Lenny Rachitsky

Anthropic’s public revenue markers describe an extraordinary scaling curve: roughly $1 billion ARR at the start of 2025, $4 billion midyear, $9 billion at year-end, and $19 billion 14 months after the starting point. Amol Avasare says that $19 billion figure was already stale because it covered only the end of February; internally, linear charts have become uncool because “everything is log linear.” Yet he says growth cannot claim too much credit: research, inference, compute, Claude Code, and go-to-market drove the lion’s share of the result.

The growth organization is designed around a bet that an AI-native product’s value could probably rise 100 to 1,000 times over two years, making large product swings more important than conventional funnel polishing. A normal grocery or trading app might add 30-50% more user value over that period; Anthropic expects the model exponential to unlock whole markets larger than their predecessors, as agentic coding already did. That is why its roughly 40-person growth team helped build the Chrome extension underpinning Cowork and Claude Code instead of limiting itself to incremental conversion tests.

Activation is the central product bottleneck because model capability is advancing faster than users can learn what to ask for. An onboarding flow optimized around Opus 4 can become obsolete when Opus 4.5 unlocks new behaviors, while even an extremely capable model delivers little if the user only asks, “What’s the weather in SF?” Anthropic’s response—including ChatGPT-memory import and detailed onboarding questions—is to accept useful friction that identifies the person and routes them toward the right product, feature, or use case.

Anthropic has begun automating growth experimentation through CASH—Claude Accelerates Sustainable Hypergrowth—and Amol says the early system can already “press play” and “ultimately print money.” It identifies opportunities, builds changes, checks quality and brand, then analyzes experiments after shipment; current work is mostly copy and minor UI, with a win rate comparable to a junior PM two or three years into the job. A senior PM remains better, but the system was not viable before Opus 4.5 and began looking promising with Opus 4.6.

AI’s immediate organizational effect may be more demand for PM and design judgment, not simply fewer people, because engineering leverage is rising fastest. Amol translates a five-engineer pod using Claude Code into something resembling 15-20 engineers in the old world, leaving PMs and designers “absolutely squeezed.” Anthropic therefore deputizes product-minded engineers as mini-PMs for projects requiring two engineering weeks or less, while PMs remain accountable for larger or unusually controversial work.

Anthropic treats focus, safety, and willingness to leave revenue on the table as mutually reinforcing strategic advantages. Amol connects the company’s early coding and B2B concentration to both research acceleration and necessity: it lacked Big Tech distribution, OpenAI’s first-mover advantage, and comparable funding, so constraints forced a narrow path. As a public benefit corporation, it will reject safety-crossing experiments regardless of measured upside; his growth doctrine is that “you just need to be okay leaving money on the table.”

The durable career hedge Amol recommends is not generic AI literacy but tool fluency combined with a sharply differentiated, interdisciplinary advantage. Product-minded engineers and PMs who can design become “absolute unicorns,” while people applying old playbooks should discard perhaps “50, 60, 70%” of how they previously operated. His own resilience thesis comes from shutting down a funded startup and surviving a traumatic brain injury: act on everything controllable, accept constraints, and learn “how to be content when you don’t get what you want.”

🔗 Original source & video: Head of Growth (Anthropic): Anthropic is automating its own growth

Google’s AI turnaround is less a single reorganization than the visible payoff from years of compounding investment, tighter product-research collaboration, and “an incredible sense of focus and urgency.” Lenny opens with Gemini reaching No. 1 in the App Store and excitement around Nano Banana. Stein describes the broader monthly product-and-model improvements as reaching a tipping point: frontier models have become useful enough that consumers can finally feel the accumulated gains.

Stein rejects the “Google is dead” thesis because AI is expanding what people search for rather than replacing Search’s vast base of navigational, transactional, and factual jobs. Google Lens visual searches are growing 70% year over year from an already “billions and billions and billions” scale, as users photograph shoes, homework, or bookshelves and ask questions that were impractical in keyword search.

Google’s competitive AI asset is the combination of frontier models, distribution, and live structured information—not merely a chatbot embedded in Search. AI Mode can tap 50 billion products updated two billion times an hour, 250 million places in Maps, finance data, and the web; AI Overviews and Lens increasingly become previews that lead into the same conversational system. The investor-relevant claim is that Google can turn existing intent into richer queries without requiring users to form a new habit elsewhere. For a forthcoming visual version, Lenny frames the opportunity as a potential threat to Pinterest; Stein distinguishes it from Nano Banana, which is an image editor.

AI Mode’s query fanout turns one prompt into potentially dozens of background searches, pairing model reasoning with real-time information, spam detection, authority signals, and links for verification. Stein’s answer to AEO/GEO is therefore evolutionary: satisfy intent, provide original and well-sourced information, and focus on the advice, how-to, and complex questions AI is causing people to ask more often. “At the end of the day, actually something’s searching.”

