long-form-interview· Amol Avasare

Anthropic's $1B to $19B growth run: how Claude became the fastest-growing AI product in history

At a product whose value is doubling on an exponential, growth-team choices invert: larger bets over micro-optimizations, quality and friction over speed-to-value, and human judgment concentrated on cross-functional alignment while AI absorbs the experimentation loop.

anthropicclaudegrowthailennys-podcastproduct-management92% confidence

Why this is in the corpus

First-hand operator view of the fastest ARR run in software history (≈$1B → $19B in 14 months) from the head of growth, with concrete plays around activation friction, agentic automation of growth experimentation (CASH), and a new PM/engineer division of labor under an AI-accelerated org.

Summary for skimmers

Amol Avasare, head of growth at Anthropic, explains why growing a product riding an exponential requires inverting classic growth playbooks: 70% of his time is firefighting "success disasters", onboarding deliberately adds friction to personalize recommendations, bets skew toward large swings, and a skeleton crew of engineers is being deputized as mini-PMs for projects under two weeks. He details CASH, an in-house system where Claude proposes, builds, tests and analyses growth experiments across four stages; and several agentic workflows he runs personally — a morning Hex-chart monitor, a weekly Slack-MCP misalignment scan, and a manager-persona coaching loop. He argues AI-first growth teams should be comfortable leaving money on the table for brand and safety, and frames Anthropic's PBC structure and early decision not to ship a consumer chatbot as the same principle applied to existential stakes.

Briefing

What survives the editorial filter

This page should feel like a smart colleague already listened for you and left only the operating logic worth keeping. Not everything said in the episode makes it through.

Trust signal

direct_practitioner_account

Guest type: practitioner.

Best used for

Amol Avasare (head of growth, Anthropic) on running growth at the fastest-scaling AI product in history — CASH agentic growth loop, two-week engineer-PM rule, adding friction deliberately, and why AI-first teams should skew toward bigger bets.

Hold lightly

No explicit downgrade reason stored yet for this episode.

Principles

Durable claims that survive beyond the speaker's biography — each with explicit limits, transferability judgment, and evidence.

Principle

Freedom through constraints

Constraints remove choice overload and compress an organization toward the few bets where it can win; treating constraints as a gift, not an obstacle, is often the leverage point.

Amol applies this at both company and personal scale. Company: Anthropic had neither Meta/Google cash flow nor OpenAI first-mover advantage, so they had to pick a very narrow focus for a generalisable technology. Personal: post-brain-injury, he had two choices — resist reality or accept the constraint and make happiness not depend on getting what you want. Dual applicability is part of the claim.

Use when: Under-resourced startups staring at a huge adjacent competitor; operators navigating personal setbacks.
Skip when: Well-resourced incumbents where the failure mode is complacency, not scatter.

When constraints hit, stop trying to break them and start using them as a forcing function — the narrow path they leave open is often the right one.

Historically we were very much like the smallest, least well-funded player in this space... you just have to really pick a very narrow focus and even for a very generalizable technology to maximize your chances of getting to escape velocity.Amol Avasare
The true freedom in life is learning how to be content when you don't get what you want.Amol Avasare (citing meditation teacher)

Durability: Ancient idea (stoicism) but rederived via concrete AI-lab strategy; endures.

Principle

Be comfortable leaving money on the table

Growth teams that squeeze the last dollar erode the things that actually compound; durable growth requires explicit willingness to leave measurable short-term upside unclaimed.

Amol pairs this with a two-bucket triage for controversial tests. Bucket 1: results don't matter because we wouldn't ship it anyway (brand/safety red lines) — don't run it. Bucket 2: uncomfortable but not a red line — run it, but require higher conviction / higher return to justify the cringe factor. He frames this as the same principle a founder should use when raising money — don't squeeze, you want people back next time.

Use when: Any growth or pricing team at a brand-sensitive or safety-sensitive product.
Skip when: Dying businesses where next-quarter cash is existential — the long-term argument doesn't apply.

Build explicit "don't test this" and "require higher ROI to justify this" tiers into your experimentation governance.

We are very comfortable foregoing metric impact in order to prioritize safety, in order to protect our brand, in order to hold a high quality bar.Amol Avasare
If you're a founder raising money, you're just trying to squeeze that last dollar. Like you don't, you don't wanna do that because you want people to come back next time.Amol Avasare

Durability: Durable operating principle; pairs with Anthropic's PBC structure and safety posture.

Principle

The right friction increases conversion

Friction that helps a user identify why the product is for them almost always improves activation and monetization, even though it lengthens the flow.

