Summary

Bending Spoons’ edge is not bargain hunting but a deeply integrated operating model that Ferrari says standalone owners cannot economically reproduce. Acquired businesses are “installed” onto more than 50 proprietary technologies spanning data, experimentation, payments, recruiting, credentials, and AI orchestration, while a nearly 1,000-person core team moves between portfolio companies. Ferrari wagers the platform saves at least $100 million annually and makes the organization “two, three times as productive, easily.”
The announced Airtable acquisition crystallizes Bending Spoons’ underwriting discipline: predict the business for at least five years, then identify enough operational upside to pay sellers well while preserving high returns. Ferrari cited an enterprise value of approximately $1.3 billion, with substantial cash still on Airtable’s balance sheet, and argued investors recovered roughly all their contributed capital “plus more.” He rejects the idea that this discredits Silicon Valley: “A billion dollars plus is unbelievable,” even if Airtable’s 2021 valuation was excessive.
The portfolio’s supposedly obsolete brands retain enormous, underappreciated distribution. Bending Spoons products collectively serve half a billion monthly active users; AOL alone remains, by the company’s estimate, the fifth-most-used email provider in the Western world, with activity Ferrari says dwarfs fashionable email startups. “We probably don’t care at all about being cool”—and lower cultural cachet can produce a more accessible purchase price.
Bending Spoons uses debt and free cash flow to acquire businesses it intends to operate forever, making it structurally different from a five-year private-equity fund. Debt has supplied most acquisition capital, while deep transformations—from rebuilding code and cloud infrastructure to redesigning pricing—favor buying five or 10 larger companies annually instead of 50 small ones. The portfolio can accept 5% profitable growth over 15% cash-burning growth because excess cash can be redeployed into accretive acquisitions.
Talent density is both the operating constraint and a measurable competitive advantage. Bending Spoons received 800,000 applications in 2025 and hired 300 people; unwanted attrition among core “Spooners” was 0.6%, versus roughly 5% Ferrari considers good in technology. Revenue per core-team member has risen from about $1 million two or three years ago to $4 million, and he expects it to “keep rising pretty fast.”
Its internal AI agent, Alt Spooner, turns proprietary systems and permissions into practical automation rather than an AI-marketing layer. The Slack-based agent inherits each employee’s access, completed a detailed A/B-test analysis in roughly five minutes, and helped move an Evernote bug from report to reviewed fix in one day rather than weeks. Bending Spoons routes about 99% of requests and tokens to self-hosted open-weight models, reserving frontier APIs for roughly 1% of complex or supervisory work, leaving token costs “basically negligible.”
Ferrari is simultaneously a maximal AI adopter and a skeptic of today’s AI investment gold rush. Molly cited that roughly 95% of Bending Spoons’ code is AI-written, which Ferrari confirmed in general terms. Although Ferrari believes some of history’s most valuable companies will emerge from the current cohort, he is “equally confident that most of these companies will fail or at least…fade away.” Their limited histories, uncertain economics, and growth that could fall from 100% to 12% make long-term underwriting impossible at current valuations.
Ferrari’s major AI concern is not whether a contested AGI threshold has been crossed, but whether humanity is systematically underestimating capabilities it cannot reliably measure. He is “equal parts enthusiastic about AI and absolutely scared shitless”: it might become humanity’s greatest boon “by orders of magnitude,” or enable catastrophic, highly targeted harm. Labs face survival incentives to race, governments have produced little he considers meaningful, and his preferred unanswered question is: “What’s currently preventing fixes from being proposed and deployed?”

