EasyHive

THE AUDIT

WHO IS ACTUALLY GETTING RICH FROM AI

Show me the receipts

We took the most-repeated "I got rich with AI" stories and chased every number back to a primary source. Roughly half did not survive. This is what held up, what collapsed, and the habit that separates the people compounding money with AI from the people posting about it.

01Why this document exists

Every AI success story this year has the same shape. A non-coder. A weekend. A number with a lot of zeros. A tool you should go buy.

So we ran the boring experiment nobody runs and traced each claim back to a primary source — the founder's own dated post, the acquirer's press release, the government record. Not a blog citing a blog. Roughly half led to nothing at all: numbers appearing in forty articles and zero sources, and savings figures credited to a startup that were actually produced by a government program predating it by three years.

The uncomfortable part: the shortcut that makes those articles wrong is the shortcut that makes AI output wrong. A model hands you a confident, well-formatted, unsourced answer, and it feels exactly like the truth. If you cannot audit a claim in an article, you cannot audit one in your own business — and that is the real subject of this PDF.

02Four viral claims, audited

These four are in nearly every AI-opportunity post right now. Here is what the sources actually say.

The claim as it circulates	Verdict	What the primary source actually says
"Anthropic raised $65B at a $965B valuation, passing OpenAI for the first time."	DISTORTED	The raise is real and first-party: $65B at $965B post-money (May 2026), against OpenAI's $122B at $852B. But "for the first time" is an invention — it is in neither announcement, and Anthropic had passed OpenAI earlier.
"Run rate hit $47B, up from $1B at the end of 2024."	HALF TRUE	$47B is solid — stated verbatim by Anthropic. The $1B baseline appears in no first-party document, only aggregators. The growth is extraordinary; the specific multiple quoted is unsourced.
"More than half of YC's newest batch builds with Claude; a year earlier OpenAI had 90%."	THIN	Traces to one offhand remark by a YC partner on a podcast. Not a survey. The 90% comparison is from early 2024 — two years earlier, not one. Directionally interesting, statistically meaningless.
"Non-coders won a state contract at 10% of consulting quotes and now save citizens $1B+/year."	FALSE	Company and contract are real (see page 3). But the $1B/year savings belongs to a state deregulation program launched in 2022, years before the startup existed. The "10%" is a founder's characterization of one bid.

Notice the pattern. None is a pure fabrication. Each is a real fact with a small, load-bearing embellishment welded on — precisely the part that makes it shareable. That is also the exact failure mode of an unverified AI answer.

EasyHive · easyhive.pro · Audited August 2026 · page 1 of 5

03What survived the audit

Every case below traces to a primary source. Where a number is self-reported rather than audited, we say so. Where the popular version is wrong, we correct it — the corrected version is almost always still remarkable.

Base44 — Maor Shlomo $80M cash · 6 months

Israel · AI app builder · acquired by Wix, June 2025 · 8 employees · built on Claude via AWS Bedrock

Six months old, eight people, bootstrapped with no VC. Acquired for $80M cash plus roughly $90M in earnout through 2029 — the earnout exceeds the headline. It cleared $189,000 profit in May 2025 alone pre-sale, and passed ~$100M ARR under Wix within nine months.

What the viral version gets wrong: it is told as "solo founder sells for $80M." There were 8 employees, and Shlomo was already a professional programmer with a prior VC-backed startup — not a beginner with a prompt. The real story is still the strongest case here.

SOURCE: Wix press room (acquirer-confirmed) · TechCrunch, 2025-06-18 · traceable in Wix earnings

Bolt.new — Eric Simons $0.7M → $20M ARR

USA · browser-based AI app builder · <40 employees, <10 in go-to-market

The company was weeks from shutting down at $0.7M ARR. It launched on a single tweet in October 2024, hit $4M ARR in four weeks and $20M within three months. Simons credits the model shift by name: Claude 3.5 Sonnet's agentic coding was "leaps-and-bounds better."

Read it right: the tool did not save a company with no product. It removed the last constraint from a team that had built the hard part and could not quite get it to work.

