How AI fatigue keeps businesses from the real gains

Before Turbo, I spent almost seven years at a fuel software startup. I joined when there were four of us and worked my way from engineer to designer to head of product, helping build software that made sourcing and dispatch autonomous for a large portion of the fuel market. We called it AI, like everyone did back then. Under the hood it was linear programming and machine learning, wild math that looked like magic from the outside, and it worked.

Then LLMs showed up, back when almost nobody knew who Claude was. As head of product, I used agents in a build process that pulled our roadmap, our velocity tracking and our projections together and assessed every commitment we'd made against what the team could actually deliver. Before LLMs, that kind of assessment was out of reach. The data was scattered across too many systems, and stitching it together would have been an engineering project of its own.

For the first time I could actually see the answer to the thing I kept circling, which was why everything felt so impossible. Our commitments to customers ran far past what our velocity could ever support, and not by a little.

I shared my ugly findings with the rest of the leadership team. The transformational move was sitting right there in front of us: rework what we'd committed to our customers and rebuild those relationships around what we could really deliver.

Instead, the evidence got stuffed down.

The conversation turned to how we could all be more productive, with a product team that was already stellar, and the energy went into work that didn't even have a clear OKR. The truth was right there, and we buried it.

And that company kept winning. It still does. Businesses get to where they are on what worked, and everything after that unfolds on its own, win or lose. The people inside carry whatever doesn't work, the friction, the broken systems, the overcommitment, and they carry it quietly because that's the job.

But that report showed us something we had never been able to see before, and it pointed at a much better path. The question I walked away with was about a kind of neuroplasticity, the business kind. When AI finally shows a company the truth about itself, how many of them have it in them to change?

When I left in the fall of 2024, I spent three months on a sabbatical, finding out what building with AI actually looked like. The plan was to make Turbo a clone of me, something that could take a product from a spec all the way to shipped, with great design the whole way. So I went deep. I fine-tuned a couple of models and built agents back when agents barely held together, and every single step was friction.

Somewhere in those three months, it flipped my world on its axis. I came out convinced that anything we do as people through a computer can now be done without us, and that there was enormous value hiding in that, even if almost nobody around me could see it yet.

Since then I've chased oversized projects on purpose, because every one of them sharpened the system. We shipped about 170 features of an enterprise freight invoicing system in six weeks. Agents migrated an edtech client's 432-feature codebase from Blazor to React, reviewing and testing every feature, and I never once opened an IDE. Each project left behind a piece of the system we run on today.

And the more leverage I build, the further from earth I feel. I'm ahead of the problems the market is still complaining about, wanting to scream that they're solved, and I keep showing up to conversations where what I do doesn't fit in anyone's box. I think about Zima Blue, the robot built to clean a pool who kept getting upgraded until he was painting across galaxies. I've never had more capacity. I've never been harder to place.

And still, when I walk into a room to help an organization, I get a dissociative stare. I feel like a cheat code trying to explain the cheat code to the NPC.

The biggest barrier I run into is belief. People don't believe anything has really changed, or they can't fathom what it would mean for their business, or they don't believe a competitor could rise up and take what they've built while they wait. I get it. If you've spent fifteen years building something that works, a stranger telling you the ground just moved sounds a lot like a pitch.

The dissonance shows up in almost every conversation I have with a business. A CTO told me his moats were thinning, and in the same conversation that no competitor was about to pounce. A head of product, deeply underwater on his roadmap, couldn't fathom bringing in a horizontal builder, someone who designs and ships across the whole product, because that person didn't fit his vendor roadmap or his org chart.

And the question I hear more than any other is some version of this one: what's the difference between someone who's been building with AI every day for three years and me with Claude and a spreadsheet? It's a fair question. The answer is everything under the waterline, and you can't see that part from a chat window.

What keeps people asleep is that none of it feels urgent enough. Buying into AI means believing there's a real threat to your business, then feeling enough urgency to change, then taking a risk for a reward you can't see yet. Most of the businesses I talk to never get past the first of those.

And the day it's finally visible in your niche is usually the day a tool already replaced your service.