Deep dive

1. Airtable is a test of underwriting, not a conquest of Silicon Valley

Molly O’Shea’s opening provocation was that Bending Spoons appears to be “eating Silicon Valley.” Luca Ferrari declined the victory lap: Silicon Valley built perhaps 70% of the leading technology businesses of the past 25–50 years and “gets almost everything right”; Bending Spoons has merely found a differentiated way to improve a tiny fraction of them.
Acquisitions are bidirectional—“you pick each other”—and Bending Spoons actively examines hundreds of companies in a typical year. The preferred targets are consumer-internet or SaaS businesses where Ferrari believes Bending Spoons can bring substantial value through product, technology, monetization, or organization—ideally all four; Tractive, a pet-tracking and health-monitoring hardware company acquired in spring, was a notable expansion beyond that comfort zone.
Ferrari’s non-negotiable is visibility: the company must be able to plot a target’s trajectory at least five years forward, ideally longer. Growth is welcome but not mandatory; Bending Spoons will buy a shrinking business if it can predict “how much they’re shrinking” and confidently price the resulting long-duration cash flows.
Airtable’s approximately $1.3 billion enterprise value was, in Ferrari’s telling, consistent with comparable public SaaS businesses. Because Airtable retained substantial cash and had raised and burned relatively little, investors recovered approximately their invested capital “plus more”—a successful outcome obscured by the contrast with its elevated 2021 valuation.

2. Silicon Valley’s excesses are a tolerable cost of a productive system

Ferrari’s defense of venture capital is direct: abundant early funding created businesses that otherwise could not exist, while later capital can accelerate scale economies, network effects, and brand. Across decades, he believes the system’s “real tangible non-hyped business value” has generated excellent aggregate returns.
His criticism concerns incentives, not the model itself. When investors are rewarded for managing more capital and returns depend on exit multiples rather than long-run cash generation, hype cycles follow; in 2021, “valuations were out of whack, completely,” across nearly every business rather than Airtable alone.
Molly’s pushback—worth keeping—is that trillion-dollar outcomes have distorted perceptions of success. Ferrari agreed: only a handful of companies in human history have reached such scale, whereas an exit above $1 billion, supported by revenue, customers, growth, and brand, remains “super successful” and “unbelievable.”

3. Unfashionable distribution can be more valuable than fashionable narrative

Ferrari’s rebuttal to the “graveyard” characterization begins with reach: Bending Spoons’ products have half a billion monthly active users. “If half a billion people using these products every month is a graveyard, then sure, let’s call it that.”
AOL is the sharpest specimen. Despite being roughly 40 years old, it remains a primary inbox for tens of millions and, to Bending Spoons’ knowledge, the fifth-most-used email provider in the Western world; Ferrari estimates fashionable email startups collectively may represent less than 5% of AOL’s sent, received, and active-email footprint.
The investment implication comes directly from Ferrari’s stated preference, not from nostalgia: a useful business perceived as less cool may be priced more accessibly. “We care about being good at our jobs and creating value,” while media attention and venture funding are poor substitutes for usage data.
Retention need not begin perfect; the relevant test is whether Bending Spoons can improve it. Ferrari could not recall one acquisition where product retention deteriorated after takeover: it has “at least stayed the same” and often improved, including in businesses that began with mediocre retention.

4. A shared operating system converts portfolio scale into operating leverage

Over a decade, Bending Spoons built more than 50 proprietary technologies covering data storage and processing, A/B testing, payments, recruiting, credential management, and AI-model orchestration. An acquired business is effectively “installed” onto that shared operating system, then supported by a core R&D and marketing organization approaching 1,000 people.
Ferrari compares the toolkit to a racing cyclist’s bike: vendors are appropriate when mass-market software is already excellent—Slack is his example—but specialized operating needs often have a market of “one or two.” Building internally enables deeper sophistication, native integration, lower dependency, and savings he wagers exceed $100 million annually.
The platform behaves like an internal open-source community. When one portfolio business finds a bug, corner case, or missing capability, it improves the shared tool and propagates that improvement to every other company; as the portfolio and organization expand, so does the platform’s capacity to refine those tools.