Amol has seen this at Masterclass (pre-purchase quiz), Calm (quiz in purchase flow), Mercury (splitting a 5-6 field screen into two screens to reduce cognitive load), and Anthropic (onboarding asks who you are and your interest areas so Claude can recommend products). Critics of the flow say "you have so much friction" — the data says it's working. The distinction is annoying friction (cut) vs. friction that surfaces the right product for the right user (add).

Use when: Activation flows where user intent is varied and the product has multiple surface areas.
Skip when: Single-purpose tools with one dominant job-to-be-done where personalization overhead is dead weight.

Stop optimizing for shortest flow. Optimize for the flow where the user leaves believing the product is for them — that usually means MORE questions, not fewer.

Adding friction and adding the right steps leads to higher conversion and higher funnel completion.Amol Avasare
A number of people look at the flow and they're like, you have so much friction, it's such a long flow. And I'm like, yeah, we have the data, we're kind of happy with how that's performing.Amol Avasare

Durability: Timeless principle but sits in mild tension with "reduce time-to-value" dogma.

Principle

Quality drives growth

In regulated or high-stakes onboarding, investing in craft and quality at the entry point outperforms conventional conversion-rate tuning.

At Mercury, the growth team suspended metric targets for a quarter to fix quality in the banking onboarding flow (beneficial-ownership details, back-navigation between address fields, complex entity paths). Amol calls it "until I joined [Anthropic], the single most impactful quarter I've ever had as a growth PM." He generalises: quality drives growth, and the best products operate this way.

Use when: Regulated, high-friction entry points (fintech/healthtech, enterprise signup, KYC flows).
Skip when: Low-stakes consumer funnels where the floor is already high and marginal craft gains are invisible.

If you are the growth team owning conversion, one of the highest-return bets is a dedicated craft-quality pass on the very first flow — especially when the entry experience is regulated or complex.

We, we said, forget metrics, forget growth, forget everything else. As the growth team on conversion, we're gonna spend a whole quarter fixing quality in this flow.Amol Avasare
Probably until I joined here, the single most impactful quarter that I've ever had as a growth PM in terms of the impact that it had.Amol Avasare

Durability: Ties to a durable principle about first-experience quality; will outlast any specific funnel design.

Frameworks

Reusable systems and operating models — including when they help and when they break.

Framework

CASH — Claude Accelerates Sustainable Hypergrowth: the 4-stage agentic growth loop

Growth experimentation can be decomposed into four evals: opportunity identification, feature build, quality/brand testing, and post-ship analysis — each is independently hill-climbable with an agent, with a human only truly required for cross-functional stakeholder management.

Led by Alex Kamis on Anthropic's growth platform team. Only viable once Opus 4.5/4.6; previously models were not capable enough. Currently scope is copy changes and minor UI tweaks; win rate is at a "junior PM with 2-3 years" level — not senior-PM yet. A brand skill containing guidelines and do's/don'ts reduces human review over time. Human in the loop currently approves; the bottleneck that won't go away soon is cross-functional coordination.

  1. Stage 1: Identify opportunities (agent scans past trends and current data for experiment ideas)
  2. Stage 2: Build the feature/test variant
  3. Stage 3: Quality + brand gate (skill containing brand guidelines, do's/don'ts)
  4. Stage 4: Ship + analyse + extract learnings
  5. (Human) Cross-functional stakeholder management
Use when: Growth teams with high-volume, low-risk experiment surface (copy, UI tweaks, upsell placements) and mature eval infrastructure.
Skip when: Teams without a mature brand-guidelines artifact, or where experiments have large safety/legal blast radius.

Decompose growth-experiment automation into four eval targets — don't try to automate "running experiments" as one monolith.

There's sort of four parts to it. One is identifying opportunities... second is then building the actual feature... third is testing and ensuring that it meets your quality bar and your brand bar. And then fourth is then once you've actually shipped the thing, analyzing the data, gathering the learnings.Amol Avasare
It's delivering results... you can push play with it and it ultimately prints money.Amol Avasare

Durability: Time-sensitive in detail (tied to 2025-2026 model generation) but the four-stage decomposition is durable.

Framework

Two-week rule: engineers become mini-PMs for short projects

In AI-accelerated orgs where engineering is 2-3x faster but PM/design are unchanged, the right response is to deputize product-minded engineers as mini-PMs for short projects — freeing PMs to focus on larger bets.

Rule is deliberately coarse: "use your head — if it's a one-week thing but extremely controversial, the PM should probably still drive it." Requires hiring product-minded engineers, which Anthropic already does. Paired insight: PMs in this world should spend less time shipping the 21st feature and more on "getting the why and what 5% better" — which is the high-leverage move when you've got 20 engineers behind you.

