The Starvard sealStarvard.
English
EnglishFrançaisEspañolहिन्दीPortuguêsBahasa IndonesiaAll languages →
Starvard University

The Starvard Blog

Notes from the free university

The One-Person AI Agency: What It Is, What It Isn't, and Whether a Single Human Can Really Run a Company of Robots

October 20, 2026 · By the Starvard professors
One founder conducting a semicircle of robot department heads like an orchestra
Illustration by Starvard (AI)

A single person can now run what used to look like a small company, but the fine print matters. This post separates the solo professional, the AI automation agency and the service-as-software company, checks the famous numbers against their sources, and ends with the sane version: judgment, trust and distribution stay human, while AI takes the production work.

In the fall of 2023, Sam Altman told an interviewer that he and his tech-CEO friends had a little group chat with a "betting pool" in it. The bet: what year would the world see its first one-person billion-dollar company?

Then in May 2025, Anthropic's CEO Dario Amodei went further and actually named a year. At Anthropic's first developer conference, he predicted that the first billion-dollar company with a single human employee would show up in 2026.

It is now October 2026. So: did it happen?

Sort of. Not really. Kind of? It depends on what you count, who's counting, and whether you read the fine print. And that "sort of, not really, kind of" is a pretty good summary of the whole topic we're about to dig into.

Because somewhere between the billionaire group chats and the YouTube thumbnails promising "$10K/month with AI agents, no experience needed," there's a real thing happening. More people really are running real businesses alone, with AI doing work that used to need a team. Some are making serious money. Many more are making a little, or nothing, or mostly making money by teaching other people how to make money.

We're going to start from the very bottom, with what an agency even is, and build up until we can answer the big questions honestly: What changed? Who's really doing this? Which numbers are real, and which are sales pitches wearing a lab coat? And if you're thinking "could I do this?", what would a sane version look like?

Let's go.

Part 1: What Is an Agency, Anyway?

Forget AI for a second. Let's start with the boring, ancient thing.

An agency is a business that does work for other businesses. A marketing agency writes your ads. A design agency builds your brand. A dev shop builds your app. A bookkeeping firm does your books. The client has a job they need done, doesn't want to hire full-time people to do it, and pays someone else instead.

Here's the first-principles truth about agencies that everything else in this post hangs on:

An agency sells hours of human attention, wrapped in judgment.

When you hire a design agency, you're paying for two different things that are usually bundled together:

  1. Judgment: someone who knows what good looks like, understands your problem, and decides what to do.
  2. Hands: someone (often a different, more junior someone) who actually produces the 40 versions of the logo, the 12 landing pages, the 300 social posts.

For a century, the agency business has had one big structural problem, which we'll call the Hands Problem: judgment is scalable-ish, but hands are not. A brilliant creative director can make good decisions for ten clients at once. But she can't personally produce the work for ten clients at once. So she hires juniors. And the juniors need a manager. And the manager needs an office, payroll, HR, and a project management tool that nobody updates.

This is why agencies are shaped like pyramids. A few senior people at the top with the judgment and the client relationships. A broad base of people at the bottom doing the production. The money is made by billing clients for the time of the people at the bottom at a higher rate than you pay them.

So if you're a talented freelancer, you've always hit a wall. You can only bill so many hours. To grow past that, you have to stop being a freelancer and start being a manager, which is a totally different job that many great freelancers hate and are bad at.

Hold onto that picture of the pyramid. The entire "one-person AI agency" idea is basically one question:

What happens to the pyramid when the bottom layer can be software?

Part 2: What Actually Changed

To be honest about this, we need to separate three different things that got lumped together under "AI."

Change #1: The machines learned to make first drafts

Starting around late 2022, language models got good enough to produce passable versions of a lot of agency output: copy, emails, blog drafts, code, summaries, research notes, and with image and video models, visuals too.

"Passable first draft" sounds modest. It isn't. In an agency, a huge share of the labor at the bottom of the pyramid is first drafts. If the first draft now takes 30 seconds instead of three hours, the economics of that bottom layer change completely.

