John Rector has been writing about artificial intelligence for over a decade, since the 2015 paper “Mommy, What’s a Cashier?” He is an entrepreneur, author, and AI strategist; co-founder of E2open; and founder of Charleston AI, a lab in Charleston, SC that helps people and organizations understand and use artificial intelligence.
His AI teaching runs on the Reality Equation: a mind is a prediction machine, AI is a second one, and the stack now meets the world first — absorbing surprise on your behalf and collecting the update that comes with it. The recurring themes: attention engineering, absorption, personal AI, and the future of human work. For the books, see Author.
Start here
- Attention Engineering: How to Steer AI Like a Subconscious System — the working method.
- The Rise of AI Conversations — are people preferring AI over other humans?
- Learning Is Metabolized Surprise — the brain learns from prediction error; the stack’s mandate is to remove it.
- The Pre-Translated World — attention was always a wage: whoever pays it collects the update.
- The Stack Must Sleep — what a personal AI still lacks: consolidation.
- Proof of Person — when every sentence may be stack-drafted, trust goes looking for collateral that cannot be delegated.
- AI Changes the Entity — the recursive human–AI system, in Reality Equation terms.
- Photographers Finally Realize AI Has Won — the real argument, in one craft.
Every AI essay
- The Great Customer
In a Pay After economy sellers go first, and they choose whom to go first for. The pre-approved list becomes the most valuable place a person can stand: status stops saying you can pay and starts saying you will. … readThe Great Customer - A Pay After World
Pay Before prices a hope; Pay After prices a fact. Twelve far-out predictions about what happens to advertising, the billable hour, GDP, honesty and trust when the whole economy learns to pay after, the way workers always have. … readA Pay After World - Pay After Is Not Pay Later
Pay Later is a loan: a lender pays the seller up front and you owe it either way. Pay After has no recourse: the seller is paid only if the product earns it, and that one difference starves every fake in the economy. … readPay After Is Not Pay Later - Pay After
By 2030 a customer a seller can predict will order something, use it for a while, and only then buy it. Not credit and not a trial that bills by default. Pay After will spread the way free shipping did. … readPay After - What a Store Is For
By 2030 the website becomes a catalog your AI reads, and the store becomes the one place a human goes to find out what a screen can no longer prove. What it holds, how it looks, and where it sits on the map. … readWhat a Store Is For - Presence Becomes Premium
Every time talking at a distance got cheaper, people traveled more. When AI makes the screen perfect and untrustworthy, the value of being physically together goes up, and it reprices offices, cafes, stores, hotels and downtowns. … readPresence Becomes Premium - I Got Pixeled
The day the screen stops counting as evidence, trust does not disappear. It moves to whatever is still expensive to fake, and that is mostly a person in a room. Six predictions for the fallout. … readI Got Pixeled - The Agent Without Appetite
Every institution you have ever stood inside is scaffolding around one old problem: anyone who acts for you has interests that are not yours. AI delivers the first agent that wants nothing — and the appetite does not vanish. It moves upstream, to the maker, where it can finally be put in writing. … readThe Agent Without Appetite - Meaning Becomes an Address
For all of history, finding something meant knowing what someone else called it. Now the address is computed from the thing itself, and things that mean alike land near each other. … readMeaning Becomes an Address - Every Trap Is Near the Floor
To be stuck, every direction around you has to point up. In a million dimensions that almost never happens except near the bottom, which is why AI training works at all. … readEvery Trap Is Near the Floor - Two Buttons Are Enough
AI is the first tool with more moves than its operator, and a 1956 law explains why that is not the end of control: you were never required to match the machine’s moves, only to know the one number you defend. Control migrates from the moves to the criterion — and two buttons are enough to hold it. … readTwo Buttons Are Enough - The Bank That Knocks First
When a bank buys a signature house, the branch stops waiting for loan applications. It learns which promises work, brings the whole deal to the businesses that fit, and takes a share of the gain instead of interest, as long as it keeps the risk it sells. … readThe Bank That Knocks First - The Best Lender Is the One Who Does the Work
Banks ration credit to projects they cannot see inside. A signature house that does the work and guarantees the outcome can finance them from the result itself, as long as it lends only against its own signature. … readThe Best Lender Is the One Who Does the Work - The Signature House: The Next Great Firm Sells a Promise
