he blames the institutions</li><li>Why "AI strategy" is the wrong phrase entirely</li><li>Doomers, gloomers, bloomers, zoomers — and the boardroom paralysis they cause</li><li>Why he hates the phrase "human in the loop," and what he says instead</li><li>The move from chatbots to agents to physical AI — he met his first humanoid robot in Amsterdam</li><li>The coming cost reckoning nobody has budgeted for</li><li>Why management, not prompting, is the skill that will separate people</li></ul>🎧 Available on <a target="_self" rel="noopener noreferrer nofollow" class="text-[#4FD1C5] underline hover:opacity-80 tBFjffeGIQFLpveCPjltPbGBTBmBprvDA " href="https://open.spotify.com/episode/2nha3JLnUxf69xINjqGBFM?si=naemg8GsR-WKekuc8CGyow">Spotify</a>, <a target="_self" rel="noopener noreferrer nofollow" class="text-[#4FD1C5] underline hover:opacity-80 tBFjffeGIQFLpveCPjltPbGBTBmBprvDA " href="https://podcasts.apple.com/us/podcast/s2e27-ai-agents-robots-the-human-gap-with-james-garner/id1843402023?i=1000775963550">Apple Podcasts</a>, and <a target="_self" rel="noopener noreferrer nofollow" class="text-[#4FD1C5] underline hover:opacity-80 tBFjffeGIQFLpveCPjltPbGBTBmBprvDA " href="https://youtu.be/QOCMTAW7j70?si=Ngww5j7cDxPQlzNI">YouTube</a>.<hr><h3>🧭 Thought to End the Week</h3>The technology is going to keep accelerating. That's not in question. James believes agents become commonplace within a year, and that most of us will meet a humanoid robot in an ordinary setting far sooner than we expect. He met his in April.What is in question is whether we close the distance, or let it harden into a divide.And here's what gives me hope, oddly. The thing separating the people thriving in this moment from the people drowning in it isn't intelligence, age, budget, or job title.Attitude.You can look at this technology as a threat to everything you've built. Or you can look at it as the most patient teacher you'll ever have, available at 3am, infinitely willing to explain, never once making you feel stupid for asking.Same tool. Same moment. Same gap.The difference is which side of it you decide to stand on, and whether you're willing to help someone else across.The code may change. And it will, faster than any of us are comfortable with.But our humanity is the constant.See you next week.— Mykel<hr><h3>📚 Sources</h3><ol><li>Fortune, "Microsoft reports are exposing AI's real cost problem" (May 2026) — Microsoft cancelled most internal Claude Code licenses across Experiences & Devices, moving engineers to GitHub Copilot CLI; Uber burned its full 2026 AI coding budget in four months; Nvidia's Bryan Catanzaro on compute exceeding employee cost; Goldman Sachs 24x token-consumption forecast; Gartner on falling token prices not yielding cheaper enterprise AI. <a target="_blank" rel="noopener noreferrer nofollow" class="text-[#4FD1C5] underline hover:opacity-80" href="https://fortune.com/2026/05/22/microsoft-ai-cost-problem-tokens-agents/">https://fortune.com/2026/05/22/microsoft-ai-cost-problem-tokens-agents/</a></li><li>MIT NANDA, The GenAI Divide: State of AI in Business 2025 — ~95% of enterprise GenAI pilots delivered no measurable P&L impact; root cause identified as the organizational "learning gap," not model quality. <a target="_blank" rel="noopener noreferrer nofollow" class="text-[#4FD1C5] underline hover:opacity-80" href="https://fortune.com/2025/08/18/mit-report-95-percent-generative-ai-pilots-at-companies-failing-cfo/">https://fortune.com/2025/08/18/mit-report-95-percent-generative-ai-pilots-at-companies-failing-cfo/</a></li><li>Pew Research Center / IEA — U.S. data centers used 183 TWh in 2024 (>4% of U.S. electricity, ≈ Pakistan's annual demand); projected +133% by 2030; 2023 U.S. data centers directly consumed ~17 billion gallons of water. <a target="_blank" rel="noopener noreferrer nofollow" class="text-[#4FD1C5] underline hover:opacity-80" href="https://www.pewresearch.org/short-reads/2025/10/24/what-we-know-about-energy-use-at-us-data-centers-amid-the-ai-boom/">https://www.pewresearch.org/short-reads/2025/10/24/what-we-know-about-energy-use-at-us-data-centers-amid-the-ai-boom/</a></li><li>IEA, Key Questions on Energy and AI (2026) — global data center electricity roughly doubling from 485 TWh (2025) to ~950 TWh (2030); AI-focused data centers grew 50% in 2025; top tech capex exceeded $400B in 2025, surpassing global oil & gas production investment. <a target="_blank" rel="noopener noreferrer nofollow" class="text-[#4FD1C5] underline hover:opacity-80" href="https://www.iea.org/reports/key-questions-on-energy-and-ai/executive-summary">https://www.iea.org/reports/key-questions-on-energy-and-ai/executive-summary</a></li><li>Lawrence Berkeley National Laboratory (2024), via Brookings — U.S. data centers at 4.4% of national electricity in 2023, projected to 6.7%–12.0% by 2028. <a target="_blank" rel="noopener noreferrer nofollow" class="text-[#4FD1C5] underline hover:opacity-80" href="https://www.brookings.edu/articles/global-energy-demands-within-the-ai-regulatory-landscape/">https://www.brookings.edu/articles/global-energy-demands-within-the-ai-regulatory-landscape/</a></li><li>City A.M. — Big Four UK graduate intake cut by up to 29%; KPMG from 1,399 to 942; accountancy graduate listings down 44% YoY. <a target="_blank" rel="noopener noreferrer nofollow" class="text-[#4FD1C5] underline hover:opacity-80" href="https://www.cityam.com/big-four-slash-graduate-jobs-as-ai-takes-on-entry-level-work/">https://www.cityam.com/big-four-slash-graduate-jobs-as-ai-takes-on-entry-level-work/</a></li><li>Axios, "AI threatens Big Law's talent pipeline" (May 2026) — junior work served both billing and training; automating it removes the apprenticeship path. <a target="_blank" rel="noopener noreferrer nofollow" class="text-[#4FD1C5] underline hover:opacity-80" href="https://www.axios.com/2026/05/02/ai-lawyers-law-firms-artificial-intelligence">https://www.axios.com/2026/05/02/ai-lawyers-law-firms-artificial-intelligence</a></li><li>Reid Hoffman, Superagency — the doomer / gloomer / bloomer / zoomer framework.</li></ol>