Change #2: The glue got cheap

A first draft is useless if a human has to copy it from one window and paste it into another all day. The second change is less glamorous: workflow tools like n8n, Make, and Zapier, which let you wire apps together ("when a new lead fills in this form, research their company, draft a personalized email, put it in the CRM, and ping me on Slack"). Plug a language model into the middle of those pipes and you get something that can take a messy input, do a bit of thinking, and push an output somewhere useful, over and over, without you touching it.

Change #3: The machines started doing multi-step work

The newest layer is what everyone calls agents: AI systems that don't just answer one question but take a goal and work through several steps toward it, using tools along the way. Coding agents like Claude Code and Cursor are the most mature example. Pieter Levels, the famous solo founder we'll meet properly in a minute, wrote on his blog in May 2026 that he hadn't written code by hand in about six months; he directs an AI coding agent running on a server instead.

Put these three together and you get a new picture of the pyramid:

Picture the old shape and the new one side by side. Before: you sit at the top, managers sit beneath you, and a row of juniors underneath them produces the drafts. After: you are still at the top, but the layer beneath you is AI agents and automations doing the drafts, research, code, follow-ups and reports.

The bottom of the pyramid gets replaced by software that costs tens or hundreds of dollars a month instead of tens of thousands per person per year. The middle (managers) disappears, because there's nobody to manage. And you're left at the top, alone, with all the judgment and all the client relationships and a small army of tireless digital interns.

That's the dream. Now let's see what reality has done with it.

Part 3: Three Different Creatures Wearing the Same Name

Here's something that makes this topic confusing: when people say "one-person AI agency," they mean at least three different business models, and they constantly blur them together. (Sometimes on purpose, because the impressive numbers come from one model and the "you can do this too" pitch comes from another.)

Creature A: The AI-Augmented Solo Professional

This is a freelancer or consultant (a copywriter, designer, marketer, developer, analyst) who sells the same kind of service they always did, but uses AI to do it faster, take on more clients, or deliver more per client. The AI is invisible to the client. The client is buying "great landing pages," not "AI."

Creature B: The AI Automation Agency

This is a business whose product is AI itself: you go into a dental clinic, a law firm, or a real estate brokerage and build them an AI chatbot, a lead-qualification system, an automated reporting pipeline. The term "AI Automation Agency" (or "AAA") was coined in 2023 by a New Zealand entrepreneur named Liam Ottley, who runs an agency called Morningside AI and teaches the model to a very large online audience.

Creature C: The Solo Company Where Software Is the Service

This one isn't technically an agency at all. It's a single founder who builds a product (an app, a website, a telehealth storefront) where AI does the work that would have been done by staff. The venture capital firm Foundation Capital has a nice name for this trend: "service-as-software." Their argument is that AI doesn't just give humans better tools; it increasingly does the job itself, so the market isn't just "software" anymore, it's all the salaries and outsourced services that software can now replace. They put a figure of $4.6 trillion on that opportunity, though it's worth remembering this is an investor's estimate of a market they're investing in, not a measurement.

These three creatures have very different economics, very different risks, and very different odds of success. Most of the jaw-dropping headlines come from Creature C. Most of the "start your own AI agency" courses are selling Creature B. And most of the people who'll actually succeed quietly will be Creature A.

Keep the three of them in mind. We'll keep checking which one we're looking at.

Part 4: The Scoreboard (Real People, Real Numbers)

Let's look at who's actually doing this. For every number, I'll tell you where it came from, because in this world the source matters as much as the number.

Pieter Levels: The patron saint of solo

If the one-person internet business has a mascot, it's Pieter Levels, a self-taught Dutch developer who has been building businesses alone for more than a decade (Nomad List, Remote OK, and more recently AI products like Photo AI and Interior AI). He publishes his revenue openly.

The numbers below are founder-reported, which means Levels posted them himself. He has a long public track record, but nobody audits his Stripe dashboard for you.

