From Zero to $19B ARR: Inside Anthropic's Growth Lead's Operating System

Source: Lenny's Podcast | Published: 2026-04-05T12:30:57Z

Anthropic's ARR skyrocketed from $1B to $19B in just 14 months — so fast that nobody bothers with linear charts anymore; the whole team demands log scale.

The Cold Email That Opened the Door

The way Amol Avasare joined Anthropic made him an outlier from day one. During onboarding, the company polled new hires on how they got in — referrals, website applications, recruiters. He was none of the above.

He was a power user of Claude, and while using it, a thought struck him: this product is incredible, and the company clearly doesn't have a growth team. So he cold-emailed Chief Product Officer Mike Krieger: love your product, think you desperately need a growth team, let's talk. He didn't expect a reply. Krieger wrote back and said he'd actually been thinking about building one.

Amol says he spent years refining the art of the cold email — subject lines that get opened, channels that avoid the noise of LinkedIn and work inboxes, body copy stripped to three things: who I am, why I'm the right fit, and we should talk. If you truly care, you follow up until the person says "please stop." Krieger says Amol is the only PM he's ever hired from a cold email.

A growth leader who landed the job using growth tactics — that's a job interview unto itself.

10x Annual Growth and "Success Disasters"

Anthropic's revenue trajectory doesn't look like corporate financials. It looks like a runaway exponential: 2023, from zero to $100 million. 2024, $100 million to $1 billion. 2025, $1 billion to roughly $10 billion. Then by February 2026, $19 billion. Amol says linear charts are "uncool" internally — everyone says: show me the log scale.

When he joined, the team was building revenue plans for 2025 — a baseline scenario and an aggressive one. CEO Dario Amodei said the aggressive scenario could go higher. Everyone thought it was insane. Then it happened. By the end of 2025, people said the law of large numbers would kick in and growth would slow. It didn't.

He spends about 70% of his time on what he calls "success disasters" — everything is going so well that other things start breaking. Every chart is green and pointing up and to the right, yet people on the team can be emotionally crushed. Acquisition, activation, monetization — every layer is blowing up simultaneously. The remaining 30% goes to more traditional, proactive growth work: choosing which products to bet on, long-term pricing strategy, timing the optimization of new products.

Why Bet Big at This Scale

At Anthropic's size, a 1% optimization translates to massive revenue. But Amol says they've flipped the allocation — where a traditional growth team might spend 60–70% of its energy on small- and medium-sized experiments, they do the opposite, putting 50–70% into big bets.

His logic is tied to exponentials. A food delivery app might offer users 30–50% more product value in two years, and small optimizations can capture a meaningful slice of that. But Anthropic's product value in two years could be 100 to 1,000 times what it is today — because each new model generation unlocks entirely new markets. Agentic coding didn't exist a year and a half ago; now its market is already bigger than the entire AI coding market was before.

On an exponential curve, small optimizations capture a fraction of current value. Big bets are how you reach the next order of magnitude. The Chrome extension was a growth team initiative — a deep AI product that one engineer on the team believed in strongly, no one else was building it, so they went for it. That kind of decision doesn't happen at traditional companies.

He added a caveat: this advice applies to companies whose core product value is AI-driven — Cursor, Lovable, and the like. If AI is just an add-on feature, the logic doesn't necessarily hold.

The Real Lever for Activation: Add Friction, Don't Remove It

Amol has validated the same insight across multiple companies: adding friction in the right places actually increases conversion.

At Masterclass, the purchase flow included a seemingly unnecessary questionnaire — what are you interested in? Most people's first reaction was "I'm here to buy something, why are you asking me this?" But the flow had been rigorously tested and was a significant revenue driver, because it helped users feel that "this product was made for me." A growth PM from his team later left for Calm, the meditation app — and their signup flow had a similar questionnaire. Not a coincidence.

At Mercury, the growth team spent an entire quarter doing one thing: fixing the quality of the bank account opening experience. No metrics-chasing, no conversion optimization — just smoothing out every form field, every jarring back-and-forth transition. The result was the single highest-impact quarter of his pre-Anthropic career.

At Anthropic, the signup flow asks users a series of questions — who you are, what domains you care about — then recommends products and features accordingly. Plenty of people see it and say there's too much friction, the flow is too long. His response: we have the data, and we're happy with the performance.

The principle: cut friction that doesn't help users understand "why this product is right for me," but if friction helps them find the right feature, don't hesitate — add it. These upfront signals also feed into downstream lifecycle marketing, churn recovery, and even lookalike audience targeting for ads.

Automating Growth Experiments with Claude

Anthropic has an internal project called CASH — Claude Accelerates Sustainable Hypergrowth. Amol says he didn't name it, and it's a little embarrassing, but the work is serious.

The team broke the growth experiment lifecycle into four stages to evaluate Claude's capabilities: identifying opportunities, building features, testing for quality and brand compliance, and analyzing results to extract insights. Each stage is scored independently, with weekly tracking of whether the model is improving at each stage and whether humans are spending less time.

