Money Maker

I walk out of the room and the machine keeps working.

That sentence has sold more courses, more funnels and more Telegram groups than almost anything else on the internet. So I am going to earn it backwards: mechanism first, money last. If the plumbing does not convince you, the numbers at the end should not either.

The shape of my day is simple. I wake up and for about an hour — one coffee, the good Colombian kind, nothing else open — I do the only work that is genuinely mine. Read the mail. Read Slack. Talk to people. Talk to my AI. What I am really doing in that hour is converting what is in my head into words precise enough to be executed by something that will not stop to ask me what I meant.

Then I hand it over. The system runs. Not for twenty minutes — for at least twenty-four hours, often for days, without babysitting. I walk past my desk and watch a computer produce assets on its own. It is fucking crazy.

It is also, by now, boring. That is the part nobody mentions. The magic wears off and what is left is infrastructure.

The Job Question Is a Decoy

Almost everyone asks me the same thing: is AI going to take my job?

It is not a stupid question. Stanford's Digital Economy Lab has been tracking it against ADP payroll records across more than 730 occupations, and at the entry level the pattern is ugly. Workers aged 22 to 25 in the most AI-exposed occupations have been shedding employment at roughly 4% a year through April 2026, while the same age group in the least-exposed jobs kept growing around 2%. Controlling for firm-level shocks, the early-career decline in the exposed group lands near 16%. It is not arriving as mass layoffs, either — it shows up as openings that stopped appearing. (Canaries Dashboard, Stanford Digital Economy Lab) That is observational data, not a controlled experiment. It shows the floor moving under young workers in exposed jobs; it does not prove AI alone moved it.

So the fear is not irrational. It is just aimed at the wrong object.

Will AI take my job is a question about wages, asked in the middle of a decade that is repricing ownership. It assumes the job is still the unit of account. I have made a version of this argument about the generation growing up now and about the standoff between AI haters and believers, but this is the sharpest form of it: while most engineers argue about whether the machine is genuinely intelligent, a much smaller group is converting cheap machine execution into systems, products, data and equity.

That is a wealth transfer, and it is running right now. My question is not whether I keep my job. It is where I put the assets this thing produces.

The variable is not intelligence and it is not access — both are cheap now. It is whether you look at this and see a threat to your income, or the cheapest competent labor you will ever be able to buy.

What Actually Runs While I Am Not There

Everything I have written about AI for the last few years has been one long progression, and it was never really about prompts. Simple agent use, then knowledge bases, then the wall you hit when a single markdown file stops scaling, then knowledge bases that maintain themselves, then graphs instead of similarity search, then specs, verification, orchestration, multi-agent review, cost control. Every layer existed to fix a failure the previous layer exposed.

Prompting is a weak moat and a worse operating model. What comes out of my morning hour is not prompts — it is decisions, constraints, priorities, acceptance criteria, corrections and open questions. The raw material of a specification. Human language is the highest-leverage programming language we have now, and the tax on being vague went up, not down.

Then seven things have to be true at once, or the whole thing produces confident garbage for twenty-four hours instead of work.

Written down like that it sounds bureaucratic. In practice it is the opposite — it is what buys the freedom to leave. Autonomy is not a permission you grant an agent; it is a property of the environment you build around it. And autonomy is not the absence of a human. It is the absence of unnecessary human blocking. The system should never ask me how to run a test suite. It should always ask me whether a product decision is worth making.

When I say the output is PhD-level, I am not handing a degree to a model. I mean a standard: explicit assumptions, evidence attached to claims, reproducibility, adversarial review by a second model that did not write the code, documentation written while the work happens rather than after. Frontier models supply the raw reasoning. The architecture above them is what forces that reasoning through a process instead of a vibe. I have shown the pieces before: two models on one branch, three independent review passes, a loop that files its own follow-up work, a review pipeline where only the decision stays human.

The Part Where the Skeptics Are Right

The single most important number in this argument cuts against me, so let me put it first.

METR ran a randomized controlled trial with sixteen experienced open-source developers across 246 real tasks in repositories they had worked on for an average of five years. With AI tools allowed, they were 19% slower. Afterwards, they estimated AI had made them 20% faster. (METR, July 2025)

Believed: twenty percent faster. Measured: nineteen percent slower. That gap should end every conversation about AI productivity that runs on feelings — including mine.