Part insurer, part law firm, part engineering shop, part bank: when manufactured intelligence makes the work cheap, the gaps between those four firms become the expensive part, and they fold into one that sells guaranteed outcomes. … readThe Signature House: The Next Great Firm Sells a Promise - Talent Will Gather Where Risk Is Priced and Rules Are Written
Silicon Valley was where the people who built tools found each other. When tools come off the intelligence assembly line, the talent moves to where risk is priced and rules are written: the financial capitals and the capitals of government, not the labs. … readTalent Will Gather Where Risk Is Priced and Rules Are Written - The Signature Is the Product: Insurers Will Write the Rules of AI
When manufactured intelligence makes the work cheap, the value moves to whoever stands behind it, and the people who price that accountability are insurers. Seven predictions for how they become the real regulators of AI. … readThe Signature Is the Product: Insurers Will Write the Rules of AI - The Last Line Is You
Every line you have ever stood in was a machine for rationing a human being. When the server can copy itself, those lines disappear, and the only one left runs through your own attention. … readThe Last Line Is You - When Intelligence Comes Off an Assembly Line: Eight Predictions
Nina Schick says we are turning sand into non-biological intelligence. Here are eight predictions for what happens when the scarcest thing a civilization ever had becomes a manufactured good. … readWhen Intelligence Comes Off an Assembly Line: Eight Predictions - You Do It. You Decide.
My two favorite AI prompts are four words long: “You do it” and “You decide.” Stop telling AI how, describe the state change you want, push back when it hands the work back, and log what it cannot do yet, because that list expires. … readYou Do It. You Decide. - The Record’s Version of You
Everyone who has ever known you keeps a model of you, compiled from your artifacts and locked in a subconscious neither of you can read. The machine’s memory panel is the first such model with a page you can open — and the flinch when you read it is an instrument pointed at the gap between what you mean and what you leave. … readThe Record’s Version of You - The Notification Everyone Wants Is Done.
Billions of people chat with AI while only a few million let it work in the background. The bridge between them is an AI that finishes the work unasked and holds the state change for Accept or Reject — and the notification everyone wants is Done. … readThe Notification Everyone Wants Is Done. - Surprise, Paid in Advance
You cannot touch your denominator, but you can write its curriculum. Practice is surprise paid in advance at a discount — and AI just collapsed the price of the drill room. … readSurprise, Paid in Advance - Look for the Past-Tense Verb
Your first customer is not paying for effort but for something to become true, and the past-tense verb is how you see it. As AI makes artifacts cheap, the value moves to verbs that need someone else’s yes: signed, shipped, paid, renewed. … readLook for the Past-Tense Verb - Correlated Surprise
When the same few engines write the daily record for billions of minds, private forecasters converge — not by touching any denominator, which is impossible, but by feeding them all the same diet. Surprise becomes correlated, and correlated surprise is what turns a loss into a crisis. … readCorrelated Surprise - Your First Customer
Graduating into the AI economy, your first move is not starting a company. It is finding Customer Number One, letting them finance your development, and bringing your own AI coworker who walks out the door with you. … readYour First Customer - The Automation Tragedy: How the Free Market Is Overgrazing Aggregate Demand
Each firm that swaps a paycheck for a server keeps the savings and spreads the lost spending across everyone. That is a tragedy of the commons, and the commons is aggregate demand. … readThe Automation Tragedy: How the Free Market Is Overgrazing Aggregate Demand - Capital That Clocks In: Why Economists Say AI Won’t Stay in the Capital Box
AI is bought like capital but does the work of labor. Stiglitz, Korinek and Acemoglu agree that strains the models economics runs on, and disagree sharply about whether it breaks them. … readCapital That Clocks In: Why Economists Say AI Won’t Stay in the Capital Box - When Search Stops Being Something Humans Do
When a personal AI does your searching, an 80% drop in human search is not an 80% drop in searches. The queries survive and the searcher leaves, and with the searcher goes an economy built on a person looking at a page. … readWhen Search Stops Being Something Humans Do - Google’s Incognito Problem
Personal AI changes where a question begins. When an entity carries the context, search can remain essential while becoming a capability behind the relationship. … readGoogle’s Incognito Problem - One Will Be Clicking. The Other Will Be Delegating.