Levels is Creature C. He's also a lesson in something that will keep coming up: his biggest asset isn't AI. It's an audience of hundreds of thousands of followers who see every launch. When he went on the Lex Fridman podcast in 2024, he reported sign-ups across his sites roughly doubled overnight. AI made building cheap. Distribution made it pay.

Maor Shlomo and Base44: The six-month, $80 million exit

In June 2025, the website builder Wix bought Base44, an AI app-building tool founded by Israeli developer Maor Shlomo, for $80 million in cash, with further payments possible through 2029 if performance targets are met. Base44 was about six months old and had raised no outside money. Shlomo had started it as a side project and documented the journey on LinkedIn and X; in his posts, he reported the company made about $189,000 in profit in the month before the sale.

This is the closest thing to a clean "solo founder hits it big with AI" story we have. But read the fine print: TechCrunch reported that Base44 had eight employees by the time of the sale, and they collectively got $25 million of the $80 million as retention bonuses. Solo-owned, yes. Solo-run, no.

Matthew Gallagher and Medvi: The near-miss unicorn (and the cautionary tale)

In April 2026, the New York Times published a story that made the whole internet sit up: Matthew Gallagher, a 41-year-old in Los Angeles, had launched a telehealth company called Medvi with about $20,000 and a stack of AI tools. Medvi sells access to compounded GLP-1 weight-loss drugs online. According to the Times, which said it was given access to the company's financials to verify them, Medvi generated $401 million in revenue in 2025, its first full year, with a net margin reported at about 16%. Gallagher's only employee was his brother.

Gallagher reportedly used ChatGPT, Claude, and Grok to build the website and code; Midjourney and Runway for ad creative; and ElevenLabs voice tools and custom AI agents for customer service. He summed it up neatly: "It's not an AI company, but I did it with AI."

For a few days, this looked like Altman's prophecy fulfilled. Then the asterisks arrived. Hold that thought: we'll come back to Medvi in Part 5, because it is the single most instructive story in this whole post.

Liam Ottley and Nick Saraev: The AI Automation Agency builders

Now Creature B. These are the two names you'll run into most if you start looking into AI agencies.

Liam Ottley founded Morningside AI and coined the "AI Automation Agency" model in 2023. His community has grown to over 300,000 members, and his YouTube channel had roughly 816,000 subscribers as of mid-2026. Ottley lays out the path as a ladder: start as a solo "automation agency" doing projects of around $3,000, $5,000, grow into a staffed "AI agency" doing $30,000, $100,000 projects, and eventually become a strategic "AI transformation partner." Independent reviews note he has claimed more than $7 million in revenue and built a team of 40+ at Morningside, which is worth noticing: the person who popularized the one-person AI agency does not run a one-person AI agency.

Nick Saraev, a Canadian entrepreneur, runs an AI growth agency called LeftClick. His own bio says his two service businesses scaled to a combined $160,000 a month in revenue (self-reported). He also runs Maker School, a paid online community on the Skool platform that teaches people to land their first AI automation client; one reviewer reported it at about $184 a month, and that it had about $290,000 in monthly recurring revenue during a Skool competition.

Saraev has one of the most useful one-liners in the whole space: "AI is B2C. 'AI for money' is B2B." Translation: consumers will play with AI for fun, but businesses only pay for AI that makes or saves them money. Remember that line. It's the key to Part 8.

The bigger picture: are solo companies actually becoming more common?

Individual stories can be cherry-picked. So let's look at data from people who see thousands of companies.

So yes: the trend is real. More people are going solo, and AI is part of why. Now let's talk about why almost none of them are actually alone.

Part 5: The Fine Print

Every number in Part 4 is true, or at least truthfully reported. And yet if you walked away from Part 4 thinking "so one person plus AI can now do what a company does," you'd be wrong in ways that matter. Here are the asterisks.

