Original video in Chinese.
Key Takeaway
- Product iteration speed has to be extremely fast: the traditional pace of releasing one feature every six months is completely outdated. Anthropic uses a “research preview” + “Launch Room” channel, which lets them ship experimental features in as little as 1 day to 1 week, quickly get real feedback, and reduce internal pressure to commit, because model capabilities are improving so fast.
- There are only two modes for AI-native companies: align every process with the company mission, cut all unnecessary reviews and meetings, and give employees a very high degree of autonomy; either become this kind of ultra-minimal, high-trust organization, or get eliminated.
- The most important new skill for individuals to develop is “product taste”: when the cost of writing code approaches zero, what’s truly valuable is deciding what to build and how to build it best; at the same time, you have to find the exact balance between the model’s current capabilities and its future capabilities — you can’t be too conservative and waste potential, and you can’t over-predict and turn everything into an empty promise.
How do you build products in the AI era? What does a company in the AI era look like?
If you watch this Lenny episode all the way through, you’ll definitely have the answer.
In this episode, Lenny interviews Anthropic’s Head of Product, Cat Wu. She’s the one responsible for the well-known Claude Code and Cowork, two star products. That’s exactly why I strongly recommend everyone watch this.
No matter how much you hate Anthropic, the fact is that they are currently the most successful AI company. Their models are strong, and their products are world-class too.
So how exactly did Anthropic manage all of this?
Take product iteration as an example.
A traditional software company or internet company might need six months to ship a feature. But Anthropic can ship in one month, or even one day.
Because in the past, model capabilities improved slowly, so everyone could safely make long-term plans. But today, model capabilities are iterating too fast. If you still work at the old pace, say by waiting six months before launching, by then the model will already be much stronger, and your product will be behind.
Cat said that you shouldn’t wait until model capabilities are fully sufficient before starting to build products. Instead, do the opposite:
First, put the product framework in place. Even if it doesn’t work that well with the current model driving it, once the next generation of model capabilities catches up, your product will already be ready and can immediately deliver maximum value.
So when Anthropic launches new Claude Code features, they do it in the form of a “research preview.” This clearly tells users: these are early, experimental features. There may be issues. They may even get cut later.
The benefit of doing this is that you can ship in one or two weeks, quickly get real feedback, and at the same time reduce the pressure of internal commitments.
Inside Anthropic, there’s a channel called Launch Room. When engineers finish developing a feature, they just drop it into this room. Other people handle the rest, like someone writing the documentation. Then the feature can be launched the next day.
You have to understand that in traditional companies, it usually takes several weeks of review, and you also have to pull multiple teams together for alignment meetings, which is a huge hassle.
Cat said they eliminated every process that gets in the way of shipping, and every unnecessary process, because they simply aren’t needed.
They do retrospectives every week, and they have clear team principles. On top of that, everyone is aligned with the company’s mission. In other words, when they run into some tricky issue, they just ask themselves: does this align with the company mission? If it does, then do it. It’s that simple.
That’s why they can give employees such a high degree of autonomy, and why they don’t have the internal friction and political infighting of big companies.
Look at the companies around you, then look at Anthropic, and you’ll realize: they’re not even the same species! Anthropic is what you’d call an “AI-native company,” a breed that only exists in the AI era.
Honestly, when I see bosses in China shouting about AI Native all day long, it’s all just empty slogans. Once you look closer, it’s still the same old playbook. It’s really ridiculous.
It’s the same point I always make: AI is an elimination race. Countries, companies, and individuals all have no way to hide from it.
So for individuals, what advice does Cat have?
She mentioned two points, and I think they’re both very valuable.
First, the most important new skill is product taste.
The logic is simple: as the cost of writing code keeps falling, what becomes more valuable is deciding what to write.
People who build products need to pay very close attention and judge: among these options, which one is worth building? What’s the best way to build it?
This ability doesn’t belong to any single role, such as a product manager. It can come from people with any background.
That’s why, when hiring, Anthropic will even prioritize engineers with strong product taste. Many engineers on the team can directly see issues from user feedback and then build the feature themselves in a short time, with almost no need for product manager involvement.
Second, the hardest thing to get right is the balance between future model capabilities and current model capabilities.
As mentioned earlier, you first have to put the product framework in place. That’s a form of prediction.
But what if you predict too far ahead?
For example, if you plan everything directly according to future AGI standards, you’ll end up designing huge products that the current model simply can’t deliver, and in the end it all becomes an empty promise.
But being too conservative doesn’t work either. That would waste the capabilities the current model already has, and you wouldn’t fully squeeze out its potential.
So finding the balance between the future and the present is the hardest part.
In plain language, it means you have to be bold and move forward, but not take such a big step that you pull a muscle.
I believe what Cat said really resonates with anyone who has worked on AI products. I won’t say more here — go watch the original video. You’ll definitely get something out of it.
OK, that’s it for this episode. If you want to understand AI, want to become a super individual, and want to find like-minded people, come join our newtype community. See you next time!