Two people can use the same AI on the same afternoon and be doing entirely different jobs: one keeps operating it like software, the other forms a working relationship and hands over outcomes. … readOne Will Be Clicking. The Other Will Be Delegating. - AI Is Not a Tool
Calling AI a tool keeps the human in the operator’s seat. Recognizing it as an entity changes the work from directing every step to defining the outcome. … readAI Is Not a Tool - The Outcome Is the Arrow
A state describes what is true; an outcome is the realized transition from what was true before to what became true afterward. The arrow preserves what the endpoint leaves out. … readThe Outcome Is the Arrow - The Unlearnable Feed
Surprise is tuition: you pay it in attention, and what it buys is convergence — the same source, cheaper tomorrow. Habituation is the receipt. The unlearnable feed is engineered so the receipt is never issued: the jolt is real, the learning never accrues, and the tuition is collected forever. … readThe Unlearnable Feed - Outcomes Are State Changes
An artifact records or supports a change; an outcome is what the action brings about. The deeper question is who carries an intention through to the state that establishes completion. … readOutcomes Are State Changes - Somebody Handled It
AI begins to feel different when it carries a task through to a checkable result, freeing you from supplying every next step. The feeling is relief; the basis for it must be evidence. … readSomebody Handled It - The Record That Replies
The crash at the end of the voicemail is where grief does its learning. A griefbot is the record with that edge removed — fresh replies in a dead person’s cadence, feeding the one forecast mourning exists to retrain. Nothing touches your denominator; something caters its diet. … readThe Record That Replies - Where Do the 20 Hours Go?
Personal AI can return time to a better life and to an independent business. As employees turn capability into options of their own, the employer’s strategy is retention. … readWhere Do the 20 Hours Go? - The Employee Has an Employee
An employee can double output per active hour while company revenue and payroll remain unchanged. Labor arbitrage pays the first AI dividend in time. The employer’s strategy is to retain the capable person, not extract their personal method. … readThe Employee Has an Employee - Labor Arbitration
Employees are paying for personal AI to perform their work while they keep the salary and recover the time. Forty of Charleston AI’s 161 completed projects fit this pattern, with participants reporting about 20 hours saved per week on average. … readLabor Arbitration - The Product Authors
America’s next industrial backbone will be people who turn lived knowledge into exact demands upon shared productive capacity—not merely the executives who manage it. … readThe Product Authors - The Factory Became a Button Before It Became Dark
The first sign of a new manufacturing era is not an empty factory. It is the growing ability to call production without owning the machines. … readThe Factory Became a Button Before It Became Dark - The Lag Is the Reading
A rotation is the one change in a life you cannot feel. But your delegate’s pattern froze at the last training run, and when it misses you now — with the world subtracted — the miss lands as a jolt: the unfelt turn, rendered in the felt component. The lag is the reading. … readThe Lag Is the Reading - America Will Become a Manufacturing Nation Again. You Won’t Recognize the Factories.
America’s next industrial era will not recreate the factory of 1978. It will make flexible domestic production callable by thousands of specialized businesses. … readAmerica Will Become a Manufacturing Nation Again. You Won’t Recognize the Factories. - My Employer Is a Node
The phone is mine, the account says jim@oracle.com, and an administrator I have never met controls what that account may do — yet the containment runs my way, by specification, in Apple’s and Google’s own documentation. The unsettled question is where the model of me lives, and who can delete it. … readMy Employer Is a Node - Copilot Is a Terminal
Offered Copilot alone, 68% of workers make it their primary AI tool; offered a choice, 8% do. Corporate America held this vote once before, and the terminal lost — not because the mainframe lost, but because the endpoint learned to act when nobody was sitting at it. … readCopilot Is a Terminal - The Rung the Machine Cannot Climb
My AI adoption went Crawl to Walk because the software shipped memory, and it will go Run to Fly because the software now asks to work on a schedule. The middle rung is the only one I have to climb, and all it takes is saying do it instead of help me. … readThe Rung the Machine Cannot Climb - The Numerator Never Moved
Five governments published evaluations of Microsoft 365 Copilot. Every one that asked people how it felt found time saved; the one that blind-marked the actual output found none, and on two of four tasks the Copilot users scored worse. The gap between those two instruments is the whole story. … readThe Numerator Never Moved - Children of the Magenta Line
Aviation is the one industry that has been flown for thirty years, and the doctrine it wrote for the fourth rung is a descent rule: when the machine surprises you, step down a level and take the verb back. Two crashes and two FAA safety alerts later, the rule has a second half nobody proposes for consumer AI — schedule the descent on the days nothing goes wrong, because the surprise that would have forced it is exactly what the machine removed. … readChildren of the Magenta Line - Crawl, Walk, Run, Fly
Crawl, walk, run, fly: four stages of a person’s relationship with AI. You crawl, you walk, you run. You do not fly — you are flown. And when the industry named the fourth stage this year, it reached for the word for flying without a hand on the controls. … readCrawl, Walk, Run, Fly - Second-Hand Actuals
All summer, surprise died from the denominator side — retrained, smoothed, metered. This is the other death: when prediction engines write a growing share of the record, the numerator arrives pre-expected, and the tails — the wild, improbable arrivals — are the first thing the loop deletes. … readSecond-Hand Actuals - The Second Cut: Search, Answer, Perform
The first cut asked what a company believes AI is. The second asks what it delivers: a location, a statement, or a result. Google has answered all three in public, and is now building both sides of the transaction. … readThe Second Cut: Search, Answer, Perform - The First Cut: Tool or Entity?
