Anthropic’s $1B to $19B growth run: how Claude became the fastest-growing AI product in history
In a Nutshell
Anthropic exploded from $1B to $19B ARR in 14 months under Head of Growth Amol Avisari by focusing on activation fixes like importing ChatGPT memory, audience-specific pods (e.g., B2B coding), and AI-automated experimentation via CASH, prioritizing massive bets over micro-optimizations amid 10x YoY growth and log-linear charts. Key tactics include adding "right friction" in onboarding for better product matching (proven at Mercury/MasterClass), a 40-person growth team blending traditional structure with Claude-powered loops for opportunity ID, building, testing, and analysis, and leaving money on the table for safety/UX as a public benefit corp. PMs/engineers thrive by mastering AI tools daily, leaning into interdisciplinary strengths, delegating small projects (<2 eng weeks) to product-minded engineers, and embracing constraints for focus—echoed in Amol's brain injury recovery emphasizing meditation and exponential thinking.
These notes were generated by AI and may contain inaccuracies.
Anthropic grew from $1B ARR at the start of 2025 to $19B ARR in 14 months. Historically the smallest, least funded player without Meta or Google's free cash flow/distribution, or OpenAI's first-mover advantage.
This is the hardest job I've had in my life. To come into Anthropic, you need to understand that 50-70% of how you operate in the past, just throw it out the door.
Clever growth move: importing memory from ChatGPT to address activation challenge in AI. Growth platform team driving CASH (Claude Accelerate Sustainable Hypergrowth) to use Claude for automating growth experimentation, delivering results. Product value in 2 years will be ~1,000x today. Internally, linear charts are uncool; everything is log-linear.
Guest: Amol Avisari, Head of Growth at Anthropic. Grew from $1B to over $19B ARR in 14 months, doubled revenue in past few months, 10x YoY growth. Adds as much ARR every few months as Atlassian, Palantir, Snowflake do annually after 15-20 years ($4.5-6B ARR each). Amol previously led growth at Mercury and MasterClass, was a founder and investment banker. Suffered severe brain injury, spent 9 months relearning to walk/work without nausea; shared in Lenny's newsletter guest post.
Amol cold emailed Anthropic's Chief Product Officer Mike Kreger: "Hey, love what you guys do. Love the product. I think you guys badly need a growth team. Want to chat?" Mike responded; they weren't publicly hiring for growth but timing was perfect. Mike is the only PM he hired from cold email.
Cold email tactics perfected as founder:
- High-open-rate subject line (secret).
- Use personal email (not LinkedIn/work email) to stand out.
- Keep short: who I am, why good fit, chat?
- Follow up repeatedly until told to stop.
Company-wide effort; growth team drives impact but credit goes to best-in-world research team, inference/compute teams, Claude Code, go-to-market, etc. Revenue trajectory: 2023 $0-100M, 2024 $100M-$1B, 2025 $1B-$10B. Joined 2024 when revenue in hundreds of millions. Early 2025 planning: Daria pushed aggressive scenarios everyone thought impossible; exceeded them. End of 2025: expected slowdown at $10B baseline due to law of large numbers, but hasn't slowed. $19B figure from end of Feb (outdated). Internally, all charts log-linear; hanging on by seat of pants, managing growth for users.
Any growth need goes to Amol's team and it works, despite magical Claude/tools.
Hardest job ever (vs. founder, investment banker). Categories same as normal company: acquisition (more people/intent), activation (signup flow, funnel to right products, success), monetization (free/paid conversion, pricing/packaging).
70% time on success disasters: scaling breaks things (e.g., Facebook/Uber/DoorDash early). Firefighting urgent issues across categories despite all charts green/up/right. Emotionally tough.
30% proactive: prioritize products with limited resources, long-term pricing/packaging amid tech/behavior shifts, timing leans (e.g., post-Co-Work launch, optimize adoption funnel).
Always improving cold start/new user experience. Activation huge AI challenge. Import memory from ChatGPT specific to moment but part of broader: make Claude understand user/needs, route correctly.
