Bending Spoons, the Milan-based owner of Evernote, Vimeo, AOL, WeTransfer and Meetup, has become the first public company to disclose granular AI productivity metrics inside a binding SEC filing. Its S-1 prospectus, filed ahead of a July 2026 Nasdaq listing, states that more than 90% of pull requests were authored or co-authored by AI by the end of Q1 2026, up from under 10% a year earlier, while revenue per full-time employee more than doubled from $1.12 million in 2023 to $2.57 million in 2025. That combination — legally attested numbers, not marketing claims — is what makes this the first genuine AI productivity case study the market can audit.
TL;DR
- Bending Spoons filed the first S-1 to quantify AI's effect on engineering output: over 90% of pull requests authored or co-authored by AI in Q1 2026, versus under 10% in Q1 2025.
- Revenue per full-time employee rose from $1.12M (2023) to $2.57M (2025), on a lean core of about 620 "Spooners."
- Hiring is extreme: 286 hires from roughly 800,000 applications in 2025, a 0.036% acceptance rate.
- The company IPO'd on 1 July 2026 at $29/share, closed the first day at $40.50, then drifted to $32.91 by 10 July.
- The model carries real risk: $4.4B in debt and $93.2M of Q1 interest expense mean the productivity gains must hold at scale.
- Because the claims sit in SEC filings, every quarterly report becomes a fresh data point.
What makes this the first SEC-filed AI productivity case study?
Companies have been claiming AI-driven productivity gains for two years, but almost always in blog posts, conference talks or investor decks, where the numbers carry no legal weight. Bending Spoons has put specific, quantitative claims about AI-authored code and revenue per employee into an SEC S-1 prospectus, which is filed under penalty of securities law. Any material misstatement would expose the company and its underwriters (Goldman Sachs, J.P. Morgan, Allen & Co) to enforcement action and class-action risk. That legal frame is what elevates the disclosure from anecdote to case study, and it means each future 10-Q will effectively test whether the reported gains hold.
What do the AI productivity numbers actually say?
Two figures do most of the work in the prospectus. First, code authorship: by end of Q1 2026 more than 90% of pull requests were authored or co-authored by AI, and roughly 70% were authored by AI with no human co-author. A year earlier that figure was below 10%. Second, output per person: revenue per full-time-equivalent employee climbed from $1.12M in 2023 to $2.57M in 2025 — a 129% increase over two years while core headcount stayed near 620.
Two caveats deserve equal billing. "Authored or co-authored by AI" is a broad definition; a pull request where a human reviews and edits AI-generated code counts the same as one shipped autonomously. And revenue per employee is flattered by acquisitions: Bending Spoons bought Evernote, Vimeo, Eventbrite, WeTransfer and Meetup, absorbing their subscription revenue while restructuring their headcount. Q1 2026 revenue of $601.3 million was up 132% year-on-year, but organic growth excluding acquisitions was 6%.
How does the company sustain this productivity model?
The prospectus describes a tight loop between hiring, tooling and compensation. On hiring, Bending Spoons processed roughly 800,000 applications in 2025; about 60,000 cleared initial screening, around 3,300 reached interview, and 286 were hired. That is a 0.036% acceptance rate, which the CEO Luca Ferrari has publicly acknowledged is "harder than Harvard." The company runs its own tests rather than relying on résumés, and it applies the same funnel across engineering, product and operations roles.
On compensation, 84% of eligible employees converted part of their cash salary into stock options in 2025, exchanging an average of 28% of salary. That aligns incentives to the operating-leverage story the S-1 tells: fewer people, higher output per head, larger equity upside. The founders retain 82.71% of voting power through dual-class shares, so strategic direction is not easily contested by public shareholders.
What are the risks the filing exposes?
Three risks are visible in the numbers. First, debt: total borrowings sit near $4.4 billion, split between $425.6M current and $3.93B non-current, and Q1 2026 interest expense was $93.2M, up 382% year-on-year. The productivity gains are effectively servicing that debt load. Second, retention: net revenue retention was 94% in Q1 2026, meaning existing customer cohorts are shrinking slightly year-on-year and growth relies on new subscribers plus acquisitions. Third, market scepticism: the shares opened at $40.50 on debut and had fallen to $32.91 by 10 July, still above the $29 offer price but a clear signal that investors want to see the AI-driven margins hold before pricing them in fully.
There is also a definitional risk. If auditors, competitors or regulators later argue that "AI-authored pull request" was measured too loosely, the headline metric loses credibility. Because the number is in an S-1, that argument would carry real consequences.
What can other teams learn from this case study?
The transferable lessons are narrower than the headlines suggest. A small engineering organisation, tight product portfolio and unified toolchain make it far easier to push AI authorship rates above 90% than a large enterprise with dozens of legacy systems. But three patterns generalise:
- Measure code authorship at the pull-request level. It is the smallest unit that ties AI output to review, tests and shipped behaviour.
- Pair AI tooling with brutal hiring selectivity. High leverage per person only works if each person clears a high bar; otherwise AI amplifies mediocre judgement.
- Publish the metric internally before externally. Bending Spoons only put the numbers in an S-1 after tracking them for at least a year.
Teams looking to build similar loops can start with our guide to the AI-native developer workflow and the broader systems-first approach to AI orchestration. For revenue-side benchmarks, the TCS AI revenue transformation blueprint covers a very different scale of adoption, and small operators can compare notes with the 2026 AI automation stack for small businesses and our walkthrough on how to build an autonomous AI employee.
FAQ
Q: Why does an SEC filing matter more than a company blog post? A: SEC filings carry legal penalties for material misstatements. Blog posts do not. That is why the S-1's AI metrics function as an auditable baseline.
Q: Does 90% AI-authored code mean 90% fewer engineers? A: No. The definition includes human co-authored PRs, and about 20 percentage points involve a human co-author. Bending Spoons still employs about 620 core staff; the leverage shows up in revenue per employee, not raw headcount cuts.
Q: How was revenue per employee calculated? A: The S-1 divides total revenue by average full-time-equivalent employees. It rose from $1.12M in 2023 to $2.57M in 2025. Acquisitions inflate the numerator, so organic revenue per employee is lower.
Q: What is the 0.036% acceptance rate based on? A: 286 hires from roughly 800,000 applications in 2025. About 60,000 took tests, 3,300 reached interviews, and 286 were hired.
Q: What should I watch in future filings? A: Whether AI-authored PR share holds above 90% as the codebase grows, whether revenue per FTE keeps rising post-acquisitions, and whether net revenue retention recovers above 100%.
Discussion
0 comments