Loading page…
General-purpose AI agent: $0 to $36M ARR in 45 days
20 people
Pushed the agent experience for office work — AI slides, research, short video — to the global front rank. The window is still open, but the game is funding + speed.
Fewer bars = easier, cheaper, or faster for an AI-assisted solo builder. Editorial judgments based on the case details.
In February 2025, MainFunc (Jing Kun's team) announced a $100M raise at a $530M valuation — behind it, annual recurring revenue that rocketed from $0 to $36M in 45 days, with a company of just 20 people. The product, Genspark, plugs into the world's strongest models plus a wide toolset, and has pushed its general-purpose agent experience for high-frequency office scenarios — AI slides, AI research, AI short video — to the global front rank. Its shipping cadence is "insanely fast", and it recently launched an MCP-enabled AI browser.
A general-purpose agent cannot explain itself, so its first users come from watching it finish something. That is why the three entry points chosen here — AI slide decks, AI research, AI short video — all produce an artifact somebody can hold up. A deck built from one sentence is a screenshot and a link; a research report is a document people forward inside their own company. The first cohort in this category is never the enterprise buyer. It is the individual knowledge worker currently losing two hours to a deck nobody will read closely, and they arrive through a free allowance large enough to finish one real task end to end rather than a demo that stops before the interesting part. Release speed is itself acquisition at this stage: a team of 20 shipping visibly every week manufactures a reason to be mentioned again, and the MCP-enabled browser that followed is one more. Zero to $36M in annual recurring revenue inside 45 days is what happens when the product and the advertisement are the same object.
The first dollar in an agent product is bought the moment someone runs out of credits in the middle of something that matters. That is the entire conversion event, which is why the free allowance has to be sized to let one genuine job complete — a person who has already seen a finished deck will pay for the next one, while a person cut off at the outline will not. Pricing here is a monthly subscription around $20 for individuals with a heavier tier above it, because the underlying cost is truly variable: every run burns frontier-model tokens plus whatever tools it calls, so cheap unlimited plans do not survive contact with power users. $36M of ARR across 20 people is roughly $1.8M per head, a ratio that only holds when the purchase is self-serve and delivery is fully automated. There is no implementation team, no onboarding call and no contract in that motion, which is also the reason revenue could move as fast as it did.
The agent application layer is confirmed as 2026's biggest wave — your way to ride it is selling shovels and building plugins.
Paste into Claude Code / Codex and get a working version of this product end-to-end.
Two production-grade prompts per idea — a full build spec and an SEO growth plan — plus 10 tailored build specs a week. Members only.
Join the Unicorn Club — $5/moCancel anytime, in one clickEvery claim traces back to a listed source — open any of them and check the number yourself.
Data credibility: Public reporting / video roundup. All revenue figures are founder-reported or estimated by the video creator, unaudited.
Three things compound for a company at this layer, and none is a conventional marketing loop. The first is model progress: the product is a layer over the strongest available models plus a wide tool set, so every frontier release improves it without a line of new code, a rare position in which someone else's research budget becomes your roadmap. The second is integrations. Each tool the agent can call widens the set of jobs it can finish, and the MCP-enabled browser shipped after launch is the clearest case — it turns the open web into another tool rather than another feature. The third is the artifact loop, where finished decks and reports circulate inside companies and recruit the colleagues who see them. Underneath all three sits shipping speed, which is the real defence available here. Twenty people releasing faster than anyone else means a copy of the product launches against a version that has already been replaced, and that gap is the only durable advantage on offer.
For an agent company the quiet stretch is a reliability problem rather than a marketing one, and the number worth tracking is the share of real tasks the agent finishes without a human rescuing it. Demo success rates mean nothing; the honest measure is a hundred tasks a customer actually cares about, counted end to end. Until that rate crosses the point where people stop double-checking the output, usage stays flat no matter how good the launch video looked. How this particular company bought its way through that period is visible on the balance sheet: $100M raised at a $530M valuation, which converts waiting into a funded sprint instead of an endurance test. That is precisely why a small team should not run at this head-on. The durable opening sits at the edges — plugins, connectors and narrow tools for the agent era — where the reliability bar covers one task instead of all of them, and where two people can clear it in weeks rather than needing a nine-figure war chest.
The agent application layer is confirmed as the biggest wave of 2026, and the window is still open — but at this level it's a war of funding + iteration speed, so solo players should not charge in head-on. The right way for you to ride this wave: build "accessory" businesses on the edge of these ecosystems (see the OpenClaw shovel-selling page), and make plugins and tools for the agent era.
That was one of the 50 breakdowns open to everyone. The Unicorn Club opens the other 1,285 at the same depth — plus pain-point filters and 10 build specs a week.
Join the Unicorn Club — $5/moCancel anytime, in one click