Inside most companies, AI is turning into a dirty little secret. Leadership teams are burnt out on chasing it as an organizational change, the strategy decks, the committees and the system overhauls that never quite land. Their people, meanwhile, are using it on their own for personal wins more than ever, and mostly not saying so. Bring it up in the Monday leadership meeting and the table goes quiet, and you can feel why:

• Every vendor email has AI in the subject line, and every one of them was built for your industry.
• The board wants to hear your AI strategy, and nobody in the room wants to be the one to answer.
• Last spring's pilot made more work than it saved, and nobody brings it up anymore.
• A competitor just posted a job for a head of AI, and you can't tell if that should scare you.
• Half your team is using ChatGPT on the side and hoping nobody asks.
• The feed says a ten-person startup replaced a department over a weekend, and the next day it says the whole thing is a bubble.
• You've sat through a dozen demos and can't name one thing in your business that changed.
• And underneath all of it is the worry that everyone else has figured out something you haven't.

That burnout goes all the way to the top. In a 2026 survey by Writer, a company that sells AI to enterprises, 54% of C-suite executives said adopting AI is tearing their company apart, and 48% of them called it a massive disappointment, up from 34% the year before.1Enterprise AI adoption in 2026: Why 79% face challenges despite high investmentWriter, with Workplace Intelligence, 7 April 202654% of C-suite executives say AI adoption is tearing their company apart; 48% call it a massive disappointment, up from 34% a year earlier

Every business is hearing two things at once right now. AI changes everything. Nobody is getting real value from it. Both are loud. Both are cited. Together they make you feel like you have to do something, immediately, without ever being told what.

Building real leverage with AI takes a lot of ingredients, and most of this journal digs into them: the context, the systems, the trust, the whole path from a chat window to a business that shapes itself. Very few companies have all of them yet, so the market has turned into a smorgasbord of where everyone is with it.

You might be a ten-person firm whose owner tinkers with ChatGPT late at night. None of it has made it into the business yet.

You might run three hundred people, with an AI committee and a dozen pilots. Nothing is in production.

You might have bought the enterprise seats and watched usage spike for a month. Then it flatlined.

You might have one person who has figured it out on their own. They do a week's work in an afternoon, and nobody else can follow what they're doing.

Or you might be that person. You get it, and you're sitting in a room that doesn't.

Everyone wants it, and most can't touch it yet. A few are getting close, and they're usually the ones who've noticed it's the same game on new physics.

And the people who actually get results tend to disappear. Some get hired away, often for a lot of money: postings for forward deployed engineers, the people who make AI products work inside a customer's business, rose more than 1,000% in the first eight months of 2026 against the same months a year earlier.2One of the fastest-growing jobs in Silicon Valley sends engineers straight to customers' offices to get AI up and runningFortune, citing Lightcast, 3 September 2026forward deployed engineer postings up more than 1,000% from January to August 2026 against the same months a year earlier The rest stay and run straight into the wall this whole piece is about, a business that can't yet believe anything has changed. It's a black hole for exactly the people a business needs most.

The people selling answers, me included, sound more certain than anyone has a right to be. There's noise and a little proof. There are no experts in the room, and there's no blueprint.

We're lost. All of us.

And when everyone's lost, the whole market starts reacting in every direction at once.

And yet people keep using it. Half of American adults have now tried an AI chatbot, up from a third two years earlier, and Pew still finds more of them expecting harm than good.3Americans and AI 2026: Chatbots, Smart Devices and Views on ImpactPew Research Center, 17 June 2026chatbot use at 49% of US adults, up from 33% in 2024, alongside more expecting negative effects than positive Gallup and Bentley University see the same slide from another angle: the share of Americans who say AI does more harm than good went from 31% to 39% in a single year.4Americans Cool Toward AIBentley University and Gallup, 28 July 2026trust in businesses to use AI responsibly at 27%, down from 31%; AI does more harm than good at 39%, up from 31%

• 27% of Americans trust businesses to use AI responsibly, down from 31% a year earlier Bentley-Gallup, July 2026
• ~40% of the time AI saves is lost to rework Workday, January 2026
• 40% of desk workers received AI workslop in the past month Stanford and BetterUp, 2025

If AI has made your week longer instead of shorter, the surveys back you up, and that pilot your ops lead still won't talk about was completely normal. In a 2024 survey by Upwork, the freelance marketplace, 77% of employees using AI said it had added to their workload.5Employee workloads rising despite increased C-suite investment in AIUpwork Research Institute, July 202477% of employees using AI said it added to their workload Workday, which sells HR software with AI built in, found that only 14% of employees consistently come out ahead.6Beyond Productivity: Measuring the Real Value of AIWorkday and Hanover Research, 14 January 2026nearly 40% of AI time savings lost to rework; 14% consistently get clear positive net outcomes Researchers at Stanford and BetterUp even gave the leftovers a name, workslop: output that looks good, lacks substance, and takes about two hours to clean up every time it lands on someone's desk.7Workslop: The Hidden Cost of AI-Generated BusyworkBetterUp Labs and Stanford Social Media Lab, September 202540% of 1,150 US desk workers received workslop in the past month, about two hours to resolve each

