Summary

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.”

Deep dive

1. Anthropic’s growth curve has outrun its own aggressive cases

Lenny Rachitsky frames the public markers as $1 billion ARR at the start of 2025, roughly $4 billion by midyear, $9 billion at year-end, and $19 billion after 14 months. His comparison: Atlassian, Palantir, and Snowflake each took 15-20 years to reach approximately $4.5-6 billion.
Amol extends the pattern backward: 2023 went from zero to $100 million, 2024 from $100 million to $1 billion, and 2025 from $1 billion to roughly $10 billion. “That 10X year-on-year revenue growth trend has been there since the beginning.”
In his second week, Amol watched Dario push the aggressive 2025 planning case while others asked, “How the hell are we gonna hit that?” Amol thought, “There’s absolutely no way”; the company hit it, then failed to show the expected law-of-large-numbers slowdown.
The $19 billion figure was from the end of February and “also out of date,” Amol says, but he resists assigning growth-team credit. Research, inference, compute, Claude Code, and go-to-market drove the lion’s share; the operating reality is “hanging on by the seat of our pants.”

2. A cold email created a job that did not yet exist

Amol was an intensive Claude user who concluded Anthropic had a great product but “obviously” no growth team. With no growth-PM openings posted, he found chief product officer Mike Krieger’s personal email and wrote: “Love what you guys do. Love the product. I think you guys badly need a growth team. Wanna chat?”
His cold-email system separates opening from replying: tested subject-line copy generates a very high open rate, reaching a personal inbox avoids the crowded LinkedIn and work-email channels, and the message stays short—who he is, why he fits, and why they should talk. The exact subject line remains his trade secret.
Amol’s persistence rule is to continue following up on something important until the recipient says, “Please stop.” Krieger answered the first message, just as Anthropic was beginning to consider a growth team, and later told Amol he was the only PM Krieger had hired from a cold email.

3. Capability overhang makes activation the defining AI-product problem

Anthropic’s ChatGPT-memory import addressed a particular competitive moment, but Amol places it inside a broader cold-start problem: how can Claude understand a new user, reveal what it can do for them, and direct them to the right starting point?
His term for the industry-wide bottleneck is “capability overhang.” Even Anthropic employees must deliberately test each internal model and update their priors; a general user could have something approaching AGI and still ask only, “What’s the weather in SF?”
Activation work can decay before it ships. A team can map Opus 4’s capabilities, test on-ramps, learn, and redesign the flow—only for Opus 4.5 to unlock new capabilities that make those findings irrelevant. Day-zero and day-one activation therefore matter more than ever, even as their target keeps moving.

4. The right friction converts better because it makes a product feel personal

At Mercury, Amol’s team devoted an entire quarter to banking-onboarding quality: “Forget metrics, forget growth, forget everything else.” It repaired confusing details such as registered-agent, legal, physical-address, and beneficial-ownership interactions—and produced what was then his most impactful growth quarter, with a significant start-to-completion uplift.
Anthropic similarly asks new users who they are and what interests them, then recommends relevant products and features. Critics call the flow long and friction-heavy; Amol’s answer is empirical: “We have the data. We’re happy with how that’s performing.”
MasterClass’s purchase quiz looked like needless obstruction to someone ready to buy, yet testing showed it materially increased revenue by demonstrating that the catalog fit the buyer’s interests. Calm also uses a quiz in its landing, purchase, or login flow, while Mercury found that splitting five or six inputs across two screens lowered cognitive load and performed well.
The governing distinction is not friction versus speed. Remove steps that merely annoy; retain steps that help users understand “why the product is for them.” The resulting identity and intent data supports activation, lifecycle messaging, look-alike targeting, and advertising after a user drops out—“juice that just keeps on giving.”

5. Anthropic’s growth team organizes around audiences and takes unusually large swings

The roughly 40-person organization includes engineers, designers, PMs, and data scientists. Horizontal teams cover growth platform and monetization across the portfolio; focused pods cover B2B, Claude Code, knowledge workers, and API growth.
One universal activation team would struggle across Claude Code, Cowork, and other products because their audiences and internal partners differ. Amol stresses that every org design is temporary, but audience focus and tight relationships—such as Claude Code growth working closely with that product’s leaders—are the current priorities.
A conventional growth team might devote 60-70% of effort to small and medium bets and 20-30% to large ones. Anthropic moves closer to 50/50 or 70/30 in favor of large swings. Lenny notes that even 1% is enormous at this scale; Amol concedes those wins could make a quarter look impressive, but they are insufficient against a 10X reference curve.