Asterisk #1: "One person" almost never means one person

Look back at the list:

That last bit about Medvi is the important one. Medvi didn't eliminate the hard human work of telehealth. It rented it. The doctors still exist; they just work for someone else. This is a pattern worth naming: the one-person company is often a one-person layer sitting on top of other companies' people. That's a perfectly legitimate business model, but it's not the same as AI replacing a workforce.

Asterisk #2: AI lets you scale your mistakes, too

Now back to Medvi, because the story got a lot more complicated after the Times piece ran.

Reporters and readers quickly pointed out several things the original article hadn't mentioned:

The New York Times later updated its article to note the FDA warning letter and the pending class action. And the headline "$1.8 billion" figure? That was Gallagher's own projection for 2026, not reported revenue.

None of this means Gallagher isn't a genuinely skilled operator. He clearly is. But it illustrates something deep about one-person companies: in a normal company, there are lots of humans whose job is to say "wait, should we be doing this?" Compliance people. Lawyers. Editors. A colleague who notices the site is down. When you replace the team with AI, you also remove most of the friction, and some of that friction was protecting you.

Asterisk #3: Survivorship and self-reporting

Notice how many of our numbers are self-reported: Levels's revenue, Shlomo's monthly profit, Saraev's agency revenue, Ottley's claims. That doesn't make them false. But we only hear from the people who succeeded and chose to post screenshots. The thousands who tried the same thing and earned $400 in six months don't make YouTube videos about it.

Asterisk #4: Follow the money in the "AI agency" world

Here's something worth sitting with. Among the most visible people in the AI Automation Agency world, a large share of the money appears to come from teaching the AI agency model, not from running one.

One newsletter writer's breakdown of Ottley's business (an outsider's estimate, not an official figure) guessed that his coaching program brought in more per month than his agency did. Saraev's Maker School is itself a substantial business. This isn't a scandal. Teaching is a real, valuable business, and both men have actual agency experience, which is more than many course-sellers can say. But it's a classic pattern in every gold rush: the most reliable money is often in selling shovels.

So when you see a video promising that you can start a $10K/month AI agency, it's worth asking: is this person's income coming from clients, or from people like me watching this video?

Part 6: What the Research Says About AI Productivity

Okay. So the headline stories come with asterisks. But surely the underlying premise holds: AI makes people dramatically more productive, right?

The research says: it's complicated, and it's moving fast. Here are four of the most cited data points, and what they actually mean.

The study where AI made experts slower

In 2025, the nonprofit research group METR ran a randomized controlled trial with 16 experienced open-source developers working on 246 real tasks in large codebases they knew well. Half the tasks allowed AI tools; half didn't.

The result: with AI, developers took 19% longer. And here's the wild part: before the study they expected AI to make them about 24% faster, and afterward, they still believed it had made them roughly 20% faster.

Important caveats: these were experts in their own enormous, familiar codebases, which is the hardest case for AI; the tools were early-2025 versions; and METR itself said AI likely helps more on smaller projects, unfamiliar tasks, or with less experienced developers. Tools have improved a lot since. But the lesson that survives is about perception: AI can feel like it's helping even when it isn't. If you're running a business alone, nobody's checking your stopwatch but you.

The "95% of AI pilots fail" report

In July 2025, a team associated with MIT's Project NANDA published a preliminary report called The GenAI Divide, which found that about 95% of organizations in its research saw no measurable financial return from their generative AI initiatives, despite $30-40 billion in enterprise spending.

That number went viral and was widely misread as "AI doesn't work." The report's own explanation was different: the failures were mostly about integration, meaning tools that didn't fit into real workflows or didn't learn and adapt, not about the models being bad. Critics also noted the sample was limited and partly self-reported, so the 95% figure shouldn't be applied to every company everywhere.

But here's the detail that matters most for our topic: the report found the strongest returns came from back-office automation, including cutting business-process outsourcing and reducing spending on external agencies. Read that again. AI's clearest wins inside big companies include replacing agencies. Which is a threat to traditional agencies, and an opening for anyone who can deliver that automation well.

