Sau Claude Code, sản phẩm bùng nổ tiếp theo của Anthropic sẽ là gì?

Video title: Anthropic's hunt to find the next Claude Code

Video author: ACCESS Podcast

Translation: Peggy, BlockBeats

Editor's note: As large language models continue to advance and AI coding tools rapidly proliferate, the industry discussion is shifting from "can the model complete the task" to "how can model capabilities be organized into products, workflows, and commercial systems."

Over the past year, products like Claude Code, Codex, and Co-work have entered the workflows of developers and knowledge workers. AI is no longer just a chatbox that answers questions; it is becoming a production interface that can call tools, execute tasks, and verify results. But as the consensus that "agents will become the next software paradigm" solidifies, a more critical question emerges: who can first translate model capabilities into reusable, distributable, and scalable work systems?

This article is compiled from an interview with Mike Krieger on the ACCESS Podcast. Mike Krieger is the co-founder of Instagram and currently the Chief Product Officer at Anthropic, leading Anthropic Labs. His goal is to guide the team in exploring Anthropic's next frontier product directions beyond Claude Code.

Alex Heath (left) and Mike Krieger (right)

In this conversation, Mike Krieger doesn't simply discuss what Anthropic's next product will be. Instead, he deconstructs the AI product competition into a set of more fundamental structural questions: how do model capabilities enter real workflows, how do AI companies organize innovation internally, how do platform companies handle boundaries with ecosystem customers, and where will human judgment be repositioned in the production chain as AI execution becomes increasingly powerful?

First, the product form is shifting from "chat" to "tasks." Previously, large models mainly existed as dialog boxes – users input prompts, models generate responses. Now, Claude Code, Co-work, and Claude Design represent a different product logic: having AI persistently work towards a goal, calling tools, generating results, and performing verification in the process. This means the key for AI products is no longer just the quality of answers, but the ability to decompose tasks, maintain context continuity, call tools, and verify results. Whoever can encapsulate these capabilities into seamless workflows will be closer to the next generation of productivity gateways.

Second, organizational methods are shifting from "large team planning" to "small team experimentation." Anthropic Labs operates more like a startup unit embedded within a large company: starting with two or three people, bi-weekly reviews, using high-frequency feedback to decide whether to continue a project. In the past, innovation labs in large companies often suffered from long cycles, vague responsibilities, and delayed projects that were "good enough." Now, models have lowered the cost of building; what's truly scarce is judgment, taste, and decision-making speed. This means organizational efficiency in the AI era depends not just on the number of engineers, but on whether smaller teams can validate directions faster.

Third, the boundary between platforms and applications is being redrawn. Claude Code's success means Anthropic is no longer just a model provider but is also defining application forms itself. The controversy surrounding Claude Design and Figma shows that when a model company builds its own applications, it inevitably touches the interests of its clients and ecosystem partners. Previously, foundational model companies mostly provided underlying capabilities, leaving vertical applications like Cursor and Figma to handle user interfaces and scenario packaging. Now, model companies also need their own products to showcase an agent-first future. This means AI platform competition isn't just about APIs; it's also about product paradigm competition.

Fourth, the stronger the AI, the scarcer human judgment becomes. Mike repeatedly emphasizes that Claude can write code faster, generate prototypes, and execute tasks, but it cannot replace the hardest part of the journey from 0 to 1: asking the right questions, understanding real users, defining the product's North Star, and judging what is "right." Previously, execution capability was the main bottleneck in knowledge work. Now, execution is being accelerated by models, and human value is more concentrated on upfront judgment, creativity, relationship networks, and organizational ability. AI won't automatically eliminate difficult decisions; it will instead amplify wrong directions faster.

If this conversation were compressed into one judgment, it would be: After Claude Code, Anthropic isn't looking for a single blockbuster product, but a method to transform AI from model capability into a production system. In this sense, what this article discusses extends beyond Anthropic's next product roadmap to the structural shift of the entire AI industry from a "model race" to a "systems race."