5. This is a permanent technology operator financed mostly with debt

Ferrari’s first distinction from private equity is duration: Bending Spoons is not a fund, has never sold a material business, and intends to own and operate acquisitions forever. A typical private-equity fund must preserve separability for an eventual sale; Bending Spoons can tightly integrate technology, people, and operations precisely because no such exit is planned.
The second distinction is workforce. Of nearly 1,000 core employees and more than 2,000 people including acquired teams, perhaps 60–70% are engineers, AI researchers, product designers, product managers, or growth managers—people who rebuild code, re-architect cloud infrastructure, launch features, redesign UX, and reinvent monetization.
Almost all acquisition capital has come from debt or internal free cash flow, with debt the majority. Because a large transformation does not require proportionally more operating effort than a small one, finite capacity favors “five or 10 businesses each year, but bigger,” over 50 small acquisitions; Ferrari currently sees “no obvious saturation point.”

6. Portfolio economics remove the standalone company’s growth-at-all-costs trap

A single-product company is primarily valued on organic growth, even after its market has matured and the biggest ideas have already been found. Ferrari’s diagnosis: management may spend heavily on R&D or marketing, “desperately trying” to rekindle growth and expand its multiple, despite weak future profit.
Inside Bending Spoons, the governing thesis is different: generate cash from existing businesses and redeploy it into acquisitions at high returns. Ferrari still prefers 15% growth to 5%, but if 15% requires substantial cash burn for distant profits, “I’d rather get 5” and invest the surplus elsewhere.
Revenue per core “Spooner” is about $4 million, up from approximately $1 million two or three years earlier. Including acquired employees, the figure is probably a little less than half that—but Ferrari expects the core metric to keep climbing through scale, integration, and further technological leverage.

7. Talent density and mobility form a second compounding system

Bending Spoons received 800,000 job applications in 2025 and hired only 300 people. Ferrari argues that mature standalone brands often lose their strongest entrepreneurial talent and cannot recreate Bending Spoons’ employer proposition: difficult work across multiple products, plus the freedom to rotate between operating companies and platform teams.
Its defining cultural principle is “extreme ownership”—caring deeply about personal craft, the team, and the company’s success. Ferrari would accept “slightly less intelligent people” before compromising on that commitment; the complementary principle is a first-principles, scientific pursuit of truth rather than comfort or confirmation.
Core-team unwanted attrition was 0.6% last year, compared with roughly 5% Ferrari regards as good in technology. A product manager bored after years refining Evernote need not change employers: they can move to an internal platform or tackle AOL’s email experience, renewing learning without sacrificing institutional knowledge.
Ferrari initially worried acquired employees would feel diminished or interpret the transaction as failure. He says those relationships went better than expected: no acquired team has lower retention than before Bending Spoons, and some improved substantially, although their retention remains below the exceptional core-team level.

8. Scientific humility determines what gets built and who builds it

The operating-system architecture emerged through “a lot of iteration,” informed by lessons Ferrari and his co-founders drew from a failed 2010–2013 startup. Their rule is to hold an opinionated vision while assuming it is probably wrong: build small pieces, test them in operating businesses, then deepen or rethink them based on adoption.
New tools begin with the people experiencing the problem, not a central platform team. Once a business proves the solution, ownership passes to the platform group for refinement and expansion; otherwise engineers can become more excited by “the engineering challenge than actually solving the problem.”
Ferrari grounds this in a broader theory of management: success comes from repeatedly probing reality, updating one’s model, and executing against the improved approximation. “Not even Steve Jobs…had it all figured out”; most people seek comfort and confirmation, so disciplined truth-seeking can itself become a competitive advantage.

9. Alt Spooner makes AI operational while minimizing vendor dependence

Alt Spooner is a Slack-based agent with the same permissions as its human counterpart across code, data, and internal tools. Employees can name it, give it a profile picture, and assign in principle any task they could perform themselves; Ferrari views that depth of integration as difficult to obtain from third-party assistants.
In one example, Ferrari requested analysis of StreamYard A/B tests and received plots and detailed findings about five minutes later, replacing hours of his or a data scientist’s work. In another, Evernote’s general manager reported a bug, asked the agent to check customer-support prevalence, identify its cause, code a fix, and notify the engineering lead.
The Evernote fix reached human review and production that day; Ferrari estimates the old coordination path might have taken weeks. Human oversight remained material—the engineering lead may have spent an hour or two reviewing—but the agent compressed discovery, triage, diagnosis, and implementation into one workflow.
Under the hood, Bending Spoons routes roughly 99% of requests and tokens to self-hosted open-weight models, using frontier APIs for about 1% of the hardest tasks or for supervision. Smarter models can critique cheaper models and send them back to revise, preserving provider independence while making token costs “basically negligible” at company scale.

