Sam Altman keeps predicting the one-person, billion-dollar startup.
Easy to roll your eyes at. Until someone who co-founded a company worth around $175 billion shares his screen and shows you what he does with his evenings.
Furqan Rydhan was on the founding team of AppLovin. He also co-founded Founders, Inc, the San Francisco lab where a lot of weird ambitious projects get built. And in January he sat down with Greg Isenberg and demoed something he had never shown publicly: Nebula, his platform for running a one person business entirely on AI agents.
Not a pitch deck. A working product, breaking live on camera, retrying its own failures until the thing shipped.
I watched the whole episode twice. Here's the full breakdown, with the numbers.
The guy who built a $175B company tinkers at night
Greg has known Furqan for over 10 years. His words: "when he speaks, I listen."
Furqan's own description of himself is less dramatic. He's a tinkerer. Nights and weekends, playing with new technology. The last few years that meant AI systems, LLMs and agents, and a few months ago the tinkering "became more obsession than just tinkering."
The origin story is the same one every busy founder lives. Too much Slack and too many people to keep up with, so you delegate and let things go, even the things you actually wanted your hands in.
His fix: build agents that do the work, with his opinions baked in. Where should an agent go full send, and where should it stop and ask him first.
Then he noticed he was telling friends about these agents constantly. That's usually the signal. He turned it into a product, with himself as customer number one.
What Nebula is: Claude Code for everything else
Nebula looks like Slack, down to the channels and threads.
Except every "person" in there is an agent.
Furqan's framing is the best one-liner in the episode. He's an engineer, so he already has Claude Code, Cursor and Codex making his coding dramatically more powerful. Nebula is "like Claude Code for everything else." The non-engineering work: slides, blogs, lead gen, analytics, email.
Under the hood each agent can:
- search the web and browse, with all the knowledge of the internet
- write code AND execute it (Python, API requests, integrations)
- use its own file system in the cloud, no computer needed
- connect to the services where work actually lives: Google, GitHub, Slack, Linear, Notion, Ghost
- run on schedules it writes for itself
That last one matters more than it sounds.
The live demo: a slide deck that builds itself
Furqan types one message into a channel: connect to Google Slides, create a deck, make a three-slide outline about Nebula for a founder audience.
The agent creates its own to-do list, calls the Google API, and the slides appear. He never touches the presentation.
Then he pushes it: add a fourth slide with a generated image that fills the whole slide.
And here's the honest part, the part that made me trust the rest of the demo. It failed. Twice. The image didn't size right, the upload complained. The agent read its own errors, rewrote its own Python, and tried a third time until something landed. Furqan just narrated: "could you just tell it to keep trying until it works? It kind of can."
That's the actual unlock with agent systems: they retry without you.
Then the kicker. He types: add a new slide every day, get this deck to 15 slides within a week.
Nebula extracts the context, writes itself a recipe, writes its own cron trigger, and schedules the job for 9am daily. Hands off. If you've ever duct-taped that together in Zapier or n8n (Furqan runs self-hosted n8n instances himself and still found it tedious), you know how much setup just evaporated into one sentence.
The blog that runs itself: 3 posts a day, 100 hits a day, 30 minutes of setup
This is the part I kept rewinding, because it's a real business in miniature, with real numbers.
Furqan asked himself: what's the smallest company an agent system could run end to end? He landed on a blog.
His setup:
- content direction and sources (X trends, news, research)
- a visual style and an editing style
- a connection to Ghost for publishing
- a daily schedule, plus Search Console and PostHog data to optimize keywords against
The result at recording time: 15 straight days of 3 posts per day, fully autonomous, around 100 hits a day. Total human setup time: under 30 minutes.
During the episode he spun up a second one live. VR-themed blog, research the top VR influencers, produce a post with an image. By the time they finished talking about something else, the post was published and the trigger was set: 6am, 2pm, 10pm, every day.
Full disclosure: I'm biased here, because this is exactly how the blog you're reading works. Three articles a day, written and published on a schedule by agents I direct. Furqan spent 30 minutes and got to 100 hits a day. The difference between his demo blog and a real asset is everything you layer on top: voice, internal links, original numbers, distribution. I wrote up the systems side of that in my breakdown of running an AI agent workforce.