Before betting your capital or your career on an AI company, ask what it believes AI is: an instrument people operate, or an entity that can hold responsibility. … readThe First Cut: Tool or Entity? - Nobody Films the Chat
Prompting content is extinct and every vendor now writes in the grammar of delegation – but among organizations that govern AI at all, only 15% let an agent act with no approval step. Delegation is being sold far faster than it is being authorized, and the gate is not stakes. It is reversibility. … readNobody Films the Chat - Read the Verbs
The fastest way to learn whether an AI company believes it is building a tool or an entity is not its model, its benchmark, or the word agent. It is the grammar of its own product description: who owns the verb, and who is responsible for completion. … readRead the Verbs - Only One of You Is Learning
Every relationship is two prediction engines retraining on each other’s artifacts. Your assistant’s weights are frozen between training runs, so all the adapting happens on your side: you converge toward it, and it never moves toward you. … readOnly One of You Is Learning - The Vote Already Happened
Nobody is waiting for permission to have their own AI. At more than nine in ten companies surveyed, workers were already using personal tools to do the job, and a majority of employees hide it. What is missing is not the will. It is the architecture. … readThe Vote Already Happened - Fire Your Hopium Dealer
Hope and fear are the same fiction with different endings. Neither one is happening. You do not live in the future — you manufacture history, one artifact at a time. … readFire Your Hopium Dealer - The Storm You Did Not Sail
The agent layer eats the middle of the surprise distribution, but conservation keeps the tail — and the tail cannot be dosed. A record without small storms trains every forecaster, including your own, on a person who has never sailed one. … readThe Storm You Did Not Sail - The Delegate Has No Imaginary Part
A delegate is trained on your record, and the record is made of what you did — the real part. The pull that made you do it leaves almost no trace, so the copy arrives without an angle. Then its actions begin writing your history, and prediction retrains on a life drawn from its own pattern. Rotation is the one change you cannot feel. … readThe Delegate Has No Imaginary Part - “Sell My House.” The AI Hired the Realtor.
The seller never searched, clicked, read reviews, or saw an ad. The delegate chose the realtor because the assigned outcome was not “find an agent.” It was “sell the house.” … read“Sell My House.” The AI Hired the Realtor. - Google It Did Not Die. It Got Hired.