SOURCE: Growth Unhinged, 2025-07-09 · Lenny's Newsletter founder interview, 2025-03-13

Pieter Levels — PhotoAI $105K/mo rev · $80K/mo profit

Netherlands · solo, zero employees · self-reported, dated March 2026

One person, no staff, and a single 40,870-line PHP file producing $105K/month revenue at $80K/month profit. He credits Claude Code for production work. His flight-sim experiment fly.pieter.com hit $1M ARR in 17 days from a ~3-hour prototype, with no game-dev background.

Two caveats: these are self-reported, not audited — and the flight sim was a viral spike that decayed to roughly $15K/month. Anyone quoting "$1M in 17 days" without the decay is selling you something.

SOURCE: levels.io first-party posts, dated 2026-03-06 and 2025 · note: his public /open dashboard now 404s

Lovable — Anton Osika $100M ARR · 8 months

Sweden · AI app builder · 45 full-time employees at the milestone

Fastest documented climb to $100M ARR: eight months, 45 people, passing $10M ARR in 60 days with a team of 15. Series B of $330M at a $6.6B valuation. Non-US — this opportunity is not geographically gated.

Disclosed, not audited: TechCrunch flags the ARR as company-stated, and 2026 headcount figures contradict each other across aggregators. Treat the trajectory as real and the precision as approximate.

SOURCE: TechCrunch, 2025-07-23 and 2025-12-18

EasyHive · easyhive.pro · page 2 of 5

Vulcan Technologies — Tanner Jones $150K pilot → $10.9M seed

USA · regulatory AI for government · founder's last code was high-school JavaScript

The genuine non-coder case with the hype stripped off. Jones built the prototype inside the regular Claude app, copy-pasting scripts out of the chat — Claude Code did not exist yet. It won a $150,000 three-month pilot with Virginia's Office of Regulatory Management against established consultancies; the company later raised a $10.9M seed. Customers now include the US Dept. of Education.

What everyone repeating this gets wrong: the "$1.2B saved for Virginians" is not Vulcan's — it belongs to a statewide deregulation program begun in 2022, years before the company existed. A $150K pilot won by a non-coder against consulting firms is a great story. It does not need a stolen billion attached.

SOURCE: Reason, 2025-07-23 · PRNewswire seed announcement, Oct 2025 · narrative detail from Anthropic's own blog (vendor source — weight accordingly)

ElevenLabs $500M ARR · ~$943K per employee

USA/UK · AI audio · 530 employees · first-party, May 2026

Included for the org chart, which is the transferable part: 530 people run as roughly 20 micro-teams of 5–10 with no middle management. AI leverage shows up as small autonomous units, not headcount growth. Compare Midjourney — revenue "significantly surpassing $200M" with ~40 people and zero outside funding since 2021.

SOURCE: elevenlabs.io first-party, 2026-05-05 · Midjourney figure via The Information, 2026-03-29 (the widely-quoted $500M Midjourney ARR is an analyst estimate — do not repeat it)

04What died — and why it matters more

These are quoted with total confidence everywhere. None traces to a primary source.

Claim in circulation	What we found when we chased it
"$300K/month from an AI headshot app" Attributed to a named, real founder	Identical figures across dozens of SEO blogs, zero traceable origin, conflicting dates, ~2 years stale. The founder's only confirmed first-party post contains no revenue figure. Textbook citation laundering.
"$1M ARR built entirely with Claude Code" A named startup	Traces to one AI vendor's marketing blog. No named founder, no company confirmation, and the real company at that domain is an unrelated product.
"Non-technical founder at $30K MRR" Forum post, widely screenshotted	Founder unnamed, figure self-reported and unverifiable — and he sells AI coding workshops. That is the conflict of interest you should assume by default.
"$8,200/month freelancing with AI automations" Various, always anonymous	Every instance traced to affiliate blogs or anonymous forum posts. Nothing verifiable exists in this category.
The solo founder who made $1,032,000 in 2025 Marc Lou — fully itemized and real	Revenue real, first-party, itemized. We cut him anyway: his post credits no AI tools whatsoever, and 2025 was a 20% decrease. A great solo-founder story that would have been a fabricated AI story.