That Monday meeting gets squeezed from both directions at once. Half the employees in a 2026 Workday survey admitted they'd hidden their AI use at least once in the past month,8AI Adoption Is Moving Faster Than Employee TrustWorkday, September 2026half of 5,944 employees hid their AI use at least once in the past month while above them, 71% of the CEOs in KPMG's 2026 outlook expect returns from AI within one to three years.92026 Global CEO OutlookKPMG, October 202671% of 1,400 CEOs expect AI returns within one to three years The people doing the work are hiding it, and the people paying for it have started watching the clock.

Then there's the parlor trick. A demo that dazzled on Friday still needs somebody to check every single line it wrote. Merriam-Webster made slop its word of 2025.102025 Word of the Year: slopMerriam-Webster, 15 December 2025slop as the word of 2025 Deloitte had to refund part of a government contract after a report showed up with invented citations,11FOI disclosure on the Deloitte reportAustralian Department of Finance, 2025a partial refund on a A$440,000 contract after fabricated references and a public database of court decisions dealing with hallucinated material now runs past 2,200 cases.12AI Hallucination Cases DatabaseDamien Charlotin, checked 8 November 20262,266 decisions documenting hallucinated material

But the cost I worry about most comes later.

When a real gain finally does show up, the people who have to adopt it may have stopped believing. They've been through the hype, the hunt for the right tool, the letdown and the retreat so many times that the next promise sounds exactly like the last one. Our piece on why AI projects fail follows that loop inside a company, and it wears people down from the inside.

By the time AI news reaches you, it has all blurred together. A viral post sits right next to a vendor's claim and a careful study, and they all sound equally sure of themselves. They aren't.

The startup that supposedly replaced a department over a weekend reached you through that top band, the one with the least behind it.

I first ran this table a year ago, setting the loudest claims next to the record. A year later, the record still hasn't caught up.

The money keeps changing shape too, and that only makes the noise louder. Nvidia's plan to put up to $100 billion into OpenAI gave way to a $30 billion investment in OpenAI's February 2026 round,19OpenAI raises $110B in one of the largest private funding rounds in historyTechCrunch, 27 February 2026OpenAI's $110B round took $30 billion each from Nvidia and SoftBank, after reports of a $100 billion Nvidia investment gave way to reports of a smaller one and Nvidia's chief executive said the original "was never a commitment."20Pledge to invest $100 billion in OpenAI was "never a commitment," says Nvidia's HuangFortune, 2 February 2026Nvidia's letter of intent to invest as much as $100 billion in OpenAI, and Jensen Huang's remark that it "was never a commitment" Then AI stocks sold off hard in the first week of November 2026.21Nasdaq heads for correction as AI stocks plunge after midtermsAssociated Press, via the Honolulu Star-Advertiser, 5 November 2026the November 2026 sell-off in AI stocks Whichever way that breaks, I don't think a macro bet you can't control should decide whether you fix the workflow sitting right in front of you.

The most useful way I've found to sort all this talk is by time horizon. Some people are arguing about a superintelligence somewhere out in the future, and whether it ends in extinction or utopia. Others are arguing about things you can measure this quarter, like jobs, bias and returns. Sorted that way, I count at least thirteen camps. And an optimist and a pessimist arguing about next quarter usually have more in common with each other than either has with someone arguing about the far future.

People move between these camps more than you'd think. LeCun doubts that scaling gets us to human-level AI and is a long-run optimist anyway. Musk signed the pause letter and started a frontier lab within months. The labs' own chief executives signed the extinction statement while racing to build the thing it warned about.

The two loudest ends of this are closer than they look. Eliezer Yudkowsky has called for bombing rogue data centers. Marc Andreessen calls slowing down a form of murder. They're both convinced AI is the most world-altering force in history, and they only disagree on which way it's going to break.

The middle of the horseshoe is the only part keeping score: the academics running experiments and the operators shipping real work. What they find almost always comes out as "it depends," which is exactly why nobody's selling it. They're who I read first.

Most of these positions come down to what's driving the person holding them. Once you can see that, you can usually guess where they'll land before they open their mouth.