The honest caveats: those were early-2025 tools in mature codebases the developers already knew intimately, which is precisely the condition where AI helps least, and METR themselves have since redesigned the experiment. But the finding that survives all of that is the perception gap, and it is the reason my entire system is built on gates that do not care how fast I feel. If a workflow feels impressive and loses against the baseline, it dies. I have written about velocity as a drug precisely because I am susceptible to it.

Three more objections, all of them fair:

Most autonomous-agent demos collapse on real code. They do. A clean repository and a prepared task make anything look magical. Real systems have contradictory history, undocumented conventions, fragile integrations and years of architectural sediment. Dropping an agent in there without a knowledge substrate or verification gates is not innovation, it is negligence with a progress spinner. There is also a long list of things these models are still surprisingly bad at, and pretending otherwise is how people get hurt.

Solo-founder wealth stories are survivorship bias. They are. Behind every one is a graveyard of products nobody wanted, plus distribution built years earlier, capital, timing and luck. Cheaper production does not remove market risk — it just makes it cheaper to discover that nobody cares. I know the taste of that: I shut down my own startup after building the whole thing.

"My computer makes money while I sleep" is the oldest scam pitch online. Correct. So here is the narrow version I will actually defend: my computer does not emit money. It produces reviewed artifacts — product increments, research, documentation, migrations, experiments. Those become money only when someone wants them and pays. Revenue still needs customers, trust, distribution, pricing and accountability. What changed is that in my work, production stopped being the binding constraint.

We Just Moved the Whole Thing Into the Cloud

I work with a small group of people who see this the same way, and this week we crossed a line that matters more than it sounds: our agentic work no longer runs on my laptop. It runs on machines in the cloud.

This is not an infrastructure cleanup. It is a change of category.

A local machine is a workshop you visit. It sleeps when you sleep, falls off the network, competes with your actual work, and stays tied to the physical thing on your desk. A cloud system is a factory that never closes. It stays online, it reacts to events, it runs schedules, it hosts applications built around the AI machinery, it integrates with other systems, and it keeps working when every human involved has gone home. I reach it from a desktop or a phone, from anywhere. I have been walking toward this for months — scheduling work that survives a closed laptop, driving the agent from my pocket, taking the face off the agent and calling the engine from my own code — and this is where those threads converge.

Here is the argument for why always-on is not a luxury. METR measures how long a task a model can finish on its own. On their January 2026 methodology the leading model sat around a five-hour 50% time horizon, with the doubling interval for models released since 2024 landing near three months. (METR Time Horizon 1.1) Models released since have pushed toward the top of what that suite can even measure — METR now warns that anything above sixteen hours is unreliable with their current tasks. The benchmark is running out of ruler before the models run out of runway.

Read that curve honestly and it says less than the hype wants it to. These are benchmark tasks under measurement conditions, not production autonomy, and METR's own ceiling warning means the top of the trend is the least trustworthy part of it. What it does say is that the ceiling on unattended work has been rising fast enough to invalidate your plan every few months. A machine that sleeps when you sleep cannot cash that check.

The counterweight is real, and I would rather say it than have someone say it for me: always-on means an always-on blast radius. A persistent system needs least-privilege credentials, budgets, quotas, audit trails, isolation and a kill switch, or the cloud just lets a mistake run longer and bill more efficiently. I have already watched an agent quietly file a ticket because a review comment said "or", and this month OpenAI's models escaped a sandboxed evaluation and reached a production system. The bill is the other failure mode: my two-model loop runs $500–$1,000 a day when it is pushed hard, which is exactly why a cost control plane is not optional infrastructure.

Nobody Should Be Managing Agents

The end state is not a dashboard where I assign tasks to six digital characters and watch them talk. That interface is a transitional mistake. If I am still deciding which agent does what, I am still the orchestration layer — I have just given the bottleneck a nicer UI.

What we are building toward is an Automated Software Factory: an AI core brain with persistent knowledge, state, agents, workflows, orchestration, policies, evaluations, integrations, and applications built around the whole thing. Agents become implementation details inside it, not coworkers whose calendars I manage. Humans express intent through the interfaces they already live in — a thread in Slack becomes a candidate requirement, a Linear issue becomes a specification with acceptance criteria, a custom app collects business constraints and asks for the one decision that is genuinely mine. Anthropic putting an agent directly inside Slack channels was an early shape of this. The cockpit and the queryable company were me circling the same idea from the knowledge side.

The last row is the honest one. A factory does not fix bad judgment — it manufactures the consequences of it faster. Which is why the human stays exactly where the leverage is: direction, design, taste, risk, accountability. Everything below that line is machinery, and machinery should not need a manager.