The following is the original content (edited for readability):

TL;DR

· AI product competition has shifted from "stronger models" to "how to ground capabilities," essentially meaning large model companies are now competing for workflow gateways.

· Claude Code's significance extends beyond writing code; it proves agents can execute tasks persistently towards clear goals, transforming AI from a chat tool into a production system.

· Anthropic Labs' core value lies not in how many products it launches, but in rapidly validating what the model's next capabilities should be with small teams.

· Co-work represents Anthropic's attempt to extend the Claude Code methodology to non-programmers, essentially abstracting "programming capabilities" into workflow automation for ordinary people.

· OpenAI's Codex catching up means Claude's advantage is no longer just technical leadership, but depends on Anthropic's ability to integrate Claude Code, Co-work, and Claude.ai into a unified experience.

· Model companies building their own applications will intensify boundary conflicts with customers, but this is also their inevitable path to defining the next generation of AI product forms.

· The faster AI can execute, the more human value concentrates on upfront judgment, product taste, and problem definition, because wrong directions will also be amplified faster by AI.

· AI's impact on employment is not a problem any single company can solve; it will fundamentally force society to re-discuss skills reshaping, distribution mechanisms, and irreplaceable human capabilities.

Original Content

Alex Heath (Host): After Claude Code, what will be Anthropic's next big product? This week, we have Mike Krieger. He is the co-founder of Instagram and now leads the team working on "moonshot projects" at Anthropic.

Mike Krieger (Chief Product Officer, Anthropic):

One of my darkest days at Anthropic was naming it 3.5 v2. I can explain why we ended up with that name.

Alex Heath: Mike and I recorded this conversation in person during Anthropic's recent Claude Code conference in San Francisco. At that conference, Anthropic announced a new large-scale computing partnership with Elon Musk. So, are you guys going to space with Elon now?

Mike Krieger: Exactly. Yes, we are looking for new, even unexpected sources of computing power.

Alex Heath: We talked about what Mike is working on now, the intense competition between Anthropic and OpenAI, and what parts of human work Mike believes will remain important even as AI gets stronger.

This is Access.

Mike, great to see you here at the Claude Code conference in San Francisco. I was just thinking back to our last conversation. You had just taken over Labs, but it's been a few months now, right?

Mike Krieger: Yes, almost four months.

How Labs Works: Bi-Weekly Pivots, Validating Big Products with Small Teams

Alex Heath: Almost four months. I want to start here for those unfamiliar with Labs. It's a pretty unique organizational structure. We talked about it when I visited your office a few months ago. What exactly is Labs, and what is its mission within Anthropic?

Mike Krieger: Simply put, my understanding of Labs – this current version, I'd call it Labs v2. We can talk about what Labs v1 did and what Labs v2 aims to do later.

But I think Labs does two main things.

First, it's about closing the gap between Claude's theoretical capabilities and the everyday experience of ordinary people. That is, Claude can theoretically do many things, but how can these capabilities genuinely enter people's daily work and life? What products, prototypes, or projects do we need to build to demonstrate how to unlock more of this potential and minimize this gap?

Second, we act more like a "frontier scout team," judging which direction the model needs to evolve to meet the needs of different users.

So, a successful Labs project doesn't necessarily have to be released as a product. It could also be a prototype. We build it and find out: the model isn't good enough yet, temporarily unable to complete this task. So we put it aside, re-evaluate it when the next generation model is released, or turn it into an evaluation metric for future model development, and continue iterating.

Therefore, unlike product labs in pure product companies, where success might be measured by "did you launch a product," at Anthropic, Labs' value can manifest in other ways: it can influence Anthropic's future direction.

Alex Heath: Labs has certainly produced some hits, right? Claude Code is one, MCP is another. What else?

Mike Krieger: Agent Skills was another important project from Labs. I can also talk about a project that wasn't released but was very helpful for research: computer use, letting Claude use a computer.

I joined Anthropic in May 2024. Next week will be my second anniversary; we call it "antiversary" internally.