Activation critical (day 0/1 experience) for long-term retention; importance exponentially higher now. Industry problem: capability overhang—models improve fast, but product diffusion lags. Even internally, busy teams struggle to update priors with new models. Users default to low-value prompts (e.g., "weather in SF") despite AGI-level capabilities.
Cycle: New model (e.g., Opus 4 → 4.5 unlocks capabilities) → test on-ramps → ship flow → next model obsoletes learnings. Hard to stay ahead.
Key learnings (timeless): match right product/feature to user characteristics; right friction helps (adding steps often boosts conversion if tested).
Ben Mann (co-founder) highlighted Amol's Mercury onboarding redo. At Mercury (fantastic product, better banking experience; recommend personal banking):
- Onboarding complex (regulated: registered agent vs. legal vs. physical address, beneficial ownership).
- Spent full quarter fixing quality (ignore metrics/growth, just best experience).
- Huge uplift in onboarding start-to-completion; most impactful quarter as growth PM.
Quality drives growth. At Anthropic: onboarding asks who they are/interests → recommends products/features. Critics say too much friction/long; data shows it performs well.
Philosophy on friction:
- Adding right steps consistently leads to higher conversion/funnel completion across roles (MasterClass, etc.).
- Test thoroughly, esp. high volume.
- Ditch annoying friction (no value); but pure time-to-value/minimal steps often fails.
- Examples:
- MasterClass purchase flow: quiz on interests → recommends classes/content; revenue driver (thoroughly tested).
- Calm.com (ex-team member): quiz in purchase/login flow.
- Mercury: broke out steps in onboarding (one screen tested).
Testing showed breaking onboarding into fewer screens reduces cognitive load. For example, one screen with five or six form inputs performs better when split into two screens. Anthropic applied similar steps in their flow that performed well. Key takeaway: cut friction when it doesn't help users understand why the product is for them. However, if friction helps users understand the product, how to use it, and what's most relevant, test and confirm it works.
The "for them" is really important—adding friction to better understand who they are so you can recommend the right thing.
This flows through activation and lifecycle. Knowing more about users enables lookalike targeting even for drop-offs at the ad layer. Initial user understanding provides ongoing value down the funnel.
Everyone will tear down Claude's onboarding, Mercury's onboarding as masterclasses. At a recent PM dinner, a PM noted AI has most impacted competitive analysis, like tearing down pricing pages and onboarding flows. Now easy with AI: ask Claude or Cursor which tool to use for teardowns. Use Co-work with Chrome extension to task it with analyzing flows, showing what's working and not.
Growth team is roughly 40 people, structured like a traditional growth team with horizontals (growth platform, monetization) across all products, and audience-focused growth pods: B2B growth, Claude code growth, knowledge worker growth, API growth. This narrow focus is essential with many products.
Team includes engineers, designers, PMs, data analysts. Shape similar to traditional growth teams but indexes more toward larger swings (50/50 or 70/30 larger bets vs. small/medium) rather than 60-70% small/medium, 20-30% large.
At Anthropic's scale, a 1% win is massive, yet they avoid over-focusing on micro-optimizations. They track 10x year-on-year growth, fixated on the exponential due to model capabilities unlocking new markets. Agentic coding example: didn't exist 1-1.5 years ago, now bigger than prior AI coding market.
In non-exponential businesses (trading app, grocery delivery), product value might rise 30-50% in two years, allowing growth teams to capture value with small/medium optimizations. For Anthropic, product value in two years is 100-1,000x today, so shift to larger bets to avoid missing massive opportunities. Still do small optimizations for compounding value, but prioritize core product swings.
Example: Chrome extension underpins Co-work and Claude code use cases—growth team built this research-heavy AI product because an engineer was bullish and no one else was doing it.
Takeaway for other companies: if primary value is AI-underpinned, shift toward bigger bets to capture future scale. Applies to AI-first products like Lovable, Cursor.
Structure combines product-specific (API growth, Claude code growth) and persona/vertical (knowledge workers, B2B). With multiple products, pure funnel structure (conversion, activation, monetization) fails due to different audiences and cross-functional stakeholders. Current structure maximizes focus on audiences/problems and ties to partners (e.g., Claude code growth works closely with Kat, Boris).
All structures are imperfect, right for a point in time.