6. Exponential product value changes the optimal growth portfolio

Amol contrasts AI-native products with a strong grocery-delivery or trading app. Even after excellent execution, the conventional product may deliver 30-50% more value two years later; Anthropic’s product value in that period is “probably like 1,000X, 100 to 1,000X what it is today.”
Each capability jump can create a market larger than the previous one. Agentic coding barely existed a year or 18 months earlier, yet its value already exceeds the prior AI-coding market—so concentrating only on high-confidence optimizations risks “missing the forest for the trees.”
The Chrome extension illustrates the larger-bet mandate. A bullish engineer saw the opportunity, nobody else owned it, and the growth team built what now underpins multiple Cowork and Claude Code use cases—“a very research-heavy product” Amol says he would not have attempted at another company.
His qualification matters: this operating model fits companies such as Cursor or Lovable when AI centrally underpins the value proposition. A conventional product with peripheral AI features should not automatically imitate it; the correct balance depends on the rest of the product and growth staffing.

7. CASH is turning the experimentation loop into an autonomous system

Anthropic’s growth-platform effort is called CASH, short for “Claude Accelerates Sustainable Hypergrowth”—a name Amol repeatedly disclaims as “a little cringey.” It began only a couple of months earlier and is explicitly still small and early.
Before Opus 4.5, Amol says the approach was not really possible; with Opus 4.6, the team began saying, “This is headed in the right direction.” It evaluates four stages: identify opportunities, build the change, test it against quality and brand standards, then analyze post-shipment data and extract learnings.
Current experiments are mostly copy changes and minor UI tweaks, but the machine is producing positive results: “You can press play with it and it’s like it ultimately prints money.” Its win rate resembles a junior PM two or three years into the role; Amol would still expect a senior PM to outperform it.
Human reviewers currently approve work and enforce brand standards, but Amol expects that burden to fall as models use skills containing brand rules, explicit dos, and don’ts. The harder missing step is stakeholder management: after one difficult meeting, Anthropic’s design head joked, “We will have AGI and it will still be impossible to get six people in a room to align.”

8. Engineering acceleration is squeezing PM and design capacity

AI is helping all three disciplines, but Amol sees engineering gaining the most leverage today. He translates a nominal five-engineer, one-PM, one-designer team into perhaps 15-20 old-world engineers, versus roughly 1.5-2 PMs and 1.5-2 designers: “PM and design are just squeezed. They’re absolutely squeezed.”
One possible response is unexpectedly PM-heavy hiring. Anthropic is actively recruiting growth PMs, while technical organizations such as Claude Code can rely more on engineering because their PMs are effectively engineers and the product itself is deeply technical.
Growth’s delegation rule makes the engineer the effective PM for projects requiring two engineering weeks or less, including conversations with security, legal, and other stakeholders. Above that threshold, the PM remains squarely accountable; a controversial one-week project is an explicit “use your head” exception.
Lenny challenges the prevailing advice that every PM should spend more time shipping. Amol’s nuance: at a small or engineering-constrained company, “Screw it, I am shipping” may be right; at scale, guiding 20 engineers 5% better on the why and what can create more value than personally shipping a twenty-first feature.

9. Prototypes and kickoffs are displacing most PRDs

Amol estimates that perhaps 60-80% of Anthropic growth work ships without a PRD. He is openly “averse to PRDs” and documentation: small changes can begin with a Slack exchange when strong, product-minded engineers can surface the missing questions themselves.
Larger work still earns rigor, especially through a 30-minute cross-functional kickoff. Bringing legal, safeguards, and other partners together to ask each, “What do you care about?” prevents far more coordination mess later.
When documentation is useful, Amol can feed rough thoughts into Cowork five minutes before a meeting, using a skill with his preferred format and projects containing previous PRDs. Yet his default remains “jump to action” and increasingly “jump to prototyping,” since showing an idea can communicate more precisely than describing it.