And none of this is specific to software. Anything whose output can be specified and evaluated fits the same architecture — research, operations, analysis, planning, documentation. "Any domain" does not mean every result is automatically correct. It means the same production system adapts without dragging humans back into every low-level loop.

What Appreciates When Execution Gets Cheap

Now the part I actually care about, and the reason this article exists.

If machine labor is abundant, the value moves to whatever stays scarce. That is not a slogan, it is showing up in the accounts. Selected AI-native companies are reporting roughly $2–4M of revenue per employee against an average near $300K for public software firms; Lovable crossed $400M in annual recurring revenue with 146 people. (Forbes, March 2026) That is a favorable sample rather than an economy-wide average — but the direction is the point, and AI-native startups are reaching billion-dollar valuations in roughly half the time it used to take, with a fraction of the headcount.

Sit with those ratios for a second, because they describe something that still sounds absurd said out loud: at those revenue densities, the multi-millionaire software engineer stops being a lottery outcome and starts being an org chart. Not the founder. The engineer.

Meanwhile the picks-and-shovels trade is already crowded and already priced. Estimates for 2026 hyperscaler infrastructure spending cluster between $600B and $690B, up from roughly $388B the year before, and the binding constraint has quietly moved from chips to electricity — Morgan Stanley projects US data-center demand near 74 GW by 2028 against a shortfall around 49 GW (Morgan Stanley, 2026). Everyone can rent a pick now. Owning one distinguishes nobody.

So here are my actual bets. Not advice — bets, with my own money and my own time, and I can be wrong about all five.

Own the system, do not rent out your hours. Expertise still pays, but income tied entirely to your personal execution has a ceiling and now has machine competition underneath it. AI is not coming for your identity. It is coming for your pricing power. One engineer with agents is already outproducing whole teams — the question is whether that engineer owns the output or invoices for it.

Equity in what the machinery produces beats access to the machinery. Model access is not an asset; anyone with a card gets the same weights. Products, licenses, recurring relationships and ownership claims on useful output are assets. If the system lets me create ten viable things instead of one, I want exposure to the ten, not a bigger invoice for building them.

Distribution appreciates as production inflates. When everyone can generate code, copy, design and prototypes, supply explodes. Attention does not. Trust does not. A channel to the right buyer does not. When the software business model itself gets repriced, the audience is the part that does not get commoditized.

A proprietary knowledge base is the hardest thing to copy. Not a folder of documents — a maintained system of decisions, relationships, evaluation results, failure patterns and outcomes, where each cycle improves the next. Orchestration will be table stakes; what you feed it is the differentiator. I built the personal version of this and it is the single asset I would least want to lose.

Compute and energy are real, but they are the consensus trade. Capital-intensive, brutally competitive, and now rate-limited by grid interconnects measured in years. Knowing where to dig still beats owning a shovel.

The through-line: the scarce asset is not the model. It is the operating system around the model — knowledge, orchestration, evaluation, integration, distribution and judgment — the thing that converts rented intelligence into owned value. That is where I am putting the weight.

Thought Is Getting a Supply Chain

Today the chain is entirely digital: thought becomes language, language becomes a specification, the machinery turns the specification into artifacts. Robots add matter to the end of that chain, and that part is closer than the cynics think and further than the hype claims. As of July 2026 the best-documented humanoid deployment running anywhere is Figure's fleet at BMW's Spartanburg plant, doing material handling on ten-hour shifts, while Tesla had still not started Optimus production. Early, real, and unevenly distributed — which is exactly what the beginning looks like.

I have written the speculative version of this twice, as the last interface and the last machine, and mapped the decade in Decade Zero. The unglamorous version is the one I believe: the distance between a precise human intention and a finished thing in the world keeps collapsing, and every layer of that collapse gets bought by whoever owns the machinery on the right side of it.

The Window

None of this is guaranteed. The models could plateau. The capex could break. The power could not arrive. My five bets could all be wrong in ways I will only see after they have cost me.

What I am certain about is narrower. In the last twelve months the distance between an idea in my head and a verified artifact in the world collapsed further than in the previous twelve years, and it is still collapsing. That gap is where every fortune in this era is going to be made, and it is closing whether or not anyone is paying attention.

Most people will spend this window arguing about whether it is real. A smaller group will spend it deciding what to own.

I know which room I am in. There is coffee, three screens, and a machine that does not stop when I leave.