Skew toward larger bets if core value is AI-backed. Automating growth is accelerating: growth platform team led by Alexei Kamisaroka (teaches growth engineering at Reforge) drives "CASH" (Claude Accelerates Sustainable Hypergrowth)—using Claude to automate growth experimentation.
Kicked off couple months ago; pre-Opus 4.5 not viable, now with Opus 4.6 heading right direction. Growth teams suited early due to small optimizations.
-
- Identifying opportunities (based on trends, past trends Claude has seen).
-
- Building the feature and preparing to ship.
-
- Testing to meet quality/brand bar.
-
- Analyzing data, gathering learnings.
Score Claude's performance in each area; testing at small scale (copy changes, minor UI tweaks). Delivers results like a junior PM (2-3 years experience), not yet senior PM level, but improving rapidly per exponential curve.
Will scale to larger experiments. Cross-functional stakeholder management remains human need, especially for large projects—PM work not going away soon.
Amal, we will have AGI and it will still be impossible to get six people in a room to get to a line.
Tool comes up with experiments; human approves; builds, ships, tests, iterates. Week-on-week: improving speed/results in each area scales the initiative. Much can be automated without human review, using skills with brand guidelines/dos/don'ts. Can unship bad ideas.
AI expanding in product/growth: from writing code to full loop. Growth is data-driven loop, ideal start. AI now good at telling what to do, not just executing orders.
Picture changes rapidly. Bottlenecks now involve tech shifts beyond ratios. Smaller companies (15-20 people): PMs blend designing/shipping. Larger orgs: unclear if roles converge or separate.
Engineers get most leverage now (e.g., Claude Code 2-3x productivity). Default team (5 engineers, 1 designer, 1 PM) now like managing 15-20 engineers equivalent, straining PM/design.
Not uniform—technical products like Claude Code fit (PMs are engineers). Hiring more growth PMs desperately; apply via crafted cold email. Also hire product-minded engineers as mini-PMs.
Frame: projects <2 engineering weeks, engineer owns as PM (security, legal, stakeholders); PM advises. >2 weeks, PM defaults as...
If a project is more than two engineering weeks, the default is that the PM should continue to be on the hook for making that go well, unless they delegate more to engineering, but they remain squarely accountable. It's not fully clean-cut; use your head. If it's a one-week thing but extremely controversial, the PM should probably still drive it. This approach deputizes engineers to be mini PMs. Not everyone can do it, so product-minded engineers' value goes up significantly, an order of magnitude. Companies will still hire a lot of PMs.
PMs joke that you can go on vacation for a couple weeks and things will be all right due to momentum and plan. This connects to the two-week rule of thumb: if it's a two-week project, you'll be all right without a PM.
Engineers are so accelerated that PMs and design struggle to keep up. One route is more PMs to keep pace; another is engineers who can PM. Great news for PMs until AI handles more PM work. Ratio of more PMs to engineers might be the future, but it depends on industry, company size.
Developer-focused products rely more on engineers. Early companies have less cross-functional coordination and stakeholder alignment, so fewer PMs suffice. As company scales, one PM becomes two from productivity gains, five engineers become 20, one designer becomes three.
Best Use of PM Time in Scaled Teams
In engineering-limited world, PMs should ship things. Today, PMs should ship to understand tools. But with 1-2 PMs and 20 engineers, consider incremental value: shipping the 21st feature or guiding the team on right opportunities? Engineers as mini PMs free PMs for high-leverage work: improving the why and what by 5%.
Counter to common thinking: instead of PMs prototyping and shipping PRs, better to help engineers become better PMs. Leverage is higher than coding in most cases.
Exceptions: smaller companies (all hands-on deck), tight budgets, limited engineers—PMs need to ship to accelerate impact. For larger, scaled businesses with 20 engineers, highest leverage is upleveling: better user insights, etc.
Shipping to learn via prototypes informs roadmap: try three things, see results. PRD becomes 'look at it' instead of talking.
Even with many PMs and engineers, prototype to articulate ideas. Company is scrappy despite big valuation: minimize bureaucracy, just go. 60-80% of shipped items lack PRD; averse to PRDs and documentation—cut blockers. 20-30% important items need good documentation.
