An interesting interview with Dianne Na Penn came out — she leads the product team for Anthropic's research org. She joined the company in 2023 as its first technical product manager. Since then her team has had a hand in pretty much everything we now use in Claude, including Claude Code.
Here's the interesting part. Turns out experience at big tech companies can just as easily get in the way as help — especially when you're building AI products.
At Anthropic, Dianne says, the people who thrive are early-career managers and former engineers who worked directly with customers. The first group simply doesn't have corporate habits to unlearn. The second group understands exactly what users need and knows the technology from the inside.
She also likes hiring former startup founders — people who had to handle product, sales, customer support, and basically everything at once.
The company doesn't need people who are good at managing processes so much as people who can get something done without a ready-made process.
The first two months are among the hardest of your career.
That's roughly the warning new hires get. You're handed a hard, high-priority task right away, and no one's going to walk you through what to do.
Most internal discussions are open across the company. So instead of an endless string of onboarding meetings, new hires are told to read through the last two months of conversations, get up to speed on the context, and show up with their own point of view.
Arguing with your boss isn't just allowed. Sometimes it's required.
Dianne says managers coming from big corporations often find it hardest to stop treating a manager's opinion as an instruction.
Anthropic's research culture is closer to academia: even a newcomer can push back on a senior scientist's take. But first you need to understand why earlier attempts to solve the problem didn't work. Being able to argue here doesn't mean you get to be confidently wrong.
Annual planning? No thanks.
Anthropic used to run a classic annual OKR (Objectives and Key Results) system. They dropped it fast — forecasts went stale within a couple of weeks.
Now the team sets hypotheses three to six months out, and only commits to concrete plans a month at a time.
The most interesting bit: when building a product, you can't just think about what the model can do right now — you have to think about what it'll be able to do in three or six months.
For example, Anthropic discussed building an interface for fine-tuning models. But then the question came up: what if a customer spends months training their own version, and by the time it's done, the base model has gone through several generations and the fine-tuning isn't even needed anymore?
In other words, you can build a great product that's obsolete by the time it ships.
I think there's something more important in this interview than career advice.
For a long time, a manager's career was built around the ability to manage complexity: aligning stakeholders, planning, delegating responsibility, tracking execution. Big companies grew an entire management culture out of those skills.
But when technology changes faster than a single approval cycle takes to run, some of those skills turn into obstacles.
That doesn't mean experience stops mattering, of course. It's more that what counts as valuable experience shifts. Experience making independent calls starts to matter more than experience navigating corporate procedure.
And maybe the most interesting part: Anthropic isn't so much inventing a new kind of management job as it is going back to a pretty old idea — a person should be accountable for the outcome, not for the process used to get there.