They're still in small-scale testing, mostly copy changes and minor UI tweaks. He says the results are real — you press the start button and it generates revenue. But he offered a reality check: the win rate is roughly equivalent to a junior PM with two or three years of experience — not yet at a senior PM level. That said, this was completely impossible a few months ago. Before Opus 4.5 it didn't work; it only became reliable with Opus 4.6.

He mentioned one area where Claude still can't substitute for humans: cross-functional stakeholder alignment. Getting six people in a room to reach consensus — that might still be hard even after AGI arrives. His design lead Joel messaged him: "We'll have AGI, but getting six people aligned will still be impossible."

The Compression and Expansion of the PM Role

Claude Code has roughly doubled or tripled engineering output. PMs and designers have seen gains too, but smaller ones. The result is that a team that was originally five engineers, one PM, and one designer now has the effective output of 15–20 engineers — and PMs and designers are becoming bottlenecks.

Anthropic's growth team has a two-layer response. Layer one: hire more PMs — they're aggressively recruiting product, engineering, and design for growth. Layer two is more interesting: for engineering projects under two weeks, engineers take full ownership of the PM role — including communicating with legal, aligning with the security team, and coordinating cross-functional stakeholders. PMs participate only in an advisory capacity unless things go seriously off track. For projects longer than two weeks, PMs formally take over.

This leads to a corollary: engineers with product sense have skyrocketed in value — they're "unicorn"-tier talent. Conversely, PMs who can design are too. In an environment where AI accelerates everything, the premium on cross-disciplinary skills has surged.

A counterintuitive take: at large companies and scaled teams, the highest-leverage use of a PM's time may not be writing code and shipping PRs, but improving the team's judgment on "what to build" and "why" by 5%. When you're looking at the output of 20 engineers, the marginal value of contributing one more feature is far less than sharpening the precision of the overall direction.

Co-work in Daily Life: Using AI to Surface Hidden Team Disagreements

Every week, Amol has Co-work scan all project-related Slack channels he's in via the Slack MCP, looking for potential disagreements. He says the results are hit-or-miss — sometimes Claude flags something completely random, but other times it catches genuinely important signals. His colleague Scott, on the enterprise team, uncovered several major cross-team alignment issues that, if undetected, would have left two teams spinning their wheels in overlapping directions.

He also set up a daily scheduled task: Co-work automatically reviews 20–25 data charts every morning and tells him which ones need attention and which have anomalies by the time he starts his day. He says he still likes to personally review a few core charts — "I just enjoy watching curves go up and to the right" — but for the mid-to-long-tail metrics, Claude's monitoring has gotten increasingly reliable, with both false positives and missed signals declining steadily.

Administrative tasks he's almost entirely handed off: conference room bookings, initial email triage, BenPass reimbursements, Brex expenses — all handled by Co-work.

Writing Your Own Weekly Review from Your Manager's Perspective

Amol does something unusual: he has Claude simulate the perspective of his boss, Ami Vora, based on her public writing and internal Slack communication, and give him weekly feedback — "Based on what I did and didn't do this week, what would Ami say?"

He admits the current quality is like "an occasionally tipsy coach" — sometimes it says things that make you roll your eyes, but other times you're grateful you saw that feedback. He categorizes this type of use as "soft coaching" and believes the capability has been unlocked — it's just that the precision is still climbing.

He thinks this area will see significant improvement in six months, as the model's ability to handle long context and cross-source information is advancing rapidly.

Why Anthropic's Focus Works

From the outside, Anthropic's strategy looks unusually focused — going deep on B2B, going deep on coding. Co-founder Ben Mann wrote a document just months after the company's founding in 2021 arguing why they should go all-in on AI coding — at a time when nobody knew how big that market would become.

Coding's strategic value operates on two levels. Commercially, it's a massive and exponentially expanding market. But the deeper layer: the best coding model accelerates your own research team, which in turn accelerates the model iteration flywheel. Dario saw this clearly years ago.

Amol believes this focus also stems from constraint. Anthropic has long been the smallest, least-funded player in the space — no Meta- or Google-scale cash flow and distribution, no OpenAI first-mover advantage. "In many ways, the fact that we're here at all is a miracle." He invokes a principle he returns to constantly — constraint creates freedom: when your options are limited, the path becomes clear, and you're forced to concentrate on the direction most likely to reach escape velocity.

A little-known historical detail: Anthropic had its own chatbot before ChatGPT launched but chose not to release it for safety reasons — the team didn't want to trigger a global AI arms race. When ChatGPT launched and captured massive consumer attention, it naturally pulled OpenAI toward the consumer direction. If history had played out differently, the two companies' positioning might be completely reversed.