PRDs are outdated; kick off with good team without them, jump to prototyping.
Depends on size, tying to two-week filter: heavy investment needs more thinking; smaller, just go. Growth team has many small changes.
Small changes (e.g., upsell for new thing): on Slack, messages back and forth. Requires good caliber, product-minded engineers who ask clarifying questions.
Larger things: proper cross-functional kickoff (30 minutes) with legal, safeguards, everyone: 'This is what we plan. What do you care about?' Streamlines mess that could happen later.
PRD approach: sometimes no PRD due to speed. Set up meeting, 5 minutes before add basic doc with thoughts, key considerations. Other times, no doc.
When creating PRD: use Claude skill with format (previous PRDs in projects). Simple: things I care about, why, problem, key considerations, cross-functional stakeholders. Default: avoid doc, jump to action/prototyping.
Future: automate with AI assistants (e.g., your assistant talking to legal assistant ironing out concerns). Legal team enables with 'how to mimic what someone might say,' co-op setups. As Claude gets more context, better at this.
Effective use: identify misalignment. With Co-Work Slack MCP, tell it: look across Slack on projects top of mind, find potential misalignments. Scheduled weekly; flags areas to be aware of.
In shipping: Claude scans company, says who to talk to, what to keep in mind.
PM jobs evolving to AI: find/align misalignments, growth automation, product improvement ideas (AI gives good ideas).
Sponsor break: Vanta automates compliance/risk for 35+ frameworks (SOC 2, ISO 27001, HIPAA). Helps get compliant fast. vanta.com/lenny for $1,000 off.
Extensive use: writing docs, brainstorming, data, UXR.
Data: Co-Work scheduled run on 20-25 charts every morning. Summarizes: things to pay attention to, concerning items, insights. Comes up in desktop app. Uses Hex links, Chrome extension, MCP.
Still check key charts personally (numbers guy). For long/medium tail, Claude proactively surfaces; confidence builds as false positives/negatives drop.
Strategy bot vision: agent watches metrics, market, roadmap; suggests pivots, wins. Close, effective later this year. Proactive across data sources, distilling insights—like alignment but broader.
Admin: Claude books meeting rooms, archives email, clears inbox first pass, files reimbursements (Benspace? Ben pass), expenses (Brex).
Manager Lens Automations
Alignment across direct reports: Claude reviews what they've done weekly, team goals/OKRs, discussion transcripts. Key takeaways, feedback suggestions. Weekly setup; quality hit-or-miss but often glad to catch issues.
Self-feedback: prompt Claude as manager Amy Vora (podcast guest, written extensively on product): based on what I've done/not done this week, feedback as Amy. Weekly.
Many things doable today; accuracy improves rapidly.
Ammy example: one-on-ones, Claude builds model from her writing/public profile or Slack MCP (last week says, priorities). Infers manager priorities from time spent.
Soft coaching unlocked, like working with sometimes inaccurate coach. Examples: Scott (enterprise lead) found major misalignments preventing wheel-spinning/overlaps. Cross-functional coordination toil reduced by AI; impossible 6 months ago, much more in 6 months.
Requires Slack data, notes, discussions.
Setup: Co-Work desktop app, connect Slack MCP (may need team/enterprise admin permissions). Then ask Claude.
Incredible growth from focus: few things, super deep. E.g., B2B deep on coding (Claude Code). Competitor shifting to similar.
Foundational from early days, from leadership. 2021 doc by co-founder Dan Mant (podcast guest): why focus on AI coding—5 years ago, pre-market clarity.
Two lenses: commercial benefit, accelerating research (best models accelerate researchers' loop). Dario saw this clearly years ago. Company DNA.
Necessity: smallest, least-funded player. No Meta/Google cash/distribution, no OpenAI first-mover. Miracle to current stage.
Broader principle: freedom through constraints. When you have a bunch of
Constraints bring freedom by eliminating excess choice and clarifying the path. With limited funding, small team size, and no distribution, Anthropic focused narrowly to maximize chances of reaching escape velocity with generalizable technology.
