Solo Founders Hitting $1M ARR With Mostly AI-Written Code

The "solo founder at $1M ARR with mostly AI-written code" story stopped being a thought experiment about 18 months ago. It is now a documented pattern, with public Stripe screenshots, monthly MRR posts on X, and Indie Hackers profiles that line up against the receipts. The question is no longer whether it is possible. The question is what those workflows actually look like in production — and what AI code still does not save you from.

This piece is about the revenue-proven side of the story. There is a separate piece on teen Replit Agent builders shipping their first apps with zero coding background; that one belongs in a different category. The founders below are technical or technical-adjacent, they were already shipping software before the current generation of AI tools existed, and they have publicly documented the leverage they got from Cursor, Claude, GPT-5, and the rest of the stack.

The five I am going to walk through are the ones whose revenue claims are verifiable against public statements — Stripe dashboards posted to X, Indie Hackers profiles with the green badge, ProductHunt comments where the founder names a number. There are dozens of others claiming similar figures with no receipts. I am not including them.

1. Pieter Levels — the canonical case

Pieter Levels, @levelsio on X, is the founder most people point at first when this topic comes up. He has been publicly running Nomads.com, RemoteOK, Photo AI, and InteriorAI for the better part of a decade, with revenue dashboards posted to a public stats page that has been live since 2014.

His combined ARR sits well north of $3M as of his most recent public updates, with Photo AI alone trending past $1M ARR inside its first year of launch — he posted the Stripe receipts on X in 2024. His bootstrappers podcast appearances and Lex Fridman conversation walk through the stack in detail.

The stack itself is the part that breaks people's brains. Levels ships single-file PHP — one index.php file, jQuery on the front end, MySQL behind it, no build step, no React, no TypeScript, no microservices. For most of 2024 and into 2025 he has been pair-coding with Cursor against Claude and GPT, posting his prompts and workflows publicly. The workflow he describes on the Lex Fridman conversation is roughly: open Cursor against the one file, describe the feature, accept the diff, deploy directly to production, debug in production if needed.

The wedge is that he was already a strong engineer who could read every line of AI-generated code before accepting it. The AI compressed the keystrokes between idea and shipped feature. It did not replace the judgment about what to ship.

2. Marc Lou — the boilerplate empire

Marc Lou, @marc_louvion, is the second canonical name. He runs ShipFast (Next.js boilerplate), BoilerCode, and IndiePage, with publicly-posted MRR running past $125K/month in mid-2025 — so $1.5M+ ARR by the time he posted the Stripe screenshot on X. He shares his monthly numbers in a public newsletter and on X.

His public workflow is heavier on shipped product, lighter on novel code. ShipFast's value proposition is that he has already written the auth, payments, emails, and SEO scaffolding that every indie founder rebuilds from scratch — so the boilerplate sells because the boilerplate worked for him first. He codes primarily in Cursor with Claude, and his blog posts describe a workflow where AI handles the 80% of code that is "boring CRUD and Tailwind styling" while he keeps tight hold of the architectural decisions.

The Marc Lou pattern is worth studying because it is not the "AI built everything" story. It is the "AI removed the parts of building software I am bad at or bored by, so I could ship five products instead of one" story. ShipFast launched in 2023, BoilerCode and IndiePage followed in 2024 and 2025. The portfolio is the leverage, not any single product.

3. Tony Dinh — multi-stack, monthly receipts

Tony Dinh, @tdinh_me, runs TypingMind (a chat front-end that hit serious revenue when ChatGPT pricing made BYO-API-key compelling), DevUtils (offline Mac developer toolbox), and Black Magic (X analytics). His monthly MRR posts put combined revenue past $80K/month — roughly $1M run-rate — and have done so consistently for over a year.

Dinh's stack is more diverse than Levels or Lou — Mac apps in Swift, web apps in Next.js, browser extensions in TypeScript. He has written publicly about leaning on Claude for code review and Cursor for the bulk of new feature writing, with a heavy emphasis on testing each diff in isolation before accepting it.

The reason Dinh belongs on the list is that he is the clearest counterexample to "AI coding only works for trivial CRUD apps." TypingMind handles streaming, custom models, plugins, and a payments flow. DevUtils is a native Mac app with file system access and a polished UI. None of those are boilerplate-shaped problems. He still ships them solo with AI assistance.

4. Danny Postma — the second-act founder

Danny Postma, @dannypostmaa, exited Headlime (AI copywriting, sold in 2022 for an undisclosed sum reportedly mid-seven-figures) and has since launched Pictorial.ai and several other AI-image tools. His public X updates put his current portfolio past $1M ARR.

What makes Postma worth including is the stack-shift. He has publicly documented moving from Cursor-only workflows to using Lovable for the front-end scaffold + Cursor for the production code path — the "AI builds the UI shell in 30 seconds, I take it into Cursor and harden it" pattern. Several other founders on the upper end of the indie scene now describe the same two-tool flow.

