Anthropic’s Cat Wu says that, sooner or later, AI will anticipate your wants earlier than you recognize what they’re

With the tech trade singularly targeted on AI fashions, Anthropic is having an exceptionally good 12 months.

The corporate could quickly pull forward of its principal competitor, because it appears to lift tens of billions of {dollars} in a funding spherical that might put its valuation at some $950 billion (OpenAI was valued at $854 billion in its March spherical), and enterprise clients more and more specific a prefererence for Claude over ChatGPT. A latest report confirmed Anthropic lately outpaced OpenAI amongst enterprise clients, quadrupling its market share since Might 2025.

Cat Wu, Anthropic’s head of product for Claude Code and Cowork, has been a key determine in that success. Since becoming a member of the corporate in August 2024, Wu has helped shepherd Claude via a crucial part, leveling it up from a purely informational chatbot to a coding device and past. Wu, who oversees the event of latest options, is steadily paired with Boris Cherny, a core member of Anthropic’s technical employees and the creator of Claude Code, main the pair to be characterized as Anthropic’s “Batman and Robin.”

Wu sat down with me eventually’s week’s second annual Code with Claude convention in San Francisco, the place she mentioned how she thinks about product technique, and the way she hopes the expertise of utilizing Claude will change sooner or later. 

This interview has been edited for size and readability.

While you’re product technique, how a lot of it’s reactive to your friends or your opponents? Do you concentrate on that in any respect?

The principle factor that we design for is staying on the exponential, so I believe, throughout our crew, we instill in everybody the lesson that AI will simply proceed to get higher. For us, we simply want to remain at this frontier. We don’t take into consideration opponents. I believe in the event you do take into consideration opponents, you find yourself being, like, perpetually two weeks, or like, a month behind how briskly you possibly can execute. And so it’s usually not one of the simplest ways to remain on the frontier.

Anthropic launched no less than six fashions final 12 months and has already launched virtually as many this 12 months. Do you anticipate this tempo of improvement to proceed?

Our hope is that it continues (laughing). I believe the fashions are nonetheless bettering at a really regular tempo, and so we should always be capable of preserve sharing these with our customers. I believe the deployments would possibly look a bit completely different—like how we dealt with Glasswing, however as a lot as doable, we wish this intelligence to profit as many individuals as doable, and it must be dealt with in a really secure manner, which is why we dealt with Glasswing [in the way that we did].

[Glasswing is an initiative that Anthropic launched in April that invited a small consortium of partner organizations — including companies like Amazon, Apple, CrowdStrike, and Microsoft — to gain access to its new cybersecurity model, Mythos. Unlike many of Anthropic’s other AI models, Mythos is not being given a general public release. The company has claimed that it fears the model — which is designed to scan codebases for software vulnerabilities — is too powerful, and could be weaponized by bad actors.]

You mentioned in a earlier interview that the way forward for work is mainly employees managing fleets of brokers. It looks like that would ultimately result in a scenario the place the brokers are higher on the job, or know the job, higher than the human.

I believe this can be very arduous to handle brokers if you cannot do the job your self. I believe the managers nonetheless must be specialists of their area. It is a new ability set that lots of people are going to should be taught, however managing brokers is definitely similar to being a supervisor of individuals, within the sense that it’s a must to perceive, like, why did the agent make this error? Did it misread my instruction? Was my request under-specified? It’s a must to have the power to debug it.

It does appear to be the long run aim is to chop down on crew measurement, although. As a result of when you have brokers doing a job, then you do not want an intern, proper?

Ideally, I believe the thought is that everybody can get much more performed. I believe that, for everybody’s job, there’s all the time this proportion of it that is actually tedious. For me, it’s responding to emails. I believe everybody has this a part of their life…So my hope is that it [the AI agents] truly does that, after which everybody has, like, all these cool issues that they’ll wish to construct [in their spare time].

What are you guys most enthusiastic about within the subsequent six months?

I believe the subsequent large factor is proactivity. Final 12 months we had been on this world of synchronous improvement. Proper now, individuals are shifting to routines, so like automating, for instance, responses to buyer assist tickets. And I believe the subsequent step is that Claude understands what you’re employed on, and simply units up a few of these automations for you.