Dianne Penn, who leads product for research and labs at Anthropic, says execution and details matter far more than strategy, putting them at about 99 percent of what makes a strategy succeed.
Penn says Anthropic's research organisation plans in the medium to long term, setting key assumptions three to six months out and planning one month at a time, after an early annual OKR process failed because adoption assumptions were far too low.
Penn says Anthropic prioritises business and enterprise customers over consumer users, whose needs change quickly, and says PMs should bring user impact into training conversations without dictating training changes.
What it takes to survive as a product manager at Anthropic
Product management at a frontier lab means turning fast-moving research into compounding user value while assumptions keep breaking. Dianne Penn argues this rewards obsessive execution, short planning cycles and people who thrive in a punishing ramp-up. Anthropic's research roots shape the job, so PMs speak for users without steering the models.
Takeaways
Execution details decide whether strategy works. Across Credit Karma, Amazon and Anthropic, Penn learned that details account for about 99 percent of a strategy's success. Her path ran from bond trading to Alexa AI model work, where transformers were replacing classic classifiers before LLMs were common.
Plan monthly inside a six-to-twelve-month outlook. Penn's team sets key assumptions three to six months out and plans one month at a time. Early annual OKRs were dropped because usage and adoption assumptions proved far too low. Long goals like life sciences become monthly bets.
Products emerge when model and harness click. Claude Code grew from a sequence of experiments. API customers used models for autocomplete long before coding became agentic, and the terminal paradigm only took off once the product harness and the model's agentic loops worked together.
Anthropic favors enterprise and current models. The company weighed a fine-tuning API but worried customers would fall two to three model generations behind. It prioritizes business and enterprise customers over consumers, whose needs change quickly.
Hard ramp-ups reveal who will succeed. Penn tells new PMs the first two months should feel like some of the hardest of their career. People who rise to it succeed, while those citing reasons they could not do the work struggle. She usually knows within one to two months.
Questioning assumptions, even hierarchy, is expected. Most channels are public, and challenging assumptions reflects Anthropic's research-lab origins. Penn says PMs from large companies often expect that whatever a VP says flows down the chain, which does not fit this culture.
PMs bring user impact, not training orders. During training of new Claude versions, Penn's team holds regular stand-ups with research leads. Researchers think about users not yet here and second-order effects on society, so PMs represent user impact without dictating training changes.
How Penn's team hires, trains and sustains pace
Strong candidates often come from varied careers, and a record as the first PM at a startup is a favored profile. Some forward deployed engineers have moved into PM roles on her team.
Tap a name to go there · tap a line to see why they are linked
On this episode 2
Show 1
Every new hire gets a spin-up buddy, a tenured team member, to help with onboarding.
An annual bootcamp began in 2024 with four or five research PMs and is now open to all of Anthropic. Even sales staff attended and learned what evals and prompting are. Penn argues much of it is replicable.
Pods of two PMs, such as across reasoning and efficiency, share context so no one is the sole owner of an area. The team takes time off and regularly discusses Penn's document on working sustainably.