The pattern: the most-shared numbers were the least verifiable, and the verifiable ones were less dramatic than their viral versions. Assume any figure living only inside content that sells a tool or a course is fabricated until a primary source proves otherwise. That one filter catches every dead claim above.

EasyHive · easyhive.pro · page 3 of 5

05What the survivors actually have in common

Five verified cases. The pattern is not the one being sold to you.

1
None started from zero knowledge of their marketShlomo was a programmer. Simons had a product that did not work well enough. Jones knew regulatory law cold and could not code. AI removed the execution constraint, never the domain-knowledge constraint.
2
They shipped something ugly and real, fastA 40,870-line PHP file. A prototype pasted out of a chat window. Not clean architecture — things that existed and could take money while better ideas sat in someone's notes app.
3
The team stayed small on purpose8 people at acquisition. 45 at $100M ARR. Micro-teams of 5–10. Leverage shows up as revenue per person, not headcount — that is the number to watch in your own business.
4
Distribution was already handledLevels had an audience. Bolt's launch tweet worked because someone was listening. Vulcan knew which office had the problem. AI collapsed the cost of building and did nothing to the cost of being noticed — now the real bottleneck.

The honest opportunity is not "non-coder builds an app and gets rich." It is: if you already know something valuable, the gap between knowing it and shipping it has collapsed to roughly the cost of a subscription. Smaller than the hype, far more useful — because it is true and available to you this week.

EasyHive · easyhive.pro · page 4 of 5

06Where Claude Code fits

Claude Code is Claude in your terminal, on your real files, with permission to act instead of advise. Four properties do the work, each mapping to a case above.

Property	What it changes
It executes	Describe the outcome in plain English; it writes, runs, tests, and fixes. Jones built a government-grade prototype by pasting out of a chat window — that manual loop is now gone.
It is agentic	Pursues a goal across many steps instead of answering one question. The property Simons named when Bolt went from $0.7M to $20M ARR.
It runs in parallel	Multiple agents on separate tasks at once — how a team of 8 produces what looks like a hundred engineers.
It remembers	Persistent context on your business, customers, and past decisions. The compounding one: the gap widens the longer you use it, and you cannot catch up on it later.

What it is not: a replacement for judgment about what is worth building. Every dead claim on page 3 came from someone who could produce output but not evaluate it. Output is now cheap; evaluation is the scarce skill, and it is what you still have to bring.

07Start this week — four steps

1
Get a paid plan and pick ONE taskThe entry plan is $20/month. Take the single repetitive thing you avoid or pay someone to do. Not ten. One — and pick the one where you already know what "correct" looks like, because you must grade the output.
2
Brief it like a new hireWhat the task is, how you do it today, what a good result looks like, and what would make it wrong. That last one matters most and almost nobody writes it down.
3
Make it argue before it buildsThese models lean agreeable and will build a bad idea enthusiastically. Force the other side: "Before writing anything: make the strongest case against this approach, list what must be true for it to work, and say what you'd build instead."
4
Audit the output like we audited these claimsWhen it says done, do not believe it. Run it on a real example and show the output. Then ask it to name the weakest claim in its own work and prove that one first. Same discipline that killed half the stories on page 3, pointed at your business.

THE ONE HABIT THAT SEPARATES THE TWO GROUPS

Everyone gets the same models. The people compounding real money with AI are not prompting better — they are checking harder. Every verified case here has a founder who could tell whether the output was good; every dead claim came from someone who could not. Your first version will come back 60–70% right. That is a draft awaiting an editor, and the editor is the job now.

EasyHive · easyhive.pro · Every figure traced to a primary source, August 2026. Self-reported figures are labeled as such. Where we could not verify, we said so and cut it.

All trademarks belong to their owners. Easy Hive is not affiliated with or endorsed by them. This site is not part of Facebook, Instagram or Meta Platforms, Inc., and is not endorsed by Meta. Savings and results shown are examples, not guarantees; yours will vary. Educational content only, not legal, tax or financial advice. Full disclaimers.