10. Cowork is already an always-on operating layer for metrics and administration

Each morning, a scheduled Cowork task reviews roughly 20-25 charts across numerous products and highlights what looks concerning or newly interesting. It follows Hex links through the Chrome extension for some sources and MCP connections for others.
Amol still opens a few charts because “I’m a numbers guy,” but delegates the medium and long tail. He says confidence can rise if the system’s false-positive and false-negative rates decline, providing coverage over metrics no executive could inspect daily.
He also hands Cowork meeting-room bookings, first-pass inbox archiving, Benepass reimbursements, and Brex expenses. The point is less novelty than removing recurring life administration: “Just hand it to Cowork. Just get rid of it.”

11. AI can now expose organizational misalignment and provide imperfect coaching

A weekly Cowork task uses the Slack MCP to scan Amol’s active projects and ask, “Go and find me areas of potential misalignment right now.” It can identify who should be consulted, what another team believes, and which constraints a proposed shipment may encounter.
For direct reports, Claude reviews weekly activity, team goals, OKRs, and discussion transcripts, then proposes observations and feedback. Amol runs the mirror image on himself, asking what feedback his manager Ami Vora would give based on her public writing, internal priorities, Slack activity, and their conversations.
The quality is hit or miss—Amol compares it to “working with a coach who’s kind of drunk at times”—but occasional findings are extremely valuable. An enterprise leader uncovered major misalignments that otherwise could have caused teams to duplicate work or spin their wheels.
Lenny imagines a strategy bot monitoring metrics, markets, roadmaps, and execution before recommending a pivot. Amol’s hedged prediction is that “later this year” this level of proactive, multi-source distillation will become very effective; that is his gut, not a certainty.

12. Coding and B2B focus came from both foresight and constraint

Amol recalls a document Ben Mann wrote in 2021, only months after Anthropic began, arguing that the company should focus on AI coding. This preceded clear evidence of the eventual market, reflecting an early leadership conviction around coding and B2B.
Coding offered two returns: a large commercial market and a research flywheel. Better coding models make researchers more effective, accelerating the work that produces better models and potentially tightening the loop again.
Necessity reinforced the strategy. Anthropic was historically “the smallest, least well-funded player,” without Meta or Google’s free cash flow and distribution or OpenAI’s first-mover advantage. Amol calls its survival “a complete miracle” and describes the strategic benefit as “freedom through constraints”: fewer viable choices forced focus.
History could have broken differently. Anthropic had a Claude chatbot before ChatGPT but chose not to launch it because of safety concerns and reluctance to start a global AI arms race; ChatGPT’s enormous traction naturally pulled OpenAI toward consumers. Amol concedes that an Anthropic-first launch might have reversed those positions: “Who knows?”

13. Safety is embedded in governance and sets hard limits on growth

Anthropic began as a public benefit corporation rather than a conventional Delaware C corporation. Amol’s distinction is that the structure permits the company to optimize for public benefit instead of treating shareholder-value maximization as its overarching legal goal.
The mission is to make the transition to powerful AI go well and remain net beneficial for humanity. Amol says Anthropic is optimistic about the upside but willing to take “a significant commercial hit” when safety requires it; withholding the original chatbot is his clearest example.
His growth framework separates two controversial tests. Category one crosses a safety, values, brand, or customer-friendliness line, so results are irrelevant and the test should never run. Category two creates “cringe or ick” without crossing a red line; it can be tested, but Amol demands proportionally high returns before accepting the compromise.
The broader doctrine is restraint: hardcore growth teams often try to “squeeze every last dollar,” but Anthropic is comfortable foregoing metrics to preserve safety, brand, quality, and user experience. Amol argues that the strongest products already operate this way and that credible safety could become a major long-term competitive advantage.

14. Mission intensity and open disagreement are Anthropic’s cultural moat

Amol joined without internal references and wondered whether Anthropic was genuinely mission-driven. His reaction after arriving was, “Oh shit. Okay”—people were even more serious internally than externally, and he says he has not met a single checked-out employee.
That shared understanding of AI’s upside and downside produces unusual energy: “Everyone is putting everything they have on the table.” Leadership transparency reinforces it, as employees are explicitly encouraged to challenge senior leaders rather than quietly defer.
Everyone can maintain a notebook channel—an internal Twitter-like feed for current thinking, updates, and provocative arguments. After an all-hands comment, one employee publicly told Dario they disliked how he framed the issue, triggering a broad debate rather than punishment.
Amol calls the talent density “playing for Real Madrid”: Mike Krieger, Ami Vora, elite researchers, and growth specialists coexist with people such as Jeff, a former U.S. ambassador to Australia, whom Amol encountered casually eating popcorn at an onsite. Culture plus talent, not any single tactic, is his “secret sauce.”

