The Idea:
David Im had a plan. A top Korean university, then a US PhD, then a life in America. He got into the university and joined a reinforcement learning lab. Then, three months before he was due to apply to grad school, his advisor pulled his research out from under him, unwilling to bet the lab’s output on someone already halfway out the door.
So Im moved into an AI hacker house in Korea and started shipping. A tutoring platform, a dating app, a concert ticketing tool. Roughly ten pivots a month for five months, with nothing to show for it. Housemates mocked it as a “thin startup.” By winter, he was a month from being kicked out.
Then Chase Huh got in touch. A self-taught coder who’d once made $2,200 from a Minecraft server in middle school, Huh had shipped three products of his own to zero users. Two serial pivoters, one rule: ship in three-day sprints and kill whatever doesn’t catch.
At 3am on New Year’s Day, they slipped into the hacker house basement to pitch a San Francisco accelerator with a product they’d faked in two hours.
This was the beginning of Sume.
The Execution:
- January 1, 2026: Five minutes into their Founders Inc. interview, partner Hubert Thieblot says he doesn’t like the product. They ditch it and pitch themselves instead, and land a spot in the Artifact batch. They have eight days to get to San Francisco, and Im sells all his Nvidia stock to pay for the trip.
- January 2026: Im’s acceptance post goes viral, so they ride it, shipping AI avatars of Steve Jobs and Roy Lee that dish out business advice. Built in two days, it pulls in 10,000 users.
- February 2026: They launch Clawra, an AI companion with its own face and personality built on OpenClaw. It hits 1M X impressions, 10,000 users and 1,000 GitHub stars in a single day.
- Early 2026: At a group fireside chat, Thieblot asks if anyone has bad news. Im admits Clawra got 10,000 users in three weeks but retention is poor. The chat turns into an investor meeting, and Founders Inc. writes a cheque the next day.
- Spring 2026: The wobble. After moving on from Clawra, they try a run of products including abgcmo. One tops $10,000 in revenue, another clears a couple of thousand a month, but none of them keep users.
- Summer 2026: They fence themselves into a single domain, AI video, and build Sume Avatar for consistent-face UGC ads. It reaches 20,000 users in eight weeks, 95% of them from organic Instagram and TikTok content.
- October 22, 2026: Im calls out the routing markups competitors charge: Higgsfield adds 40% even at its best discount, Fal adds 100%, Runway 414%. Sume API launches with 0% markup, a Higgsfield-style layout and MCP access for Claude and ChatGPT. On Seedance 2.5 at 1080p, it comes in 50% cheaper than Fal.
- October 26, 2026: They go after Higgsfield’s Genjutsu with Sume Remix, the same clone-any-video workflow at 2.6x cheaper, starting from $5.
- October 28, 2026: Six days after launch, Sume crosses a $1M annualised run rate with a team of two.
The lesson? Sume didn’t out-build Higgsfield. It looked at what the market leader was charging for simply routing someone else’s models, decided that margin was the weakest point in the business, and gave it away. Same layout, same models, 0% markup. Two founders who’d been mocked for pivoting every three days found the one pivot that worked by aiming it straight at a competitor’s pricing page. If your rival’s moat is their margin, it isn’t a moat. It’s your marketing budget.