Inside Anthropic's Growth Team: Big Bets, Automated Experiments and Firefighting Success

Host Lenny Rachitsky opens with the scale of Anthropic's growth: from about $1 billion in annualized revenue at the start of 2025 to roughly $19 billion 14 months later. Avasare, the company's Head of Growth, got the job in an unusual way. As a heavy Claude user he thought the company obviously lacked a growth team, so he cold-emailed then-chief product officer Mike Krieger. There was no open growth role, but the company was just starting to think about one.

Avasare says he honed cold emailing as a founder. He keeps his high-open-rate subject line a secret, but shares the rest of the method. He finds a personal email address, since LinkedIn and work inboxes are saturated, keeps the message to who he is, why he fits and a request to chat, and follows up until the person asks him to stop. Krieger, he says, told him he was the only PM he had hired from a cold email.

Avasare gives most of the credit for Anthropic's growth to its research, inference and compute teams, and to products such as Claude Code, rather than to growth. He recounts the revenue arc as roughly zero to $100 million in 2023, to $1 billion in 2024, and to about $10 billion last year. When he joined, the company's aggressive planning case alarmed people, and the company then beat it. Internally, he jokes, linear charts are uncool and everything is on a log-linear scale.

His team covers the usual growth areas of acquisition, activation and monetization, including free-to-paid conversion and pricing. The difference is that about 70% of his time goes to success disasters, meaning things breaking because growth went so well. He compares it to the early days at Facebook, Uber or DoorDash. It can be emotionally tough even when every chart points up.

The remaining 30% is more proactive work. It includes deciding which products to put resources behind, what long-term pricing and packaging should look like as behavior shifts, and when to start optimizing the adoption funnel for a new product like Cowork. He also mentions an import-memory feature that eased the cold-start problem for new users.

Avasare calls activation a big challenge for AI products, because people often arrive with trivial questions and never find what the model can do. New models also keep changing what is possible. A flow built to explain what Opus 4 could do may be outdated by the time tests finish and the next model arrives. The durable lesson, he says, is to match the right feature to the right user.

At Mercury he led a quarter where the growth team ignored metrics and simply fixed the quality of a complex banking onboarding flow. He says it was the most impactful quarter he had had on a growth team until joining Anthropic, with a significant lift in completion. At Anthropic, onboarding asks users who they are and what interests them, then recommends products. Critics call it too long, but he says the data supports it.

His rule is to cut friction that adds no value but keep friction that helps users understand why the product is for them. He points to MasterClass's purchase quiz and a similar quiz at Calm as examples that were tested thoroughly. The answers also feed lifecycle messaging and look-alike ad targeting later. Everything should be tested, he says, but the idea that cutting all steps is always best usually fails.

The growth team has about 40 people, structured as horizontals for growth platform and monetization plus audience-focused pods such as B2B, Claude Code, knowledge workers and API. Avasare says it resembles a traditional growth team except for the mix of bets. A conventional team might spend 60-70% of its time on small and medium bets, and Anthropic flips this to roughly 70/30 or 50/50 in favor of larger swings.

His reasoning is the exponential, a theme he says Anthropic raises constantly. A grocery delivery app might offer 30-50% more product value in two years, so small optimizations capture much of the gain. For an AI-driven product he expects value in two years to be a hundred to a thousand times today's, so teams should take bigger bets. They still do small optimizations, because the compounding value matters.

As an example, the growth team built the Chrome extension that now underpins several use cases in Cowork and Claude Code, because one engineer was bullish and nobody else was doing it. He says this approach fits companies where AI is the core of the value, such as Lovable or Cursor, and probably not those with AI as a side feature. He also explains that multiple products require audience-focused pods with close ties to product teams, such as Claude Code's.

Avasare says the growth platform team, which includes growth engineer Alexey Komissarouk, started an effort called CASH, short for Claude Accelerates Sustainable Hypergrowth. It uses Claude to automate growth experimentation. It began only a couple of months before the interview and was not workable before Opus 4.5, but results with Opus 4.6 convinced the team it was heading in the right direction.

He splits the experiment lifecycle into four parts: identifying opportunities, building the feature, testing against quality and brand bars, and analyzing results after shipping. The team scores Claude on each part and tries to improve them. So far the work is mostly copy changes and minor UI tweaks, and he says it delivers results at a win rate he would expect from a junior PM two or three years in, though not yet a senior one.

Humans currently approve ideas, and brand is a concern, but Avasare expects review to shrink as skills encoding brand guidelines improve. The piece he says will not disappear soon is cross-functional stakeholder management. He quotes his head of design saying that even with AGI it will be hard to align six people in a room.

Avasare says engineers are getting the most leverage from AI right now, more than PMs and designers. A team of five engineers, one designer and one PM may behave like 15 to 20 engineers with Claude Code while PM and design headcount stays flat, so product and design are squeezed. His team is hiring growth PMs and may need many more, though the answer varies by company size and industry.

His team also hires product-minded engineers and applies a rule of thumb. If a project takes two weeks of engineering or less, the engineer acts as the PM, handling security, legal and stakeholders, with the PM advising. Larger projects stay under the PM's accountability. In his view this makes product-minded engineers dramatically more valuable.

