Bending Spoons hires 286 from 800,000 as AI writes 90% of code

Getting hired at Bending Spoons in 2025 was, on the numbers, harder than getting into Harvard. The Italian owner of AOL, Vimeo, Evernote and Eventbrite made just 286 hires from roughly 800,000 applications, keeping its core headcount, the Spooners, at around 620, according to reporting in the Wall Street Journal. That is a rejection rate a shade above 99.9%. The number underneath that headline is the more interesting one to me though.

Revenue per full-time equivalent Spooner reportedly climbed from $1.12 million in 2023 to $2.57 million in 2025, more than double in two years, on a headcount that barely grew. The company's IPO disclosures attribute a chunk of that to AI inside the engineering org: the share of pull requests authored or co-authored by AI went from under 10% in the first quarter of 2025 to more than 90% by the first quarter of 2026, with roughly 70% written by AI alone. A company that publicly refuses to hire fast is, on its own telling, also one that has moved most of its code output onto machines in about a year.

Why this matters beyond one Italian acquirer: Bending Spoons has quietly become a template. It buys tired consumer software, keeps the brand, thins headcount, and runs the codebase with a small core team on heavy AI leverage. If these productivity numbers hold up on the other side of the S-1, every roll-up shop and PE operator is going to point to this and ask the same question of their own portfolios.

The honest caveat is that these are self-reported IPO-adjacent metrics from a company with an incentive to make the AI story sound tidy. 'AI co-authored' is a wide bucket. Revenue per employee also flatters a firm that fires most of the staff it acquires and then counts only the survivors, so the ratio is doing double duty as productivity story and headcount story. What the reporting doesn't give you is the retention picture inside those 620, or how much of the AI-written code is durable versus rework a year from now.

The direction is still the part worth watching. If a lean acquirer can plausibly claim mid-seven-figure revenue per head with most of the code written by machines, the floor on how few humans a legacy consumer software franchise needs just moved, and the roll-up model gets a lot more interesting to capital currently sitting on the sidelines.