5. Riley Brown — the Cursor-native generation

Riley Brown, @rileybrown_ai, is the youngest of the bunch and the most aggressive about claiming AI-first workflows. He runs VibeCode and the Vibecode School content brand, with publicly-stated ARR past $1M.

The reason to put him on the list with a caveat — his MRR screenshots are less consistent than the four above and some of the early claims have been revised. But the workflow he documents on YouTube is the closest thing to a "Cursor-native solo founder" template that exists publicly, and his audience reach has made his methods the default mental model for the next cohort of indie builders. Whether his exact numbers hold up to a Stripe screenshot is less important than the fact that his playbook — voice-prompt into Cursor, accept, deploy, repeat — is the playbook a lot of $50K-200K ARR builders are now copying with their own receipts.

What the workflow actually looks like

Strip the X marketing veneer off all five and the production workflow is roughly identical:

• Prompt into the editor. Cursor is the dominant tool. Windsurf and Zed are growing. The prompt is usually a one-paragraph feature description plus a pointer to the relevant existing file.
• Read the diff. Every founder on this list still reads the proposed diff before accepting it. None of them are running "agent mode" against production unattended. The closest is Levels accepting most diffs without review on his single-file PHP app — but he is the original author of every line in that file and has the entire architecture in his head.
• Ship to staging or directly to production. Levels deploys to production. Lou ships to a staging environment. Dinh ships native apps to TestFlight first. The pattern is "fast feedback loop," not "skip QA."
• Debug with the AI. When something breaks — and it does, daily — the workflow is paste the error into Claude or GPT-5, get a hypothesis, test it, iterate. Cursor's built-in debug-this-error shortcut is the version everyone uses.
• Refactor in chunks. Once a feature stabilizes, the AI gets asked to extract helpers, rename functions, write tests. This is where AI is unambiguously better than human attention — it never gets bored renaming the 47th variable.

The 60-90% AI-written code figure quoted in headlines is roughly accurate by line count. By judgment count — which lines matter, which architecture survives — it is closer to 30-40%. The founders are still the ones deciding what gets built, what gets shipped, what gets rolled back.

What AI code does not save you from

The honest reframe is that AI coding tools removed the cost of writing code as the binding constraint on solo SaaS. The new binding constraint is everything else — taste, distribution, customer trust, operational discipline. The founders at $1M+ ARR are the ones who already had those before AI showed up. The AI made them faster, not different.

The wedge for solo founders

What this means, structurally, is that the gap between a solo founder and a 5-person team narrowed by something like an order of magnitude in the last 24 months. The solo founder bypasses the team-coordination cost — no standups, no Slack threads, no architectural arguments, no waiting for a code review — and ships in a tighter loop than any team possibly can.

But the solo founder only gets that leverage if they are already technical (or willing to become technical fast). Every name on the list above could code before AI tools existed. They got faster, not transformed. The "non-technical founder ships a SaaS" story is different — that one belongs in the teen Replit Agent builders piece and the broader AI agent retrospective write-up, where the tradeoffs are different.

For the technical solo founder, the playbook now is roughly:

• Pick a tight, defensible niche (Levels: digital nomads; Lou: indie hacker tooling; Dinh: developer tools; Postma: AI-generated images for marketers).
• Build a minimum-viable product in 2-4 weeks of Cursor-driven coding.
• Launch publicly with the Stripe metric visible.
• Iterate weekly based on user payments and refund requests, not feature requests.
• Compound through portfolio — once one product hits $10K MRR, build the second on top of the audience.

That is the pattern the public revenue receipts validate. The founders who try to skip the niche selection step or the public-launch step do not show up on the list.

What to do this week

If you are a technical-adjacent person sitting on a SaaS idea and watching this happen from the sidelines, the cost of finding out has never been lower. A Cursor Pro license is $20/month. A weekend of focused work will tell you whether the AI-coding workflow clicks for you. The Levels-style single-file approach is the lowest-friction starting point — pick one feature, build it as a static page with a Stripe button, ship it, see if anyone pays.

If you are non-technical, the path is different and slower, and the teen Replit Agent builders piece is the better entry point. The five founders profiled above are not the template for that path. They are the template for the technical-founder path, which is what this piece is actually about.

The "solo $1M ARR with mostly AI-written code" story is no longer aspirational. It is observed behavior. The question for everyone reading this is whether the constraint between you and a similar outcome is the code (it probably is not anymore) or the distribution, taste, and discipline that the founders above already had before AI showed up.

"Most of the work of running a $1M solo SaaS is not writing the code. AI made the code part easier. The other 70% is what still separates the founders with receipts from the founders without them."