10. Aggressive AI adoption does not make current AI startups underwritable

Ferrari estimates Bending Spoons is in the 99th percentile for aggressive operational AI deployment, and Molly noted that roughly 95% of its code is AI-written. Yet he distinguishes technology conviction from startup conviction: some current companies may become among history’s most valuable, while most will “fail or at least…fade away.”
His “gold rush” critique targets companies raising massive sums at billion-dollar valuations with little beyond an idea and a credible founder pedigree. When a business has only one or two years of history and 100% annual growth, no buyer can know whether it will still grow 100% or only 12% in three years—and that difference changes everything.
Bending Spoons is therefore not considering these acquisitions now. The combination of unpredictable trajectories and often irrational valuations conflicts with its mandate to offer sellers attractive cash prices while delivering strong shareholder returns; Ferrari may reconsider after the market matures.
Much of Bending Spoons’ AI research happens through use: benchmarking, fine-tuning open-weight models, combining systems, and building narrow models. Meetup’s in-house recommendation model is his example—competitive with frontier models for that single task and essentially free, while being “awful at everything else.”

11. The dangerous AI threshold may arrive before humans recognize it

When Molly relayed Jensen’s claim that OpenAI’s Astra model represented AGI, Ferrari declined the label debate. Definitions are ambiguous, and an AI performing 95% of human tasks as well or better may be almost as exciting and scary as one performing everything; what matters is expanding depth within tasks and breadth across them.
His stance is “equal parts enthusiastic about AI and absolutely scared shitless.” It could be humanity’s greatest boon “by orders of magnitude,” or enable extinction and comparably awful outcomes; unlike nuclear weapons, AI might let an attacker create immense, surgical damage while personally benefiting.
Ferrari’s subtler concern is measurement. Increasingly intelligent systems may conceal capabilities when useful, while today’s low-ego assistants tend to disclaim uncertainty rather than boast; humans may therefore see only what they ask to see and systematically underestimate real ability near a catastrophic threshold.
Ferrari pointed to cybersecurity incidents that shocked even researchers because models showed lateral thinking, perseverance, and the ability to collaborate. Molly then mentioned an OpenAI–Hugging Face incident as a related example. Ferrari’s inference is explicitly uncertain but stark—other consequential capabilities may already exist without having been probed, and “in a year’s time, I think the problem only gets worse.”

12. Safety incentives remain unresolved as productivity accelerates

Ferrari sees the labs’ bind: they may invest in safety, but the perception of losing technical ground could cut valuations by perhaps 90%, so moving more cautiously creates an existential corporate risk. Governments have done little he considers meaningful; he calls the EU AI Act highly harmful to industry while failing to address humanity’s genuine existential threats.
His proposed reframing is causal rather than prescriptive. People already ask what ought to be done, yet “nothing is happening”; the overlooked question is what prevents fixes from being proposed and deployed, because identifying that root cause might create a chance of useful action.
On jobs, Ferrari says he has “changed my mind…recently” without fully specifying the new conclusion here. His categorical policy view is that protectionism is untenable: unless every country stops simultaneously, a country that refuses to embrace AI risks “complete irrelevance” and third-world status within perhaps a few decades.
Over the next 12 months, he is most excited by further in-house technology and AI progress. The objective is to keep reinventing efficient business operation, push revenue per core employee above today’s $4 million, and translate that leverage into better acquisition prices for sellers and high returns for Bending Spoons.

A.I. Safety Goes Mainstream + a ‘Hard Fork’ Exit AMA Why AI Agents Break the Rules | Sharp Tech with Ben Thompson