The agent hype check

In June 2025, Gartner predicted that over 40% of agentic AI projects will be canceled by the end of 2027, citing escalating costs, unclear business value, and weak risk controls. Gartner also coined a useful term: "agent washing," meaning vendors slapping the "agent" label on ordinary chatbots and automation tools. It estimated that only about 130 of the thousands of companies claiming to sell agentic AI were the real thing.

This is a forecast, not a measurement. But it's a useful warning for anyone thinking of selling "AI agents" to small businesses: a lot of what's being sold under that name isn't, and a lot of what is will get canceled.

The Klarna boomerang

Finally, a story from the real world. The fintech company Klarna became the poster child for AI replacing workers when it said in early 2024 that its AI assistant was doing the work of roughly 700 customer-service agents. Then, in May 2025, CEO Sebastian Siemiatkowski told Bloomberg the company would invest in human support again, admitting that cost had been "a too predominant evaluation factor" and quality had suffered. The AI didn't vanish; Klarna moved to a hybrid model, with AI on simple cases and humans for the rest.

So what do we conclude?

Put these together and a pattern appears. Think of any piece of work as having two parts:

Here is the uncomfortable truth for the one-person AI agency: clients have always mostly paid for the last 20%. Nobody hires an agency because they can't produce a logo. They hire an agency so that someone competent takes responsibility for producing the right logo. AI collapsed the cost of the first 80%. The last 20% is still a human job. Which means it's your job, and there's still only one of you.

Part 7: The Economics, From First Principles

Now we can actually reason about this, instead of just collecting anecdotes.

Law #1: When something gets cheap to make, its price falls

This is the oldest rule in economics, and AI doesn't repeal it. If AI makes a blog post nearly free to produce, then blog posts become nearly free to buy. Your clients aren't dumb. They have ChatGPT too.

Upwork's 2026 data shows this happening in real time, and it's the single most useful statistic in this entire post for anyone thinking of going solo:

Read those first two lines side by side. The same technology is making one kind of freelancer richer and another kind poorer. The difference isn't whether you use AI. It's whether AI makes you a cheaper producer of a commodity, or a more capable solver of a hard problem.

If your business is "I use AI to make social posts," you are in a race to the bottom against everyone else with the same tools, including your clients. If your business is "I use AI to rebuild your sales follow-up so you stop losing $40,000 a month in dropped leads," you're selling something scarce.

Law #2: The bottleneck moves; it doesn't disappear

Remember the Hands Problem from Part 1? AI largely solves it. But every system has some bottleneck. If it isn't hands anymore, what is it?

For a one-person AI agency, the new bottleneck is almost always one of these:

  1. Your attention. Every client, every automation, every agent needs someone watching it. Software can run 24/7; your judgment can't. A solo operator with 15 automated client systems has 15 things that can break while they're hiking.
  2. Trust. A client handing their customer conversations, data, or brand to a one-person shop is taking a risk. Why should they trust you? That trust is built slowly, through track record, referrals, and reputation, and AI can't speed it up much.
  3. Distribution. Who knows you exist? Look at our scoreboard again: Levels has a huge audience. Shlomo built in public on LinkedIn. Ottley and Saraev are YouTubers with hundreds of thousands of subscribers. AI made building nearly free. It did nothing to make being found free. If anything, by flooding the world with AI-generated stuff, it made attention scarcer.
  4. Accountability. When something goes wrong (and in an automated system, it will), someone has to fix it and own it. In the Medvi story, the chatbot inventing prices was a judgment problem, not a technology problem.

Notice that all four of these are human things. That's not a coincidence. When software takes over the production layer, the value concentrates in whatever software can't do.

Law #3: Recurring beats one-off

The AI agency world has learned this lesson the hard way. Building someone a chatbot for $3,000 is a one-time payday, and then you need a new client. But AI systems break, drift, need updating when the underlying models change, and need someone to watch the outputs. That's not a bug for an agency; it's a business model. The healthiest AI agencies tend to sell ongoing management (a monthly retainer to run, monitor, and improve the system) rather than one-off builds.

