Contrary to the 'minimize steps to value' mantra, adding friction like user questionnaires to onboarding often boosts conversion. By asking users about their goals, you can personalize their experience, make them feel the product is for them, and guide them to the right features, improving funnel completion.
To manage the strain on product managers from hyper-productive engineering teams, Anthropic has a rule: if a project is two engineering weeks or less, the engineer is the PM. They are responsible for stakeholder management (security, legal, etc.), with the official PM acting only as an advisor.
The traditional Product Requirements Document (PRD) is too slow for a hypergrowth environment. Amol Avasare states that his growth team at Anthropic skips PRDs for ~70% of their work, preferring to kick off projects on Slack for smaller tasks and jump directly to prototyping for larger ideas.
Anthropic's initial position as the "smallest, least well-funded player" without the distribution of Google or first-mover advantage of OpenAI was a blessing in disguise. These constraints forced a laser focus on narrow areas like B2B and coding, preventing distraction and allowing them to achieve escape velocity.
Instead of being a generalist, the best way to stay valuable is by combining deep skills. For example, a PM who can also design, or an engineer who is highly product-minded, becomes a "unicorn" in an AI-augmented team. This interdisciplinary spike makes you far more valuable and less replaceable.
Anthropic is developing a system called "CASH" to automate growth work. It uses Claude to identify opportunities, build features (like copy changes), test them, and analyze results. The system is already delivering results comparable to a junior PM and is expected to handle increasingly complex experiments.
A powerful, non-obvious use for AI assistants is proactive stakeholder management. Amol Avasare runs a scheduled task for Claude to look across his Slack channels and projects to find potential areas of misalignment. This helps him surface and resolve issues before they derail projects.
To maintain culture during hypergrowth, Anthropic employees have "notebook channels" in Slack—like personal Twitter feeds for internal thoughts. Leaders use these to model behavior and scale their beliefs (e.g., "why we leave money on the table"), quickly aligning new hires without meetings or formal docs.
At rapidly scaling companies, the growth team's primary focus isn't just proactive optimization. Amol from Anthropic spends 70% of his time on "success disasters"—firefighting issues where extreme success in one area breaks another part of the system, from acquisition to monetization.
Anthropic's intense focus on AI for coding wasn't just a market strategy. The core belief, held since 2021, was that creating the best coding models would accelerate their internal researchers' work, creating a powerful flywheel that improves their foundational models faster than competitors.
Amol Avasare, a user of Claude, identified a need for a growth team at Anthropic and cold-emailed CPO Mike Krieger. He used a perfected, high-open-rate subject line and contacted Krieger's personal email to bypass noise, which led to the interview process and ultimately the job offer.
Amol Avasare uses Claude to generate weekly feedback from the perspective of his manager. He instructs the AI to analyze his manager's public writing and internal communications to create a model of her priorities and style, then asks it to evaluate his week's work and provide feedback as if it were her.
As AI tools dramatically increase engineering leverage (2-3x), the traditional 5-engineer, 1-PM, 1-designer team structure breaks. The PM and designer become bottlenecks, struggling to manage what is effectively a 15-20 person engineering team's output, forcing a rethink of team ratios and roles.
For AI-first products, future value is exponentially greater (e.g., 1000x in 2 years). Therefore, Anthropic's growth team flips the typical 70/30 optimization/big-bet ratio, focusing on larger swings that unlock new markets because small optimizations can't capture the massive potential value created by model improvements.