On whether PMs should ship code, he says it helps for learning the tools and for expressing an idea through a prototype. But on a team with 20 engineers, a PM adds more by improving the why and the what by even 5% than by shipping the 21st feature. Smaller, resource-tight companies may need PMs to ship.

He says 60-80% of what Anthropic ships has no PRD and small changes are handled in Slack. Larger projects get a short cross-functional kickoff of about 30 minutes with legal, safeguards and others. When he does write a PRD, he uses a Claude skill and a project holding previous PRDs, and often prototypes instead.

One of Avasare's favorite uses is asking Claude, via the Slack connector in Cowork, to scan Slack for areas of potential misalignment on projects he cares about. He runs this weekly. A colleague leading the enterprise team found major misalignments this way that would have caused teams to duplicate work. Setup, he says, means connecting the Slack connector in the Cowork desktop app, which may need admin permission, and then asking Claude.

Cowork also checks 20 to 25 charts every morning through scheduled tasks, using the Chrome extension and connectors, and summarizes what needs attention. He lets it book meeting rooms, triage email and file reimbursements and expenses in Brex. Rachitsky imagines a strategy bot that watches metrics and the roadmap, and Avasare thinks something like it may be effective later this year.

He also uses Claude as a soft coach. It reviews what his direct reports did against team goals and transcripts and suggests feedback, and it can simulate his manager's perspective on what he did or did not do that week. He describes the quality as hit or miss, like a coach who is sometimes off but occasionally catches something important. He expects accuracy to keep improving.

Rachitsky observes that Anthropic has gone deep on a few things, such as B2B and coding. Avasare says the focus comes from leadership and has been there from the start. He mentions a 2021 document by a founder arguing for a focus on AI coding. The logic was both commercial and that the best coding models accelerate research, which Dario Amodei has seen clearly for years.

He adds that necessity played a role. Historically Anthropic was the smallest, least well-funded player, without the free cash flow or distribution of Meta or Google or the first-mover advantage of OpenAI. He calls this freedom through constraints, since limits free you from excess choice. He also notes that Anthropic had a chatbot before ChatGPT launched and chose not to release it for safety reasons, which may have kept it away from the consumer pull that followed.

Avasare says safety is the reason the company exists and is built into its structure as a public benefit corporation, which lets it legally avoid treating shareholder value as the overriding goal. He says Anthropic has repeatedly been willing to take a commercial hit for safety. For controversial growth tests he sorts them into two kinds. Some are off the table because they violate brand, customer or values lines, and others are uncomfortable but allowed if they produce a high enough return.

A common growth mistake, he says, is squeezing out every last dollar. His team is comfortable forgoing metric gains to protect safety, brand, quality and user experience, and he thinks the best products operate that way. He expects the stance on safety to become a competitive advantage as the stakes rise.

Asked how to thrive as AI reshapes jobs, Avasare's first piece of advice is to use the tools. People should use Claude Code and Cowork and recheck what each new model release can do, since something that failed before may work now. Doing so improves productivity and also builds product sense for AI products.

Second, he says to lean into where you have an unfair advantage and almost forget your weaknesses. He argues that interdisciplinary people are increasingly valuable, such as a product-minded engineer or a PM who can design. His own mix of founder, investment banking, sales and growth experience gives him an edge in certain situations.

Third, be adaptable. He says that coming to Anthropic means discarding perhaps 50, 60 or 70% of how you worked before, and clinging to old playbooks creates friction.

Avasare calls the culture and people the most defensible thing Anthropic has. He says it is truly mission-driven and that he has not met anyone checked out, unlike every other job he has had. Leadership is transparent, and employees keep notebook channels in Slack, a Twitter-like internal feed of thoughts that anyone can follow or challenge. Employees can disagree publicly with leadership, and that has sparked big debates.

He says the notebook channels also scale leadership's beliefs as the company grows quickly, so new hires see principles like the importance of leaving money on the table. They also give Claude context about how the company thinks. He describes the talent density as like playing for Real Madrid.

Rachitsky asks why Anthropic warns about AI's dangers. Avasare says the company thinks exponentially, believes in the risks more strongly than it says externally, and wants to stay in the game to push a race to the top, because voices on the sidelines have little influence. On the idea that SaaS tools will be replaced, he says Anthropic still uses Slack, Figma and Workday heavily and sees little value in building its own.

In his failure story, Avasare describes founding a mental health company that spent three years and raised a couple of million dollars, with seven to ten employees at its peak, before shutting down. It was very painful and took years to get over, but it taught him the skills that led to his product career. His advice to struggling founders is to keep sending monthly investor updates, especially in bad months, to avoid surprises.

Rachitsky then recalls how Avasare's guest post about a traumatic brain injury made him a better product manager. In early 2022 a kick during sparring left Avasare off work for nine months, unable to look at screens or listen to music for more than seconds. A second injury in mid-2023, about a month into his Mercury job, kept him off for a further stretch, and he says he is still not fully healed.

He says the experience helped him in the long run. He avoids alcohol and caffeine, takes short breaks even on the busiest days, and goes on a meditation retreat at least once a year. He credits meditation with giving him space between awareness and the chaos around him. His takeaway is again freedom through constraints and learning to be content when you do not get what you want.