1. Fund returns, book deals, consulting work and lab fundraising all need the story to keep moving. Once someone has made a forecast in public, walking it back costs them, so the position slowly hardens into who they are.

   Incentive

2. Doom and hype both get the headlines and the funding.

   Attention

3. AI goes after thinking work, which is the commentators' own work. People draw a line in the sand to protect the years they spent learning a craft, and they'll defend that line harder than a paycheck.

   Identity

4. Utopia and apocalypse are both old religious stories. The prophecy just has a date on it now, and the ending is a superintelligence or extinction. People are certain about it long before there's any evidence.

   Faith

So when you hear a loud claim, ask yourself whether you can check it on a date that's coming soon. If you can't, the claim mostly tells you which camp the speaker belongs to. The forecasts below all came with dates, and most of them were made by people with every reason to be early.

In April 2024, Musk said AI would be smarter than the smartest human within two years. That date came and went. At Davos in January 2026 he moved it, the usual way, to the end of 2026 or 2027.22Elon Musk on AI, leadership and speedNorges Bank Investment Management, In Good Company, 8 April 2024"smarter than the smartest human ... within two years"23AI smarter than any human is coming by next year, says Elon MuskBusiness Today, from the World Economic Forum, 22 January 2026"smarter than any human by the end of this year, and I would say no later than next year" Amodei said AI would be writing 90% of code within three to six months, and that window closed with no audit anywhere near it. Developers themselves put the share of their code AI writes or assists at around 42%.24CEO Speaker Series with Dario Amodei of AnthropicCouncil on Foreign Relations, 10 March 202590% of code written by AI in three to six months25State of Code Developer Survey 2026Sonar, 2026developers' own estimate of the share of code AI generates or assists at 42%

The rest are still open, and they've started to soften. OpenAI now talks about "very small discoveries" in 2026, and Anthropic calls powerful AI by early 2027 plausible.18AI progress and recommendationsOpenAI, April 2026"very small discoveries" expected in 2026, more significant ones from 202826Frontier Safety RoadmapAnthropic, July 2026powerful capability by early 2027 described as plausible; the March 2025 OSTP submission had said late 2026 or early 2027 The white-collar bloodbath has until 2030 on its window, and its clearest signal so far shows up in hiring: employment of 22 to 25 year olds in the most exposed jobs runs about 19% below trend.27AI jobs danger: Sleepwalking into a white-collar bloodbathAxios, 28 May 2025half of entry-level white-collar jobs and 10 to 20% unemployment within one to five years15No Widespread Displacement, but the AI Employment Gap for Young Workers Has Widened to 19%Brynjolfsson, Chandar and Chen, Stanford Digital Economy Lab, 12 August 2026employment of 22 to 25 year olds in highly exposed occupations about 19% below trend, mainly through reduced hiring

Herbert Simon said in 1960 that machines would do any work a man can do within twenty years. In 2016 Hinton told the world to stop training radiologists, and the number of US radiologists has kept growing ever since.28The Radiologist Shortage: A Workforce UpdateACR Bulletin, 2026the number of US radiologists grew 17.3% from 2014 to 2023 IBM sold off Watson Health. Zillow's home-buying model ended in $569 million of write-downs. And the driverless coast-to-coast trip Musk promised for 2017 finally showed up as a robotaxi service in seven cities in 2026.29The museum of confident predictions30RobotaxiTesla, November 2026rides in Austin, Dallas, Fort Worth, Houston, Miami, Orlando and Tampa

Most of these got the direction right and treated the date as marketing, which makes sense when so many of the people making them had something to sell. A few were worded so carefully they could never cleanly lose.

Medicine runs on trials, and a trial reports the average effect across a big sample. Doctors have been writing about what that hides for decades. One well-known paper says an average can "fail to reveal the potentially complex mixture of substantial benefits for some, little benefit for many, and harm for a few."31Evidence-Based Medicine, Heterogeneity of Treatment Effects, and the Trouble with AveragesKravitz, Duan and Braslow, The Milbank Quarterly, 2004an average can hide large benefit for some, little for many and harm for a few

AI studies split the same way. In one study of customer support agents, the gain from an AI assistant shrank as skill went up, until for the most skilled agents it was too small to measure.16Generative AI at WorkBrynjolfsson, Li and Raymond, The Quarterly Journal of Economics, February 202515% more issues resolved per hour on average, 36% for the lowest-skill quintile, no significant change for the most skilled