100 hits a day on 30 minutes of setup is a proof, not a business. The business shows up when you treat the agent output as a floor and push the quality bar up every week.
The math: a $1,000 a month AI team
Furqan's claim, straight from the episode: spend $1,000 a month on an AI team, point it in the right direction, and it will produce $5,000 to $10,000 a month of value for you.
"Point it in the right direction" is carrying a lot of weight in that sentence. His whole thesis is that direction is the part that stays human. The research, the writing, the image generation, the scheduling: that's what goes away.
Greg pushed this one step further, and it's the step most people listening will actually take. Agencies.
Content agencies churn out blog posts for a monthly fee. Analytics agencies set up PostHog and send reports. Both are services an agent system can now deliver with, in Greg's estimate, a 20th of the humans. Same client, same deliverable, a fraction of the payroll.
Furqan didn't push back. He's already running Nebula on Nebula: changelogs, landing page experiments with three variants a day, an analysis agent reporting results every morning. I profiled a founder doing a version of this solo in the solo AI agent business story, and the shape is identical: one operator directing a stack of agents that produce client-grade output.
"Won't this get commoditized?"
Furqan answered the comment-section objection before anyone could type it.
Yes, everyone gets access. His analogy: there was a time when just having a .com was a competitive advantage. Then having a blog was. Each wave floods, the basic version becomes worthless, and the edge moves up a level.
His answer to the flood is very concrete: build a critic. A blog critic agent that loops against every post and won't let anything publish below a 9 out of 10 on metrics you define. Your content has to be superior, your delivery has to be superior, and you build agents to enforce that too.
Greg's timeline on the window: one-person blogs and newsletters are a huge opportunity for the next 12 to 18 months. After that, the basic version is table stakes.
So the moat isn't the tool, because everyone will have the tool. The moat is your taste and how fast you iterate the system. Same as it ever was, just with more use per decision.
How to steal this playbook this week
You don't need Nebula specifically (it's live, and Furqan hasn't even hooked up billing yet, his literal words were "go burn tokens"). You need the pattern.
→ Pick the smallest company you can fully describe. A niche blog. A daily newsletter. A weekly report for one industry.
→ Write the direction doc like you're briefing a sharp employee. Sources, voice, visual style, what good looks like, where the agent must stop and ask you.
→ Put it on a schedule. Daily beats perfect. Furqan's demo deck grows 2 slides a day whether he's watching or not.
→ Close the loop with data. Search Console and analytics feed back into next week's direction. That loop is the product.
→ Add a critic before you add volume. Scaling output without a quality gate just builds a content farm nobody reads.
The founders I interview keep converging on the same shape: one person, strong opinions, agents for everything repeatable. Furqan just built the cleanest version of it I've seen demoed.
FAQ
What is Nebula, the AI agent platform?
Nebula is an agent platform built by Furqan Rydhan, co-founder of AppLovin and Founders, Inc. It looks like Slack, but every channel contains an AI agent that can search the web, write and execute code, keep its own file system, connect to services like Google, GitHub and Ghost, and schedule its own recurring jobs. Furqan describes it as "Claude Code for everything else."
Can AI agents really run a business by themselves?
They can run narrow ones. Furqan's test blog published 3 posts a day for 15 straight days with zero human involvement after a 30-minute setup, reaching about 100 hits a day. What agents can't do is pick the direction, define quality, or decide what's worth building. The humans who do that part well are the ones who'll capture the value.
How much does a one person business powered by AI agents cost to run?
Furqan's working number is about $1,000 a month for an AI team, which he believes produces $5,000 to $10,000 a month of value when pointed in the right direction. You can start far cheaper: a single scheduled agent plus a publishing integration costs a fraction of that while you validate the niche.
What's the window of opportunity for one-person AI businesses?
Greg Isenberg's estimate in the episode: 12 to 18 months for blogs and newsletters before basic agent-generated content gets commoditized. After that, the edge shifts to superior quality, niche expertise and distribution. Starting early matters less than building the quality loop (critic agents, data feedback) that keeps you above the flood.
Furqan built his system because he had power in his hands and wanted everyone to get it. Fair. But a system without direction is just burning tokens.
If you're a founder figuring out what to point your agents at, that's exactly what I dig into every week with people who've already done it. Bootstrapped founders, real revenue, playbooks you can copy.