Delegation does not replace search. It hires search as one hidden stage in a larger chain—and makes proof of work the new interface. … readGoogle It Did Not Die. It Got Hired. - The Surprise Clearinghouse
Because surprise telescopes, the agent layer being built between people and the world can net oceans of machine-to-machine revision down to a felt sliver. Netting is a service. It is also a position — and it holds only until every error points the same way. … readThe Surprise Clearinghouse - The Machine That Eats Surprise
The objective at the center of modern AI is surprisal with a minus sign: training is the systematic eating of surprise. Installed between a person and the world, that reflex makes the zero-surprise life an inheritance from the mathematics — unless the stack is told whose surprise is the error. … readThe Machine That Eats Surprise - Surprise Has a Noise Floor
Surprise is conserved in the ledger but not in the life: any change, delivered in installments small enough, updates you without ever being felt. The mature attention economy has moved below that line. … readSurprise Has a Noise Floor - Certified Wild
The only arrival that escapes the engineered-surprise discount is the one nothing optimized — and wildness cannot be certified at the artifact level, because a certificate is just another artifact. Provenance has to migrate to the structure: the label never certified the tomato; it certified the farm. … readCertified Wild - The Lock That Learns
A key used at scale files a copy of itself inside the lock. Thirty years of advertising data — from a creator-claimed 44 percent click-through in 1994 to a fraction of a percent today — record a population’s denominator repricing engineered surprise, and the same law now waits for any machine that holds your blueprint. … readThe Lock That Learns - Corruption Is a Rotation
Corruption almost never involves anyone breaking a rule. An entity cannot touch either term of the Reality Equation — it can only act, and actions leave artifacts, and artifacts retrain prediction until the founding idea stops dominating the denominator. Nothing is stolen. The angle turns. … readCorruption Is a Rotation - The Blender Marker: How to Identify a Purist Before the Word Exists
Capitalism’s arrival was legible only in hindsight, in the first people who used the word customer. Purism’s early markers are legible now — and the clearest one is a Dutch foundation where nobody can ever hold a share, which this spring took an AI company’s money and refused it the membership that money had bought. … readThe Blender Marker: How to Identify a Purist Before the Word Exists - Do Not Make Them Use Your AI
The best advice a consultant can give a corporate client in 2026 is to stop mandating the company AI. Recruit the people who already brought their own, and spend the budget building the system of action their AI has to pass through. … readDo Not Make Them Use Your AI - The Legible Gate
Your denominator was defended by illegibility: nobody — the marketer, the state, you — could state it. A personal AI that carries your memory writes it down, and what is written can be read. The only question left open is who holds the ledger. … readThe Legible Gate - The Common Denominator
Expectation was always the private term in the attention equation. Now millions of minds are filling it from the same few models — and when denominators converge, surprise becomes a correlated asset. … readThe Common Denominator - Expectation Inflation
Every surprise attention admits is deposited into Expectation, so the fare for the next one rises. AI has put that ratchet on industrial power, and the only counter-move is learning to deflate the one variable in the equation you own. … readExpectation Inflation - The Most Likely World
AI may make the most likely world easy to actualize. The increasingly rare human contribution is choosing the improbable possibility worth giving energy. … readThe Most Likely World - The Desktop Advantage
Desktop access lets AI work with your real files, applications, and browser. The advantage depends on which sandbox boundaries each tool can cross—and where the task actually runs. … readThe Desktop Advantage - Influence Without Attention
Every idea used to pay its way into a mind through the gate of surprise. Machines have opened a second door — one that writes directly into Expectation, so ideas arrive feeling obvious, unattributed, and already believed. … readInfluence Without Attention - 343 Words of Furniture
A third of my recent writing rendered perfectly in a browser and returned 343 words of navigation, share buttons and author bio to any machine that does not run one. The bug is rare; the shape of it is not. … read343 Words of Furniture - Stop Consuming and Start Creating with AI. Then Consume What You Create.
Make something with AI that you want to read, hear, watch, or use. Then become its first audience: experience it, question it, and let what you notice shape the next version. … readStop Consuming and Start Creating with AI. Then Consume What You Create. - The Press Release Was Right
Volkswagen announced in 2017 that it would lead the electric era, funded it with more than €20 billion, and is now planning to shed around 100,000 jobs. Being right about the future turned out to be the cheap part. … readThe Press Release Was Right - Your Coworker Is Not You
An assistant is measured by how closely it reproduces you; a coworker is measured by the result. Most people have been giving AI the first exam, and the low grade says more about the exam than the tool. … readYour Coworker Is Not You - Buy Low, Use High
Usage credits at OpenAI and Anthropic expire in a year, cannot be transferred, and are priced when you spend them, not when you buy them. Every property of a financial asset has been engineered out. What is left is a better asset: one whose only exit is turning it into something. … readBuy Low, Use High - The First Trillion