Consultants at BCG using GPT-4 finished 12% more tasks, 25% faster and at higher quality when the work sat inside the model's abilities, then were 19% less likely to reach the right answer on a task just outside them.32Navigating the Jagged Technological FrontierDell'Acqua et al., Organization Science, 11 March 202612.2% more tasks, 25.1% faster and higher quality inside the frontier; 19% less likely to be correct outside it Across firms, about a quarter of US businesses use AI in some function, against 50 to 60% of very large firms in information, professional services and finance.33Business Trends and Outlook SurveyUS Census Bureau, October 202624.6% of businesses used AI in a business function in the two weeks to 4 October 202634The Microstructure of AI DiffusionBonney et al., NBER Working Paper 35141, April 2026adoption around 50 to 60% among very large firms in information, professional services and finance Average the firms where it works with the firms where it doesn't, and the line can come out completely flat. A study of about 25,000 Danish workers in eleven exposed occupations found exactly that, with no effect on earnings or hours larger than 2%.35Still Waters, Rapid Currents: Early Labor Market Transformation under Generative AIHumlum and Vestergaard, March 2026rules out effects on earnings or hours larger than 2% two years after ChatGPT

And studies age fast. METR's 2025 trial found experienced developers were slower with early-2025 tools. Its follow-up on newer tools found them somewhat faster, then called even that unreliable, because developers increasingly refused to work without AI at all.36We are changing our developer productivity experiment designMETR, 24 February 2026returning developers about 18% faster and new ones about 4% faster, both intervals including no effect; METR calls the signal unreliable The BCG experiment ran on GPT-4 in 2023 and reached its journal in March 2026, by which point the tools it measured were two generations old.

Ask the market what AI is and you'll get an answer with a product attached. Depending on who's selling, it's a labor force, a tool, a threat to your thinking or a whole new way to build.

The labor-force shape needs layoffs as its proof, so cuts get labelled as AI. Challenger counted about 55,000 announced US job cuts attributed to AI in 2025, 5% of the year's total, while the New York Fed found that just 1% of AI-using service firms had laid anyone off because of it.41October layoffs at three-year lowChallenger, Gray & Christmas, 5 November 202654,836 AI-attributed announced cuts in 2025, 5% of the total42Are Businesses Scaling Back Hiring Due to AI?Federal Reserve Bank of New York, 4 September 20251% of AI-using service firms laid off workers because of AI Some of those cuts are real automation. Others are ordinary cost-cutting that sounds a lot better when you blame it on AI.43Evidence of an AI-driven shakeup of job markets is patchyOxford Economics, 7 January 2026traditional causes cited more often than AI

The threat to thinking has something real in it too. In one Polish study, endoscopists who'd worked with AI for three months went back to working without it, and found a precancerous growth in 22% of colonoscopies, down from 28% before they'd ever used it.44Endoscopist deskilling risk after exposure to artificial intelligence in colonoscopy: a multicentre, observational studyBudzyń et al., The Lancet Gastroenterology & Hepatology, 12 August 2025adenoma detection without AI fell from 28.4% to 22.4% The lesson I'd take is to watch which skills your team stops practicing, and that gets lost the moment the headline turns into "AI makes you dumb."

Google's DORA research, drawn from nearly 5,000 technologists, found that AI "magnifies the strengths of high-performing organizations and the dysfunctions of struggling ones."45State of AI-assisted Software DevelopmentDORA, Google Cloud, 2025AI "magnifies the strengths of high-performing organizations and the dysfunctions of struggling ones" In a business where it's working, I see some mix of these:

1. It makes more of what already works.

   Amplification

2. It takes the drag out of the work: the slow handoff, the guess where a number should be, the step only one person in the building knows how to do.

   Friction

3. It lets you do things you couldn't do before. An analysis that took a team a quarter comes back by morning, and a small firm can take on work that was always out of its reach, without hiring for it.

   Leverage

That's why the studies split. Novices gained because the friction dropped. Firms with a working loop pulled ahead while firms with broken processes lost ground, so averaged together they can land right near zero, like they did in Denmark. None of the shapes catch this, because each one looks at AI from a single seat, and what actually gets amplified is the whole loop.