Safety Isn't Posturing — It's Built into the Corporate Legal Structure

From its founding, Anthropic chose to incorporate as a Public Benefit Corporation rather than a standard Delaware C-corp — meaning it's not legally obligated to maximize shareholder returns as its highest priority and can weigh public benefit in its decision-making.

Amol says he was skeptical when he joined — sounds great on paper, but do they actually walk the talk internally? He got his answer quickly after starting: "They might take it even more seriously internally than they do publicly." He's seen the company walk away from commercial upside multiple times for safety reasons.

He shared the growth team's framework for handling controversial experiments: Category one — experiments so controversial that they won't ship regardless of results, because they violate brand, values, or user trust. These simply don't get run. Category two — experiments that feel uncomfortable but don't cross a red line. These can be tested, but if the "ick factor" is high, the returns need to be correspondingly high to justify it. AI safety is firmly in category one. Non-negotiable.

He also offered a broader growth principle: be willing to leave money on the table. Don't extract every last cent, because you want users and partners to come back next time. In the short term, you'll sacrifice metrics. In the long run, every great product operates this way.

Notebook Channels: Built for Humans, and Now for Agents

Inside Anthropic, everyone has a Slack "notebook channel" — essentially a personal internal blog where you share what you're thinking about, what you're paying attention to, what you have opinions on. Employees can subscribe to anyone's channel, including those of the research team and leadership. Someone who disagreed with Dario at an all-hands would go to his notebook channel and publicly raise the objection, sparking a debate.

This transparency serves two functions. For humans, it helps a rapidly scaling organization internalize the principles and thinking patterns across teams — a new engineer who reads the growth lead writing "we should be willing to leave money on the table" will calibrate their own judgment accordingly. For AI, these channels are becoming context sources for Claude. The HR team has already started tagging certain documents: "Please confirm before editing — this is a key document that Claude references."

The company is increasingly deliberate about capturing institutional knowledge in structured ways — serving both human onboarding and the operation of various internal agents.

60–70% of Projects Skip the PRD

About 60–70% of projects on Anthropic's growth team have no product requirements document. Amol says he "hates documents" — his default is: if you can skip the doc and go straight to action, skip it. If you can go straight to a prototype, do that.

Small projects are handled entirely in Slack — a few messages back and forth, and a product-minded engineer can figure out on their own, "What about that audience segment?" For large projects, he still insists on one cross-functional kickoff meeting — pulling in legal, security, and relevant teams to spend 30 minutes aligning on everyone's concerns. The ROI on that meeting is enormous because it prevents weeks of downstream chaos.

When a PRD is genuinely needed, he dumps his thoughts into Co-work five minutes before the meeting, has it generate a draft following the format of all previous PRDs, then discusses it in the meeting. But he repeatedly emphasizes: documentation is a means, not an end. At this speed, every dispensable ceremony should be dispensed with.

A Failed Startup, a Brain Injury, and "Constraint Creates Freedom"

In his twenties, Amol founded a mental health quantification company. Three years, several million in funding, a team of 7–10 at peak — and ultimately, shutdown. Calling investors to tell them the money was gone and the vision wouldn't materialize — he says it was the most painful experience of his life at the time. But it was during that startup that he learned cold emailing, product thinking, and how to build from zero. Without that failure, his subsequent career path wouldn't exist.

In early 2022, he took a kick to the head during a Muay Thai training session. An ordinary sparring day, one hit that landed wrong, and his entire life collapsed. For the first two or three months, his wife handled everything for him except showering and using the bathroom — including replying to friends' text messages. Twenty seconds of music made him nauseous. Screens were completely off-limits. It took about six months before he could walk normally again, nine months before he returned to work. During that period, he and his wife discussed what they'd do if he could never work again.

In mid-2023, he published an article about the experience on Lenny's newsletter. A month later, in the course of everyday life — a bag bumped his head while getting off a plane — he was injured again. He had just joined Mercury a month earlier and had to tell his new company he needed to take three months off. Because the first injury hadn't fully healed, the risk of re-injury was elevated. To this day, he still hasn't fully recovered and occasionally experiences dizziness and headaches.

The core lesson he distilled: that recurring principle — constraint creates freedom. No alcohol, no coffee, mandatory rest breaks in the morning and afternoon — these aren't choices, they're hard constraints. Even on the most chaotic model launch days, he'll go sit in the office meditation area. He does at least one meditation retreat per year. He quoted a meditation teacher: true freedom is being at peace when you don't get what you want.

"She'll Be Right"

Amol's go-to life motto is an Australian colloquialism — "She'll be right." When things get hard, you say with understatement: it'll be fine. It's not avoidance — it's an instinct for calibrating emotional responses.

His other favorite is something Discord's CMO told him repeatedly during his startup days: "Sometimes in life you just have to go for it."

Cold-emailing Anthropic's CPO, returning to work after a brain injury, telling everyone "we can go higher" in the face of growth targets nobody thought possible — put those two sayings together, and you've got a pretty good picture of his operating system.