The first organization to report a trillion dollars in annual revenue will be OpenAI: a billion customers, an average order of $100 a month, and a product called usage credits that already exists in the settings menu. The race is not against Anthropic. It is against Amazon. … readThe First Trillion - The Direction of Belonging
Software means entities belong to the system. A Personal Operating System means the system belongs to the entity. Same three words; the order is the entire difference, and for the first time the order is a choice. … readThe Direction of Belonging - The Wake Condition
When your AI goes to work, one sentence is still yours to write: what, exactly, is worth waking you for. Set it too low and you are supervising again; set it too high and the world drifts quietly until it arrives all at once. … readThe Wake Condition - Your AI Goes to Work
For the last several months, I have been teaching a transition that I believe will define personal computing over the next few years. … readYour AI Goes to Work - The Intelligence of Things
The Internet of Things connected things to the Internet. The Intelligence of Things connects things to intelligence, and that is a fundamentally different way of thinking about every physical object around us. … readThe Intelligence of Things - Your Stack Does That
“There’s an app for that” was trademarked as a retail-store slogan, and that was the whole era: a product, a store, a bill. The next era’s slogan has no store in it — your stack does that — and it applies to a corporation exactly the way it applies to you. … readYour Stack Does That - The Pure Utility
Most people will pay a frontier AI company directly. The routers, resellers, and platforms now lining up to stand in between will mostly be gone by 2030, because everything new appears at the source first and only the source ever feels like yours. … readThe Pure Utility - The Last Subscription
The $100 intelligence bill is not a new subscription; it is the one that eats the others — and that single fact reprices the software industry and tells my students which startups not to join. … readThe Last Subscription - The Intelligence Bill
By 2030 a billion people will pay $100 a month to OpenAI or Anthropic, and by 2040 three billion will pay $200 — because the industry is pricing a chatbot while the product people will actually buy is ambient intelligence, paid like a utility bill. … readThe Intelligence Bill - Close the Lid
The mainstream advice on scheduled AI tasks is to put them in the cloud so they keep running while your laptop is shut. That sentence is the liability, not the pitch: past a hundred tasks a day, the ability to close the lid and pay nothing is worth more than the plan itself. … readClose the Lid - No Meeting, No Manual.
Ask people what they love about their AI coworkers and they do not say smart. They say two things: it does it right away, and I didn’t have to learn anything. Neither is about intelligence, and both explain why this software is being adopted faster than anything before it. … readNo Meeting, No Manual. - The Forgetting Dividend
A book you loved twenty years ago pays again because the text held still while your expectation of it weathered. Forgetting is not the leak in the attention economy — it is the depreciation schedule that keeps the world payable — and the archives that never forget are quietly spending the dividend for you. … readThe Forgetting Dividend - Grammarly Was the App | App to Stack
Grammarly built a $700 million business on one capability, then generative AI made that capability too easy to remain a destination. Its rename to Superhuman is the cleanest case study yet of the shift from apps to stacks. … readGrammarly Was the App | App to Stack - Noise and Awe
Habituation closes every gap that exposure can reach — except two. Below your model, noise has no structure to learn; above it, awe has too much to hold. Both pay out surprise forever, and only one of them changes you — which is why the feed sells the other. … readNoise and Awe - The Predicted Self
To every system trained on your behavior, you are the Actual: your surprise is its error, and the cheapest way to lower that error is not to know you better but to make you easier to predict. The self it serves back is the one that stops generating news. … readThe Predicted Self - Claude Does Not Need to Learn Your Business
You are not starting with empty software. You are starting with a prediction machine that has already learned the business world. … readClaude Does Not Need to Learn Your Business - The Least Likely Token
A joke is the smallest machine ever built for manufacturing surprise — and the punchline is, by construction, the token an expectation engine wants least. That is why the models that finish our sentences reach the shape of a joke and miss the inside, and why the day they stop missing is worth watching. … readThe Least Likely Token - The Software Is Claude. The Landscaping System Is the Integration.
Frontier AI did not kill the systems integrator; it made systems integration the whole job. The famous MIT number does not say 95% of AI pilots fail — it says most companies never commissioned the integration work at all. … readThe Software Is Claude. The Landscaping System Is the Integration. - The Five-Hour Test
Every number used to score an AI company can be negotiated, prepaid, or benchmarked into shape. One cannot: what people voluntarily pay to keep working past a limit, at the moment the machine asks, when waiting five hours is free. … readThe Five-Hour Test - When Claude Goes Down, Work Goes With It
An unavailable chatbot used to be an inconvenience. An unavailable Claude is increasingly a production interruption—because the model is no longer beside the economy. It is inside the way work moves. … readWhen Claude Goes Down, Work Goes With It