We forget how early this is. ChatGPT is four years old. Electricity took about forty years to show up in factory productivity, and the gains only came once factories ripped out the single engine and its line shaft and put a small motor at every machine.46The Dynamo and the ComputerPaul A. David, American Economic Review, 1990electrification's gains arrived once factories were redesigned around unit motors In 1987 Robert Solow said the same thing about computers: "You can see the computer age everywhere but in the productivity statistics."47We'd Better Watch OutRobert Solow, The New York Times Book Review, 12 July 1987"You can see the computer age everywhere but in the productivity statistics"

Computers paid off the same way, once people rebuilt the work around them. With AI, that rebuild has barely started.

You make one person ten times faster. But the work feeding into them doesn't change, and neither does the work after them. So now they sit and wait, faster than before, and the company ends up with faster silos.

Leadership feels it and goes looking for a Head of AI to own the whole thing, and plenty have hired one: IBM's 2026 CEO study counts a Chief AI Officer at 76% of the organizations it surveyed, up from 26% a year earlier.48IBM Study: CEOs are reshaping C-suite roles for the AI eraIBM, 4 May 202676% of surveyed organizations have a Chief AI Officer, up from 26% The hire usually turns out to be one of these:

• Someone who knows AI cold and has never touched a real organizational problem.
• Someone business-savvy who, in practice, hands the team better prompts.
• Someone who believes the tools are the answer and runs initiatives around them.
• Someone brilliant who disappears into a corner to tinker, when the team needed a hand on its shoulder.

Every one of those is a real, fundable hire. None of them can own the work, because the work lives in the handoffs between teams and every messy gap in between, and you can't own a problem that lives in the gaps. The role needs operations, business judgment and engineering all at once, plus the know-how below the surface that makes AI compound. Very few people have all of that.

More than anything, owners are yearning for the truth. What will AI actually do for my business, and what will it cost me to get there? Most of them are still waiting for an answer, and in the meantime they get a demo, a deck, a forecast with a date on it, or a study about somebody else's average.

Every good thing my team has found came from trying, again and again, with no blueprint. A business making payroll can't just go try, so it does the safe version:

• A chat window open in another tab, when an agent could be running a whole workflow overnight.
• The AI built for its industry, when the business is one of one.
• A pilot on the vendor's clean sample data, instead of last month's real exports, the ones with three naming conventions and the CSV nobody owns.

Those are sensible moves, and they keep AI above the waterline, where it can never compound.

I believe consolidation is coming, and in some industries it's already here. By August 2026, Current, owned by Thrive Holdings, had bought 48 local accounting firms to rebuild their work on AI. In May 2026, Long Lake, backed by General Catalyst, agreed to buy the travel company Amex GBT for $6.3 billion on the same bet.49Buy the Firm, Then Rebuild It With AIPYMNTS, 4 August 2026Current, owned by Thrive Holdings, has acquired 48 local accounting firms; Long Lake, backed by General Catalyst, agreed in May to take American Express Global Business Travel private for $6.3 billion In your niche it'll look like an acquisition down the street, or a competitor quoting in a day what takes you a week.

And compounding leverage is real. BCG's 2026 survey of 1,330 executives found the 7.5% of companies it calls future-built growing revenue 2.4 times as fast as the laggards.50AI Is Starting to Pay Off. Almost 50% of Companies Now Generate Value with It.BCG, 30 September 20267.5% of companies classed as future-built, with 2.4 times the revenue growth of laggards, in a survey of 1,330 executives BCG sells the transformations its research recommends, so take the multiple with a grain of salt.

But the direction matches everything I see. The first workflow is slow and ugly. The second one borrows from it. By about the fifth the snowball is rolling, and the business is doing work it couldn't have quoted a year earlier.

You'll want a guide for this, because most of AI sits under the waterline. Everyone sees the chat window. What makes it compound is underneath, in the context, the systems and the agents you've slowly learned to trust.

Getting one workflow to run on your real data, the way your best person would do it, takes years of battle scars, and you can't earn those while you're making payroll. So find someone who already has them, someone who'll sit with your team on its real bottlenecks and build working proofs until your people don't need them anymore.

Then start applied. Map your business on the Transformation Wheel, find the weak box sitting next to the most revenue or cost, and move one number there. Measure it before you change anything, and again after, with the review and the rework counted in.

When an average can't say what a treatment will do for one patient, doctors can run a trial of one, measuring the same person with and without it over repeated periods.51N-of-1 and Novel Within-Subject Trial MethodsNIH Office of Disease Preventiona planned, repeated comparison for one person Your one number is that trial, run on your own work, and it's as close to the truth as anyone is going to get you.

So next Monday, when someone brings up AI and the table goes quiet, bring a number. If you could move one number in your business in the next ninety days, which would it be?