AI Mode is positioned as an information product, not a general-purpose therapist, creative companion, or spreadsheet workbench. Google is betting that users will move from “keywordese” to five-sentence natural-language requests—such as finding outdoor date-night options after excluding four restaurants and accommodating an allergy—while retaining access to authoritative sources and follow-up questions.

The speed of the launch is evidence that Google can still operate like a startup when conviction is high. A five-to-10-person team began roughly a year before the conversation, found a few “moments of brilliance,” tested with about 500 outsiders who were encouraged to say when it sucked, expanded through Labs, and then launched broadly in the US. Stein believes “the next year or so of product” might establish consumer habits for many years.

Stein’s operating system combines relentless dissatisfaction with hard instrumentation: vision identifies the better world, while retention curves and root-cause analysis show whether the product is actually getting there. Teams should watch day-seven, day-30, and day-90 retention, find where an S-curve is flattening, and move investment toward new engines where individual changes can still generate 10%, 20%, or 4% wins. His counter to the “cult of lean” is that difficult breakthroughs often die because teams remain understaffed after internal conviction arrives.

Instagram Stories and Close Friends illustrate how to adopt proven formats without merely cloning them—and how long compounding improvement can take. Stories became Instagram-native through camera-roll uploads, pausing, creative tools, and a coherent placement; Close Friends took two or three years to recover from confusing design and mistranslation, then worked when lists reached roughly 20–30 people and could produce two DM replies in Stein’s example. The broader rule is “clarity instead of cleverness,” paired with enough humility to admit the first version failed.

🔗 Original source & video: Inside Google’s AI turnaround: AI Mode, AI Overviews, and vision for AI-powered search | Robby Stein

Scale AI remains independent after Meta invested a little over $14 billion for 49% of its non-voting stock, according to new CEO Jason Droege. Meta received no new board seat or preferential data access; roughly 15 of Scale’s 1,100 employees moved, while Alex Wang joined Meta and retained his Scale board seat. Droege says Scale’s two major businesses each generate hundreds of millions in revenue, the business has grown every month since the deal, and it recently signed two $100 million government contracts.

Enterprise AI’s delivery gap is measured in reliability and implementation time, not an absence of economic value. Proofs of concept often reach 60%–70%, but closing the remainder resembles adding successive “nines” of data-center uptime: legal, policy, regulatory, accuracy, and change-management work makes important automation a six-to-12-month project. Droege’s summary is “easy to learn, hard to master.”

Frontier-model training has moved from quick preference rankings to hours-long demonstrations by elite professionals. A task that was choosing between two short stories 18 months ago can now require a top developer to build and explain an entire website or a PhD to teach nuanced cancer knowledge; 80% of Scale’s expert network has at least a bachelor’s degree and roughly 15% has a PhD. Behind seemingly magical models is persistent “operational chiseling”: compute, model improvement, and increasingly specialized data all improve together.

The next enterprise bottleneck is digitizing institution-specific judgment rather than ingesting more raw data. Scale’s healthcare example turns 200–300 pages of mixed-format records into five to ten considerations and once surfaced an allergy that conflicted with a planned medication. Off-the-shelf models, RAG, and fine-tuning can only go so far because identical words may carry different importance across companies; the local experts themselves must increasingly label “what good looks like.”

Models are moving from knowing things to doing things, making reinforcement-learning environments a critical infrastructure layer. Agents must learn inside realistic sandboxes—navigating a configured Salesforce instance, handling business data, completing goals, and escalating uncertain decisions—while labs seek training tasks generalizable enough to avoid collecting “forty-five trillion combinations.” Droege expects the technology could become close enough within two to three years to force difficult change-management and policy decisions.

Droege rejects a near-term white-collar apocalypse while openly acknowledging Scale’s incentive to keep humans involved. He does not think the transformation will happen in the next year and calls it within two years “very far-fetched,” though “nothing’s impossible here”; longer term, he argues that if these systems are to work for people, humans will need to remain in the loop for consequential decisions. His deeper thesis is that labeling has a “history of new beginnings”: as old needs fade, new human knowledge and skills become valuable.

Droege’s company-building framework combines independent insight, buyer urgency, structural economics, and survival. Uber Eats reconstructed restaurant economics, initially charged 30% before the market settled around 25%, and pursued incremental demand that, if demand tripled while labor stayed fixed and only ingredients scaled, could carry 70%–80% incremental gross margin. The business then grew from zero to roughly $20 billion in four and a half years. Founders still need a reason they uniquely see the opportunity, the willingness to spend five to ten years on it, and the discipline to remember that “not losing” is a prerequisite to winning.

🔗 Original source & video: Scale AI CEO on Meta’s $14B deal, scaling Uber Eats to $80B, & what frontier labs are building next