AI agent defeats Excel world champion in live head-to-head competition

Shortcut is an AI-powered Excel agent targeting financial professionals. Nico Christie, its founder, flew to San Diego last week to put the product to a live test: a head-to-head competition against Michael Jarman, the 2024 Excel World Championship winner, across four official championship models.

Shortcut won.

“I flew down to San Diego last week, nervous as hell, calendared a studio, we rented it out, and competed live on four official championship models, me versus Michael. And of course, I had Shortcut, but Michael had his big brain. We turn an operating model build out with 15 quarters of historic data — that can take thirty hours for a hedge fund to build — in one hour, then ten minutes.”

The competition

Jarman describes the Excel World Championship as deterministic problem-solving, closer to a hackathon than a speed-run. Contestants model board games, navigate maps, or track complex systems — the year Jarman won, competitors tracked 20 World of Warcraft characters leveling up through quests. Answers are graded against pre-calculated results, no style points. AI is currently banned under an honor code that Jarman says is getting harder to enforce as outputs become indistinguishable from human work.

Christie didn't enter as a hobbyist. He spent the prior weekend running practice cases, tuning verification loops, and building multi-agent adversarial review into his prompting — spinning off sub-agents across different frontier models specifically to catch errors before submission. His takeaway is that winning required genuinely skilled AI use, not just access to a good model. He deliberately chose Opus 4.8 Fast over a more capable model because the accuracy was already near 100% on verifiable tasks, and the speed advantage was decisive.

Where the real product challenge sits

The competition format is, by Christie's own admission, the easier case for AI. Tasks are programmatically verifiable — the answer is either right or wrong. Real financial modeling isn't. A DCF has 20 defensible structures, and the right one depends on company-specific context that no model inherently knows. That institutional knowledge, Christie says, is what Shortcut has to learn customer by customer.

The speed argument is more immediate. An operating model covering 15 years of quarterly historicals — roughly 30 hours of hedge fund analyst time to build, two hours to update each quarter — takes Shortcut about one hour to build and ten minutes to update. Christie believes a 10x further speed improvement is achievable, and points to the real unlock it would create: hedge funds currently track around 200 companies per pod, constrained by Excel throughput. At sufficient speed, that number could reach 2,000.

The harder constraint is that Excel doesn't run headlessly. Running hundreds of models simultaneously on a single machine isn't currently viable without expensive VM workarounds. Christie expects Microsoft to solve this, and is betting the product roadmap on it.

After the contest

Jarman, who now works in private equity investing in freight rail across North America, asked to get his team onto Shortcut before he left San Diego.