15. Notebook channels now serve humans and the agents advising them

Leaders use notebook posts to scale beliefs during rapid hiring. When Amol writes about “being comfortable leaving money on the table,” new growth engineers acquire a decision principle without requiring another meeting, reducing strategic drift as the organization expands.
Those records also become machine context. Some onboarding documents warn employees to consult an owner before editing because Claude treats the document as an important reference; growth, safeguards, and other functions are increasingly documenting their thinking partly so agents can represent it accurately.

16. Frontier AI companies still buy established SaaS

Despite its ability to build internal tools, Anthropic heavily uses Slack, Figma, Workday, and other SaaS products. Amol calls the future complicated—more software will be built internally as Claude improves—but says he does not see those core products disappearing in the immediate future.
Lenny’s pushback to the “vibe-code SaaS away” thesis is opportunity cost: Anthropic has more valuable things to build than another Slack, while mature software embodies years of hidden sophistication. Amol agrees that some disruption claims contain truth but others are overblown, and notes that many of these vendors are also valued Anthropic customers.

17. Staying commercially competitive is part of Anthropic’s safety strategy

Amol argues that linear and exponential thinkers perceive the timetable differently. Someone extrapolating from today asks how much better models could be in two or three years; someone internalizing the exponential expects major consequences sooner and sees more urgency in discussing both upside and downside.
Anthropic’s public warnings are sometimes softer than its internal beliefs, he says, not manufactured fear. Most employees remain optimistic, but they do not treat a good outcome as guaranteed and believe participants building the systems can describe specific risks more surgically than commentators outside the work.
The company’s strategy is to “drive the race to the top.” If Anthropic quits and shouts from the sidelines, “no one cares”; if it remains a leading, commercially successful player, it can influence competitors and the broader conversation toward safer principles.

18. AI-era careers will reward sharp spikes, adjacent skills, and adaptability

The baseline advice is relentless tool use: test Claude Code and Cowork after every model release, including tasks that previously failed. “One model launch later” a broken workflow may suddenly work; without retesting, someone can miss months of productivity and fail to develop AI-product judgment.
Beyond fluency, Amol recommends doubling down on an unfair advantage tied to impact. A stakeholder mediator, exceptional craft PM, product-minded engineer, or PM who can design becomes an “absolute unicorn” when neighboring functions are overloaded.
His own combination came from founding, investment banking, finance, sales instincts, and growth. He points to Nick Lin’s investment-banking and private-equity experience while building Claude for Sheets and Claude for Excel: “I know this. I know this. He’s built for this.”
Adaptability is the final filter. Joining Anthropic means throwing perhaps “50, 60, 70%” of prior operating habits out the door; people who keep applying inherited playbooks will create friction for themselves as the job changes underneath them.

19. Failure and injury turned constraints into Amol’s operating system

Amol’s largest professional failure was a mental-health startup he pursued for three years, funded with a couple of million dollars and staffed by roughly 7-10 people at its peak. Shutting it down meant telling employees and investors that a deeply held vision had failed—and took him years to process.
His tactical advice is to send monthly investor updates even when events are going badly. Keeping backers informed did not remove the pain, but it avoided surprises; the failed company also taught him product work and cold outreach, creating a career path he could recognize only in retrospect.
In early 2022, a Muay Thai kick caused a traumatic brain injury. He spent nine months off work, roughly half a year before he was comfortable walking again, and months unable to tolerate screens or even 20 seconds of music; an everyday bag strike after a flight reinjured him in mid-2023, forcing two more months off shortly after joining Mercury. He remains mostly, but not completely, healed.
The constraints produced durable practices: no alcohol or caffeine, a short break in the morning and another between lunch and the end of the day even during model launches, annual meditation retreats, and space between awareness and “reality’s insane.” His guiding lesson is that action and acceptance can coexist: do everything controllable, but pursue the “true freedom” of learning “how to be content when you don’t get what you want.” His paired mottos capture the posture: “Just go for it” and the Australian “She’ll be right.”

The State of Modern War: Palantir & Anduril Execs on Drones, AI, and the End of Traditional Warfare Training the AIs' Eyes: How Roboflow is Making the Real World Programmable, with CEO Joseph Nelson