In November 2025, a Dutch developer named Pieter Levels tweeted a revenue dashboard for one of his apps. PhotoAI, an AI photo generator, was making about $138,000 a month. He runs it alone. No team, no office, no investors.

That's a real number from a real person who posts his stripe revenue in public. It's also the exception, not the rule. Most AI projects don't end like that. So this is an honest look at both sides: the people winning with AI, the numbers behind them, and the well-documented reason the majority still fail.

The solo founders with public receipts

The most credible AI success stories are the ones where the founder shows the money.

Pieter Levels (@levelsio) built PhotoAI from nothing. By his own public figures, it went from $5.4K in month one to over $100K a month within 18 months, and around $138K/month by late 2025 — roughly 70% of his income. His wider portfolio of small products runs at about $3M a year, still solo, and he builds with AI coding tools rather than a payroll (Indie Hackers case study).

Marc Lou did something braver than most: he published a post titled "I made $1,032,000 in 2025," broken down across roughly 15 income streams, with zero employees (his newsletter). His products ShipFast and CodeFast each make around $20K a month, and more than 7,200 developers have bought ShipFast. He documents the wins and the flops openly.

Two things stand out about both of them. They ship products fast, and they sell something people pay for. AI is the tool that lets one person do the work of a small team — not a magic revenue button.

The small-business numbers (with sources)

Away from the headline founders, the aggregate data on ordinary businesses is genuinely positive — when you cite it carefully.

  • The average small-business worker saves about 5.6 hours a week using AI tools; owners and managers save 7+ (Business.com research, 2026).
  • 91% of small businesses using AI report a revenue increase (Salesforce).
  • McKinsey puts the average return at about 3.7x on AI tool spend for small businesses.
  • Around 66% of small firms say AI saves them $500–$2,000 a month.

A typical documented setup looks unglamorous: a chatbot for customer questions, ChatGPT for content, and an automation tool for email. One small e-commerce owner reported saving three hours a day on inquiries and about $1,200 a month versus hiring help — on tools costing roughly $70 a month.

The honest part: most AI projects fail

Here's the number the hype posts leave out. In 2025, MIT's Media Lab published "The GenAI Divide," a study of 300 public AI deployments plus interviews and surveys. It found that 95% of enterprise generative-AI pilots delivered no measurable profit-and-loss impact (Fortune coverage).

The reason matters more than the number. MIT didn't blame the models. It blamed the "learning gap" — companies buying AI, then failing to wire it into real workflows, teams, and habits. The same study found that buying tools from specialist vendors and integrating them succeeded about 67% of the time, while grand internal build projects succeeded roughly a third as often. The biggest wins were boring: back-office automation, cutting outside agency and outsourcing costs.

Forbes made a sharper version of the point after surveying tens of thousands of small businesses: plenty say AI is working, but the measurable data is thinner than the enthusiasm.

What the winners do that the 95% don't

Put the two sides next to each other and a pattern shows up. The people winning with AI aren't running pilots. They're doing three things:

  1. They point AI at work that ships or sells. Levels builds products and charges for them. The e-commerce owner automates the exact task that was eating her day. Nobody's running a "GenAI initiative" — they're removing a specific cost.
  2. They integrate instead of admiring. MIT's finding is the whole game: value comes from wiring the tool into a real workflow, not from buying access and hoping.
  3. They keep the loop short. One person, fast iterations, honest measurement. If it doesn't save hours or make sales, they drop it.

We've felt this first-hand. Running a small software studio, the AI tools that stuck were the few we used on work that had to ship — writing, research, and code that ends up in front of customers. We wrote an honest log of the AI skills that actually earned their keep, and the same pattern held: the wins came from shipping, not experimenting. It's also how we built and sell our own Maya animation toolkit as a one-person product.

Takeaway

AI success stories are real. The numbers from Levels and Marc Lou are public, and the small-business gains are measurable. But the honest version includes the 95% who saw nothing — and the difference between the two groups is rarely the model. It's whether AI got attached to work that actually ships and sells. Start there, measure honestly, and drop what doesn't earn its keep.