Inside PM at Anthropic: Who Makes It, and Who Doesn’t

Summary

Dianne Penn, who leads product for Anthropic’s research and labs teams, describes product management in a fast-moving research organization. The role combines long-term direction with hands-on execution: PMs need to understand users, work closely with researchers, test products themselves, and revisit assumptions as models evolve.

Strategy, planning, and product development

• Pair a long-term direction with short planning cycles. Anthropic’s research organization considers medium- and long-term goals, while Penn’s team plans month by month against assumptions about the next 3–6 months, informed by a broader 6–12-month view.
• Avoid planning rituals that quickly go stale. The team found that annual OKRs could become obsolete when usage and adoption differed from expectations. Monthly plans provide a way to adjust without losing sight of longer-term aims.
• Treat execution as the bulk of strategy. Penn estimates that “99%” of making a strategy work happens after the initial plan—through working through user flows, reviewing feedback, examining model outputs, and resolving product details.
• Develop products bottom-up through experiments. Teams test whether a technology can produce a useful experience before committing to a larger product. The key question is whether users experience enough “magic” for the idea to be viable.
• Build products that improve with the models. Anthropic’s product bootcamp focuses on products that “ride the exponential”—becoming more useful as model capabilities improve.
• Example: Claude Code. Its development was not a one-step shift to agentic coding. Users had already used models for autocomplete; a sequence of product, environment, and model improvements made terminal-based agentic loops practical.
• Evaluate investments against model progress. The team questioned whether a fine-tuning API was a worthwhile investment if rapidly improving core models could make the work obsolete—or leave a deployed solution several model generations behind—before it reached customers.

What Anthropic looks for in PMs

• Adaptability and initiative matter more than following a standard career ladder. Successful hires have come from varied backgrounds, including research, startups, and customer-facing roles. Early PMs at startups can be especially effective because they have often covered customer success, IT, business development, and product.
• The first two months are deliberately demanding. Penn says new PMs should expect the first two months to feel among the toughest of their careers. Those who adapt quickly can enter an “acceleration phase”; those who focus on why they cannot act tend to struggle.
• Move quickly during onboarding. New PMs are expected to understand users, write an evaluation, and produce a roadmap within roughly two to three weeks.
• Challenge assumptions constructively. PMs should not assume a senior leader’s view is settled. They should do the work, explain their reasoning, and question ideas with curiosity and respect—not simply declare someone wrong.
• Demonstrate product judgment through hands-on work. Rather than spending all their time writing requirements documents, strong PMs test prototypes and new models, then share specific feedback in public channels.
• Unlearn excessive process. Penn sees autonomy and a “make it happen” mindset as particularly valuable, including among less-senior PMs. Experience overcoming organizational inertia can also be a strong signal.

Working with research and customers

• Balance current users with future users and broader impacts. Research priorities can include future use cases and societal considerations, not only the needs of today’s customer base.
• Prioritize user problems explicitly. In regular discussions with research leaders, PMs help classify issues as P0, P1, or P2, explaining the impact on users and whether a problem blocks adoption or migration.
• Consider whether a limitation will persist. Some issues may become less important as model intelligence improves. PMs can sometimes reduce user pain through product or interface choices while waiting for model improvements, rather than treating every complaint as a training priority.
• Build trust with researchers. PMs should explain the user impact and importance of a problem while recognizing that researchers determine what is technically feasible.
• Use customer-facing experience as a path into product. Advanced deployment engineers who move into PM roles have brought valuable knowledge of customers’ tools, workflows, and motivations. Penn notes that doing deep customer work and deep research work simultaneously is difficult, so teams need ways to combine those perspectives.

Team development and sustainable pace

• Run a recurring product bootcamp. Started in 2024 by a team of four or five, it initially shared research-PM practices—such as evaluating user-consented feedback, writing reviews, prompting, and creating evaluations—and later opened to the wider company. The team also shares training videos with new employees.
• Make company context accessible. Anthropic’s public-channel culture lets new hires review prior discussions. The advice is to read back through the conversation, form a view, and then ask informed questions.
• Pair new and experienced team members. Senior employees act as support partners for newer PMs. Small teams and responsibility across multiple areas are intended to encourage ownership and urgency.
• Build sustainability into team operations. Penn’s guidance includes taking time off after intense sprints, fully disconnecting on vacation, and using flexible time off when work allows.
• Use PM “pods” and copilots. Having two people share context on an area reduces single-person dependency, makes time off more restorative, and distributes knowledge. It also helps leaders scale by giving capable team members enough context to act without waiting for direction.

Metrics and timelines mentioned

• No revenue, margin, CAC, LTV, churn, or growth targets are provided.
• Planning horizons include monthly plans, 3–6-month assumptions, and a broader 6–12-month outlook.
• New-hire ramp: two to three weeks for initial user understanding, evaluation, and roadmap; the first two months are expected to be especially demanding.
• Training: an annual bootcamp, first run in 2024.

Presenters and source

• Dianne Penn — Anthropic product leader for research and labs. Her name appears as “Diane” in parts of the subtitles.
• Nakul — host of The Skip podcast. The subtitles also render his name as “Nikhil.”
• Source: The Skip podcast.