  1. ≤2 weeks engineering → engineer owns PM duties (security/legal/stakeholder coordination); PM advises
  2. >2 weeks → PM squarely accountable; delegates execution but owns outcomes
  3. Override: controversial short projects still go to PM
Use when: Larger orgs that have already shifted to AI-accelerated engineering but haven't grown PM headcount proportionally.
Skip when: Early-stage startups where everyone is doing everything; solo-PM teams without engineers who can handle stakeholder work.

Pick a time threshold, define clearly who owns the PM job on each side, and hire product-minded engineers so the hand-off works.

If a project is less than is two weeks of engineering time or less, then the engineer is on the hook to effectively be the PM for that.Amol Avasare
It's not like fully clean cut. It's like use your head — if this is a one week thing, but it's extremely controversial, the PM should probably still drive it.Amol Avasare

Durability: Durable response to a durable imbalance (engineer throughput > PM throughput in AI-accelerated orgs).

Signals

What appears to be shifting, for whom it matters, and what happens if you ignore it.

Signal

Capability overhang is the binding product-side constraint in AI

The bottleneck to realised AI value has shifted from the model layer to the product/activation layer — even internal users struggle to update their mental models when a new version ships.

Amol describes the treadmill: Opus 4 unlocks capabilities → team runs tests and builds on-ramps → new model ships → learnings are now irrelevant. He generalises: for most users, if their instinct is to come to a powerful model and ask "what's the weather in SF", capability overhang is the default state. Activation importance has gotten "exponentially higher" as a result.

Use when: AI-native product teams planning 2026 activation and onboarding investment.
Skip when: Non-AI products.

The competitive edge in AI products is now activation craft, not model quality — staff accordingly.

One of the biggest problems in the industry is capability overhang, where the models are just getting better so quickly... the real challenge is on the product side of how do we start to diffuse those benefits to people.Amol Avasare
If people's instinct is to come there and be like, hey, what's the weather in SF, then they're not gonna get the most out of the product.Amol Avasare

Durability: Current signal; will stay relevant until models stop improving or activation catches up.

Opportunities

Only included where there is a buyer, a real wedge, and a plausible revenue path — not vague idea theater.

Opportunity

AI-driven cross-functional misalignment detection is an unlock

A new class of agentic product — "find me where teams are about to misalign" — is now feasible because models can traverse Slack+doc context and summarise at supervisor quality, and this addresses the real bottleneck at scaled companies.

Scott, who leads Anthropic's enterprise team, uses this workflow and found major areas of misalignment that would have caused teams to spin their wheels. Amol runs it weekly himself. He explicitly ties the opportunity size to shipping velocity at big companies being "constrained at often by all the cross-functional coordination." Says six months ago this wasn't possible.

Use when: Teams building enterprise AI products, internal tools, or agentic workflows.
Skip when: Small orgs where misalignment is solved by a 10-person standup.

There's a standalone product category hiding here — misalignment-detection-as-a-service for scaled orgs.

Go and find me areas of potential misalignment right now. And it does a really, really good job. So this is something that I've scheduled runs every week.Amol Avasare
Six months ago that wasn't possible and I'm like, shit, like six months from now, what is going to be possible there.Amol Avasare

Durability: Time-sensitive in mechanism (depends on current-gen model context handling) but durable as a product category pointer.

Lessons still worth keeping

Useful takeaways that did not fully clear the bar for durable principle status.

Lesson

If the product rides an exponential, bias growth work toward larger bets

Traditional growth logic (compounding small wins) underweights large bets for AI-first products because the opportunity surface itself is growing exponentially; the optimal portfolio shifts toward swings that position for the future surface.

Amol contrasts grocery-delivery (product value grows 30-50% in two years) vs. AI-first products (100-1000x). At a traditional company, small optimizations capture a meaningful share of the modest future; at Anthropic, small optimizations still matter — "the compounding value is not immaterial" — but you have to take much larger core-product bets too. The Chrome extension is his example: a research-heavy swing with no competitors, built because the team was bullish, that now underpins Cowork and Claude Code use cases.

Use when: AI-first products where the model capability curve is the primary driver of value.
Skip when: Products where AI is a side feature and the core value is not on an exponential.

If your product value is on an exponential, flip your growth portfolio toward larger bets — a 70/30 traditional mix underweights the future.

Linear charts are just not cool. Like no one cares about linear charts. Everything is log linear.Amol Avasare
The product value that we will deliver in two years time is probably like a thousand x what it is today.Amol Avasare

Durability: Durable as a conditional principle; the specific thresholds shift with model progress.