Putting it together: the new shape

So here's what the one-person AI agency pyramid actually looks like, once you account for all this:

The shape, drawn plainly: you at the top, carrying judgment, trust, sales and accountability. Beneath you, AI agents and automations produce the first 80 percent. Beneath them sit rented humans and platforms: contractors, partners, specialist firms and SaaS tools.

It's not one person replacing a pyramid. It's one person standing at the top of a pyramid made mostly of other people's software and other people's staff, holding the client relationship and the judgment. That's still a genuinely new and exciting thing. It's just not magic.

Part 8: What a Sane One-Person AI Agency Looks Like

Alright. Enough deflating. Let's build something.

If you took everything above seriously and still wanted to start a one-person AI agency, here's what the evidence suggests a sane version looks like. Think of these as principles, not a get-rich formula.

1. Start from a problem you already understand

The best solo AI businesses are almost never started by "AI people." They're started by people who deeply understand some specific, unglamorous corner of the world (dental clinic scheduling, freight broker paperwork, insurance claims intake, real estate lead follow-up) and then notice that AI can now fix a painful part of it.

This is the cheapest competitive advantage that exists. Anyone can learn n8n in a month. Not everyone has spent five years watching exactly how a property management office loses tenants.

2. Sell the outcome, never the AI

Remember Saraev's line: businesses pay for AI that makes or saves money. Nobody wakes up wanting "an AI agent." They want fewer missed calls, faster invoices, more booked appointments. If your pitch leads with the technology, you're selling a toy. If it leads with a measurable result, you're selling a business improvement, and you can price it against the value of that result instead of against your hours.

3. Pick work that's on the right side of the Upwork line

Ask honestly: "Is AI making what I sell cheaper for everyone, or is it making me able to solve problems that used to be out of reach?" Choose the second. Complex, high-context, high-stakes work is where solo operators are seeing earnings rise. Generic content production is where they're seeing earnings fall.

4. Keep a human in the loop, on purpose

The cautionary tales in this post (a chatbot inventing prices, AI support quality falling, ads that misled) share a root cause: removing human review in the name of speed or cost. Design your systems so that anything customer-facing, legal, financial, or reputational gets a human check, at least until it's earned trust. That human is probably you. Price that time in.

5. Productize what you can

The trap of any agency is that every client is a custom snowflake, which eats your attention. The escape is to standardize: the same core system, deployed for many clients in the same niche, with small adjustments. That's how a single person can serve more clients without the attention bottleneck crushing them. (Push this far enough and you drift from Creature B toward Creature C, an actual product. Many successful agencies end up there.)

6. Sell maintenance, not just builds

Monthly retainers: predictable income for you, and systems that don't quietly rot for the client.

7. Build your distribution from day one

Every success story in this post had an audience, a network, or a public build log. If you're not going to be a YouTuber, at least be findable: write about the specific problems you solve, in the specific places your specific clients hang out. One deep, credible piece about how you cut a clinic's no-show rate is worth more than a hundred generic "AI will change everything" posts.

8. Know the regulatory edges

Medvi's troubles are a reminder that some industries come with rules about who can say and do what: healthcare, finance, legal, anything involving personal data. AI makes it very easy to move fast into those industries. It does not make you exempt from their rules. If you serve a regulated industry, learn the regulations, or partner with someone who knows them.

A day in the life (an illustration, not a promise)

To make this concrete, here's a made-up but realistic sketch of a solo AI agency owner who serves physiotherapy clinics:

8:00 Check the overnight dashboard: AI handled 140 booking inquiries across nine clinics. Three got flagged as "unsure." Review them; one is a patient asking about a medical symptom, which the system correctly escalated to the clinic's front desk.

9:00 Weekly call with a clinic owner. Show her that no-shows dropped since the automated reminder-and-rebook system went live. She asks whether it could also chase unpaid invoices. That's next month's upsell.

