Expedia Just Told You What an AI-Skilled Engineer Is Worth: 2 to 10x

Expedia’s CTO Ramana Thumu put a number on something most hospitality-tech leaders have only guessed at: after a nine-month study of 5,000 engineers, the company found AI-coding productivity gains aren’t evenly distributed — they concentrate almost entirely in the top 2-10% of performers, who become two to ten times more productive with AI assistance, while the rest of the engineering org sees little measurable gain. Expedia didn’t just publish the finding, it restructured hiring around it: a new San Jose office and at least 10 hires from major tech companies since June, recruited specifically to capture that outsized leverage. The result showed up fast on a real project — a checkout-platform migration originally scoped at 80 engineers over 15 months instead shipped with a smaller, AI-augmented team in seven.

The pattern lines up with what Anthropic has published about its own internal migrations. Bun co-founder Jarred Sumner used Claude Code to move Bun from Zig to Rust — roughly a million lines of code in under two weeks, with 100% of Bun’s existing test suite passing before merge — at a measured cost of about $165,000 in API tokens. Different company, different codebase, same underlying shape: a small number of engineers who already know how to direct an AI coding tool compress a multi-year project into weeks, and the cost of getting there is now a line item you can actually put a number on.

For hospitality operators watching labor costs and technology budgets converge, Expedia’s finding is the more useful one to sit with than any vendor’s AI pitch: the leverage isn’t in the tool, it’s in who’s driving it. A property or brand that treats “AI-skilled engineer” as a hiring category — the way Expedia just did — is making a bet that the 2-10x gap is real and durable, not a one-time honeymoon effect. Whether that bet pays off for a hotel tech team is still unproven; whether it’s worth testing on your next migration is not really a question anymore.