The Plays

Try these this week

Verb-first executable actions — each one tied to a stated outcome in the episode.

Cold-email-your-way-into-a-growth-role at a hot company with no growth team

I just sent Mike Krieger a cold email. He was the chief product officer.
Amol Avasare
Single target: 2-6 weeks of sequenced follow-up per
  1. 1

    Pick a company you're already a power user of and that has no posted growth role

    Amol was a heavy Claude user before emailing; conviction and signal come from use, not research.

  2. 2

    Find the CPO's or relevant exec's PERSONAL email, not work or LinkedIn

    Everyone else is emailing LinkedIn and work addresses. Use public records, mutual connections, or paid email-finders to get a channel with lower competition.

  3. 3

    Write a tested high-open-rate subject line

    Amol keeps his exact line private; the point is: A/B test your subject line until you have a repeatable high open rate before spending it on a target that matters.

  4. 4

    Send a very short message: who you are, why this company needs a growth team, why you're a good fit, explicit ask to chat

    Amol's pitch paraphrase: Love what you do, love the product; I think you badly need a growth team; want to chat?

  5. 5

    Follow up as many times as it takes — stop only when they tell you to stop

    Amol: if I really care about it, I keep reaching out until they tell me, please stop.

Scripts

pitch

Hey [name] — love what you guys do, love the product. I think you guys badly need a growth team. Want to chat? Here's who I am: [one line]. Here's why I'd be a good fit: [one line].

Before you start

  • · Genuine power-user product conviction
  • · A high-open-rate subject-line template you've already tested
  • · A channel into personal (not work/LinkedIn) email
  • · Willingness to follow up many times
career-movescold-outreachhiring-marketearly-stagegrowth-stage

Manager-persona coaching loop: "As [manager], what feedback do you have?"

I say, Hey, based on what you know of Army — both publicly (she's written extensively about product) and then internally, and then our discussions — what feedback do you have for me as Army? And I get that every week.
Amol Avasare
Weekly; permanent rolling per
  1. 1

    Build a manager-persona context pack

    Collect your manager's public writing, podcasts, and (if permitted) their internal Slack posts and 1:1 transcripts with you. Put them in a project or skill.

  2. 2

    Connect relevant MCPs for your own activity

    Slack MCP so Claude can see what you did this week; doc/calendar MCP optional but improves signal.

  3. 3

    Schedule a weekly task

    Prompt: 'Based on what you know of [manager] publicly, internally, and from our past 1:1s, and based on everything I've done or not done this week, what feedback would [manager] give me?'

  4. 4

    Treat the output like a drunk coach, not an oracle

    Keep the 'holy shit, glad I caught this' items; discard the rest. Over time, refine the persona pack based on what feedback was actually useful.

Scripts

prompt

Based on what you know of [name] publicly (their writing about product and leadership), internally (their notebook-channel posts and Slack), and from our past 1:1 transcripts, and based on everything I've done or not done this week (pulled from Slack, calendar, and docs), what feedback would [name] give me as my manager? Be specific and include evidence.

Before you start

  • · Access to manager's writing (public minimum)
  • · Slack MCP and optionally doc/calendar MCPs
  • · A persona document or project Claude can reference
self-coachingfeedbackcareer-developmentgrowth-stagescalehyper-scale

Weekly Slack-MCP misalignment scan

Go and find me areas of potential misalignment right now. And it does a really, really good job. So this is something that I've scheduled runs every week.
Amol Avasare
Weekly; 4-6 week calibration before trusting it unsupervised per
  1. 1

    Get Slack MCP connected on Cowork

    Depending on org size you may need team/enterprise admin permissions to enable the MCP at the workspace level.

  2. 2

    Maintain a short list of the 3-7 projects you are currently driving

    The agent needs this context to scope which channels/threads to read — without it, signal-to-noise tanks.

  3. 3

    Create a weekly scheduled task

    Prompt Claude: 'Here are the projects I'm driving. Look across Slack this week for potential misalignment between teams or leaders on these projects. Surface specific threads, the disagreement, and who is on each side.'

  4. 4

    Triage the report in <=15 minutes on Monday

    For each flagged item: resolve directly, escalate, or mark as false-positive to improve the prompt over time.

Scripts

prompt

Using the Slack MCP, look across Slack for the past week. The projects I'm currently driving are: [list]. Find me areas of potential misalignment — threads where teams or leaders appear to disagree, assume different outcomes, or are solving the same problem in conflicting ways. For each, include a link, the disagreement in one line, and who is on each side.