Anthropic developed a version of Claude and a chatbot before ChatGPT launched but chose not to release it for safety reasons, avoiding kicking off an AI global arms race. ChatGPT launched, gained insane traction, and pulled OpenAI toward consumer products. In an alternate history, if Anthropic had launched first, outcomes might differ.
People don't realize how far behind Anthropic was—when raising money, it seemed impossible to compete with OpenAI's lead. It shows the power of focus.
Anthropic's success stems from its strongest leadership team from top companies worldwide, plus incredible people.
Deep bet on coding targets huge TAM and creates a feedback loop: excelling at coding accelerates research and better model building.
Dario has discussed this publicly.
Anthropic is officially an AI safety research company. Growth balances expansion with mission of AI safety and alignment.
Founders left OpenAI to start Anthropic, embedding safety in corporate structure as a public benefit corporation (PBC) instead of Delaware C-corp. PBC legally prioritizes public benefits over maximizing shareholder value.
Mission: Ensure transition to powerful AI is net beneficial for humanity. Excited about upsides but aware of risks; willing to take significant commercial hits.
Examples: Delayed Claude release due to safety risks, yet it worked out well.
Growth Framework for Controversial Tests
Two types:
- So controversial you shouldn't run it—results irrelevant because you won't ship due to brand, customer friendliness, values (red line).
- Controversial but not a red line—run if strong hypothesis and high expected return justifies cringe/ick factor.
At Anthropic, AI safety is firmly type 1.
Core Growth Principle: Leave Money on Table
Growth teams err by squeezing every dollar, harming long-term relationships (e.g., founders raising money). Anthropic growth team comfortable forgoing metrics to prioritize safety, brand, quality bar, user experience.
Best products operate this way, driving more long-term growth. Safety stance becomes competitive advantage as risks/stakes rise.
Focus on creating good content; micro-optimizations don't matter—it grows through sharing if useful.
- Use the tools daily: Master Claude, Cursor, track model releases, experiment to discover applications/boost productivity/gain AI product sense. Revisit failures as models improve.
- Lean into competitive advantages: Double down on spike skills (e.g., stakeholder mediation, craft). Ignore weaknesses; become best at high-impact strengths.
- Be interdisciplinary: Engineer who is product-minded or PM who designs = unicorn. Lowers layoff risk, increases usefulness.
Speaker's background: Founder, investment banking (finance/numbers), near-account exec (sales), growth → outsized impact in certain areas.
Example: Nick Lynn runs Claude for Sheets/Claude for Excel (tanked market); ex-investment banking/private equity, built for financial services product.
- Be adaptable: Discard 50-60-70% of old playbooks; friction from sticking to past methods.
Echoes advice from Jenny (Claude/CL Work design leader): Go deep on specifics (sideways E vs. T-shaped). Marc Andreessen said similar.
Secret sauce: Mission-driven culture no one can replicate. Leadership (esp. Dario, Daniel) invests heavily.
Speaker skeptical pre-joining but found internal seriousness exceeded external talk. Everyone viscerally grasps AI upsides/downsides, divergent futures → passion/belief.
No checked-out employees; all pour everything in, creating indescribable energy.
Open culture:
- Transparent leadership.
- Slack maze for discussions.
- Notebook channels: Personal internal Twitter feeds for thoughts/priorities. Join anyone's (e.g., research) to learn.
- Encouraged to publicly challenge leaders (e.g., disagree with Dario on all-hands via notebook → debate).
Builds trust/togetherness unlike anywhere else.
Talent density: Best researchers worldwide; product has Vora, Mike Krieger (co-founded Instagram); growth has John Eggan (OG growth engineering), Alex Guy (Reforge growth engineering teacher).
Anecdote: At LA onsite (October), spotted Jeff—US ambassador to Australia—eating popcorn alone; chats with Australia's prime minister.
Notebook Channels Purpose
Scale beliefs/views as org grows: Model behaviors/principles publicly (e.g., post on leaving money on table → new engineers align).
Prevents strategy/perception drift from rapid hiring; runs tighter ship.
AI Integration
Notebooks feed Claude (e.g., onboarding docs reference-checked before edits). Arming Claude with team contexts (growth, safeguards) for better help.