11:00 A model update changed how the AI phrases cancellation messages. Two clinics' messages now sound oddly formal. Fix the prompt, test it, roll it out.

13:00 Write a short post for a physiotherapy business forum about what actually reduces no-shows (with real, anonymized numbers from clients who agreed to share).

15:00 Discovery call with a new clinic. Spend most of it asking about how their front desk works today. Mention AI maybe twice.

16:30 Use a coding agent to build the invoice-chasing feature. Review every line it touches the payment system.

Notice what the human is doing all day: judging, explaining, fixing, selling, and writing. The AI is doing the volume. That's the real job.

Who this is (and isn't) for

This path tends to suit people who already have deep knowledge of an industry, are comfortable selling, enjoy talking to clients, and like tinkering with systems. It tends to frustrate people who were hoping AI would let them avoid sales, avoid clients, or avoid learning a domain. AI removed the need to manage juniors. It didn't remove the need to be good at business.

Part 9: So, Will There Be a One-Person Billion-Dollar Company?

Back to that betting pool.

As of this writing, the honest answer is: nobody has clearly done it yet. Medvi came closest on revenue, but it had more than one person, leaned heavily on partners' staff, and came with regulatory baggage. Base44 had a team and sold for $80 million, not a billion. The "one-person unicorn" is still, as TechCrunch put it in 2025, closer to a mythical creature than a documented one.

Will it happen eventually? Probably, in some form. If you squint and allow contractors, partner companies, and a sibling, it arguably already has. Billion-dollar valuations are also partly a function of hype: when investors are excited about something, valuations can get ahead of reality. Bloomberg columnist Dave Lee made roughly this point in April 2026, observing that shrinking unicorn headcounts reflect both real AI efficiency and investors rushing to get into promising AI companies earlier.

But here's the thing I'd like you to take away: the billion-dollar question is the wrong question.

It's a fun bet for tech CEOs. It's irrelevant for almost everyone else. The far more interesting change is happening in the middle of the distribution. A decade ago, a one-person business earning, say, a few hundred thousand dollars a year with healthy margins was rare and usually required unusual skill or luck. Today, with AI handling much of the production work, that outcome is becoming reachable for a lot more people who have real expertise in a real niche.

Not guaranteed. Not easy. Not passive. And definitely not "no experience required." But reachable.

That's the actual revolution. Not one person making a billion dollars. Many people building a durable, well-paying business on their own terms, with judgment as their product and AI as their staff.

The Short Version

If you skimmed (no judgment):

  1. An agency sells judgment plus hands. AI is now very good at the hands part.
  2. Real people are making real money solo with AI, but almost none are truly alone, and many of the numbers are self-reported.
  3. When AI makes something cheap to produce, its price falls: generic AI work is getting less lucrative, complex AI-augmented work more so.
  4. The bottleneck has moved from hands to attention, trust, distribution, and accountability, all of which are human.
  5. The sane version: deep niche knowledge, sell outcomes, keep humans in the loop, productize, sell maintenance, build distribution early.

Go build something real. And when someone shows you a screenshot, ask who's counting.

Why this belongs at Starvard

Starvard is a free university, so this question belongs here in its practical form: not whether a company of robots sounds exciting, but what one person can learn, build and responsibly sell next. Start with AI Agents: Your First $10,000 for the agent layer this post keeps coming back to. Add Freelancing 101 for the client-finding craft that decides whether the agency eats, and use Startups 101 to pressure-test the offer before you build a pyramid of one on top of it. All three are free, like everything at Starvard.

Watch

Sources

Tagged: Technology · Business · AI Agents · AI Entrepreneurship · One-Person Business · Startups

Back to the blog

A note on accuracy: Much of what you read at Starvard was created with help from AI agents, working alongside people. We check what we can, but we cannot guarantee that everything here is true, current, or complete. If something really matters, your health, your money, a grade, a job, please verify it with a trusted source before you rely on it.