Before you start

  • · Slack MCP connection on Cowork
  • · Workspace admin willingness to enable MCP scope
  • · A maintained project list in the prompt
cross-functional-coordinationmisalignment-detectiongrowth-stagescalehyper-scale

Daily metrics-anomaly agent on Hex + Chrome extension + MCP

I personally have like a Cowork that runs on a schedule and looks at sort of 20, 25 different charts every morning... Cowork will tell me, okay, here are the things that you should pay attention to. Here's what is concerning and here are just some interesting insights.
Amol Avasare
Daily; 2-4 week calibration period before trusting the signal per
  1. 1

    Identify the 20-25 Hex (or equivalent BI) dashboards you currently check by hand each week

    Pick charts that matter at your altitude but you don't have time to eyeball daily — exactly the long-tail set.

  2. 2

    Download the Claude (Cowork) desktop app and install the Chrome extension

    Chrome extension is how Claude reads the dashboards; MCP connectors handle structured sources.

  3. 3

    Create a scheduled task in Cowork that runs each morning

    Give it the chart links and instruct it to return (a) what's concerning, (b) what's interesting, (c) a plain summary for the charts you don't need to eyeball.

  4. 4

    Calibrate for 2-4 weeks

    Track how often Claude's 'concerning' items were real vs false-positive; tune the prompt until false-positive and false-negative rates satisfy you. Only then trust the long-tail coverage.

Scripts

instruction

Every morning at 8am, visit these 20-25 dashboard URLs. For each, report: (1) any metric concerning anomaly vs prior 7 days, (2) any interesting pattern worth my attention, (3) a one-line summary. Return as a prioritised list.

Before you start

  • · Cowork desktop app + Chrome extension
  • · MCP connectors for any non-web data sources
  • · A stable set of dashboards whose URLs don't churn
growth-and-activationmetrics-monitoringai-opsgrowth-stagescalehyper-scale

Decision Moments

Actual decisions, real outcomes

Specific decisions narrated in the episode with their outcomes and transferable lessons.

Amol was a heavy Claude user. Anthropic had no growth team and no posted growth roles. He had no warm intro.

Did: Sent Mike Krieger (CPO) a cold email via a non-obvious personal channel with a pre-tested high-open-rate subject line and a one-line pitch that Anthropic needed a growth team; committed to following up until told to stop.Outcome: Mike responded. Amol was hired as (per Mike) the only PM hired from a cold email, just as the company was starting to think about a growth team.

High-conviction, channel-optimised cold outreach to the right person at the right moment beats applying to posted roles you qualify for on paper.

Part of an emerging decision pattern across multiple episodes

A growth engineer at Anthropic was bullish on a Chrome extension for Claude; no competitor was doing it; the case for near-term conversion impact was weak.

Did: Green-lit the Chrome extension as a large swing despite the research-heavy profile and speculative ROI, because the team's conviction was high and no one else was building it.Outcome: The Chrome extension became the underpinning for multiple use cases on Cowork and Claude Code.

At AI-first companies with exponential product-value curves, portfolio allocation should skew toward conviction-led large swings; waiting for small-optimization evidence underweights the future surface.

Part of an emerging decision pattern across multiple episodes

AI-accelerated engineering was producing 2-3x engineer throughput at Anthropic's growth org; PM and design ratios were unchanged and visibly squeezed.

Did: Instituted the two-week rule: if a project is <=2 weeks of engineering time, the engineer owns PM duties (security, legal, cross-functional coordination); if >2 weeks, the PM is squarely accountable. Override for controversial short projects.Outcome: Growth team of ~40 could sustain its shipping velocity without immediate proportional PM hires; engineers with product instincts saw their leverage multiply.

In AI-accelerated orgs, a simple time-threshold rule can redistribute PM overhead to where it actually pays back — and turn product-minded engineers into the highest-leverage hires.

Part of an emerging decision pattern across multiple episodes

At Mercury, the onboarding flow was a compliance-heavy maze (registered agent, legal, physical addresses; beneficial-ownership details). The rest of the product was high-craft; onboarding was not. The growth team owned conversion with metric targets.

Did: Declared a quarter-long craft pass: "forget metrics, forget growth" — the team just fixed quality in the onboarding flow field by field.Outcome: Significant uplift in onboarding started-to-completion; described by Amol as the single most impactful quarter of his growth-PM career to that point.

Quality is a first-order growth lever, not a hygiene item; a metric-free craft pass can outperform any conversion-targeted experiment cycle.

Part of an emerging decision pattern across multiple episodes

Corpus connection

Where this episode fits for retrieval

What kinds of decisions this briefing is best pulled into.

Primary decisions

  • growth-strategy
  • activation-onboarding
  • org-design

Temporal flag

partially dated