Companies must structure thought-sharing for agents.
Use Slack, Figma, Workday heavily despite AI hype replacing SaaS. Build some tools internally; Claude ability improves but no immediate change.
Many SaaS are Anthropic customers, valued heavily.
Not worth building Slack—value accrues elsewhere; shows sophistication of long-evolved products.
Critics dismiss Anthropic's humanity risk warnings (jobs, fear) as fundraising/attention tactics, but
Anthropic creates headlines to communicate beliefs about potential AI futures, even if not guaranteed, to help people understand and avoid downsides. Building AI responsibly is preferred over staying out and hoping nothing bad happens.
Three key points:
- View progress through an exponential lens rather than linear: current state suggests limited improvement in 2-3 years linearly, but exponentials mean much happens sooner.
- Discuss downsides to avoid them and push upsides; most at Anthropic are optimists but recognize good outcomes are not guaranteed. Not enough productive discussion from those with real knowhow.
- Internally believe risks more strongly than stated externally; public statements are sometimes softer versions.
To influence the race to the top, must participate as a leading player; sideline shouting has no impact. Staying in the game with commercial success allows pushing principles and influencing others.
Biggest failure: founding a company, raising a couple million, having 7-10 employees, then shutting it down after three years. Focused on quantifying mental health for early predictors of generalized anxiety and major depression. In early 20s with no idea what they were doing.
Advice for struggling founders: send monthly investor updates even when things are bad—it's hard but keeps people in the loop and avoids surprises. Shutting down was brutal; tied to identity, took years to recover.
Positives: gained PM skills, product experience, cold emailing—essential for current path. Failures are a long game; grateful now as it led here. Big failures don't end careers; progressed to Masterclass, Mercury, Anthropic.
In early 2022, suffered traumatic brain injury from a kick to the head during normal Muay Thai sparring (after years of MMA/Muay Thai without issues). Off work 9 months; first 2-3 months brutal—wife handled everything including texting friends. Couldn't walk comfortably for ~6 months, listen to music >20 seconds without nausea, or look at screens.
Recovery: slowly increase tolerance, expose gradually without overpushing to avoid setbacks. Discussed with wife if he'd ever work again.
Mid-2023, after guest post, reinjured (bag hit head getting off plane)—off work 2-3 months, one month into Mercury job. Still not 100% healed; elevated reinjury risk until fully healed. Manages dizziness, headaches.
Transformative: best thing that happened.
- Habits: no alcohol, no caffeine; forced breaks even on craziest days (e.g., model launches) in office meditation area.
- Annual meditation retreats (e.g., Spirit Rock—changed his life; one upcoming).
- Meditation builds awareness vs. reality space for choice, level head amid insanity, avoids unhealthy coping.
Freedom through constraints: forces focus, adaptation. Choice: resist reality and suffer, or accept emotionally while acting (diet, every action). True freedom: be content without getting what you want.
Books:
- Joy of Living by Yongey Mingyur Rinpoche: rethink life experience, practical tactics.
- Awareness by Anthony de Mello: similar from different angle.
- Thinking in Bets by Annie Duke: quantify likelihoods (e.g., % chance task completes on time)—tactically helpful for product.
Favorite recent movie/TV: Marty Supreme (insane, ridiculous, loved it). No recent TV; Olympics if counts.
Loved product: Maruhachi Shinsui Pro pillow (Japan hotel discovery; beads allow unconscious height adjustment for back/side sleeping, relieves neck/upper trap pain). Ships to US via Amazon Japan.
Life mottos:
- "She'll be right" (Aussie saying: dismisses severity, assumes it'll be fine).
- "Sometimes you just have to go for it" (from Discord CMO Eros Resmini, adviser during founding).
New hobby: Tight on time; prioritizes work, body function, key relationships. Sports—football (Wolverines fan after wife's Michigan Big House game; 49ers fan). Watching, not playing.
Find Amal on LinkedIn. Anthropic jobs: growth engineering, growth product, growth design roles open—apply via jobs page.
Be helpful: try products, give harsh feedback; refer great people for team.
Keep Lenny's Podcast in your library
Save the videos and channels worth coming back to, and find them again in one place.