Alex Heath: Is it anniversary?

Mike Krieger: It's antiversary. Everything at Anthropic has to relate to ants. I was resistant at first. We don't say dogfood; we say antfood.

After I joined, we started building Labs. One of the earliest projects proposed was: why not try letting Claude use a computer?

Alex Heath: That's computer use.

Mike Krieger: Yes.

Alex Heath: What model era was that?

Mike Krieger: That was Claude Sonnet 3.5. It was also the first model I participated in launching. I started working on the launch in my third week. We often joke that Anthropic doesn't have an onboarding project; it just throws a really hard project at you. And I was directly involved in the launch in my third week.

Sonnet 3.5 is an interesting model because it was one of the first to really unlock some coding scenarios. Not quite full agentic coding, but you could see the beginnings.

So, we put Sonnet 3.5 in and built a computer use product around it. But it had many problems. It was too slow using the computer, not accurate enough, and its visual capabilities weren't good enough. It would see the screen, say "I need to click that button," and then click somewhere else.

But building this "not fully functional" testing framework was incredibly helpful. Later, when we got to Sonnet 3.5 v2 – we can discuss that naming later, it was truly one of my darkest days at Anthropic – we could just drop the new model into this framework and test it.

Later we tried 3.6; it still wasn't good enough, but showed some improvement. Then came 3.7. I remember that day very well. I was on a business trip in New York, meeting the New York team. Suddenly someone messaged me: we think that thing Labs built, the computer use project that had been sitting for nine months, is really starting to show signs of life on Sonnet 3.7. We think it's time to open up computer use as a capability for public discussion.

This took about nine months in total. Every few months, we'd drop the new model into the same testing framework. Even though Labs had temporarily shelved the project, it was still very useful because it became a test suite for evaluating the evolution of computer use capabilities in models.

Alex Heath: When you first joined Anthropic, you were Chief Product Officer. I remember thinking: Mike Krieger, the Instagram co-founder, a consumer product guy in my mind, why would he join an enterprise AI company?

Mike Krieger: Yes.

Alex Heath: We might have talked about this before. I thought it was a very interesting choice. In hindsight, it was the right choice. And the timing was very good.

I'm curious, you joined as CPO, overseeing the entire product line. But "AI product" as a concept is quite vague and changes fast. How did you transition to Labs about four or five months ago? I understand you're more of an IC now – an individual contributor? Do you still manage people?

Mike Krieger: I don't manage people now. We're just heading into the performance review cycle.

Alex Heath: So this is what you wanted, right? You're escaping writing performance reviews?

Mike Krieger: Exactly. I opened the system to see what reviews I needed to write and found out: I only need to write my own self-review and my manager's review.

Alex Heath: That's it?

Mike Krieger: That's it.

Alex Heath: Claude is writing all the performance reviews now.

Mike Krieger: Claude does help write some reviews, which is useful. It doesn't write everything for you, but at least it helps you remember: what did I actually do in the last six months?

I think companies go through different stages, and the match between those stages and what I'm truly passionate about varies.

When I joined, the product and engineering team was only about 30 people, maybe half and half. Of course, we had engineering teams working on research infrastructure, scalability, etc., but just looking at people actually building products, it was mainly Claude.ai and what we then called the API – we didn't even call it Claude Platform yet – probably only 30-35 people total. Very, very small.

It still felt a lot like an early-stage startup. Many things were still being defined. Like, "what this product actually is" was far from settled. Claude.ai back then didn't have Projects, Artifacts, or essentially anything beyond a list of conversations with Claude.

So joining Anthropic then felt a lot like joining a startup trying to find its product-market fit. Of course, it already had tailwinds.

Alex Heath: When you joined, the Claude 3 series had already been released, right? Opus, Sonnet, and Haiku.

Mike Krieger: Yes. That was Anthropic's first time delivering a model series that was at least near the frontier. There was still so much to do on the product side: what exactly should this product become?

Although my background is more consumer-oriented, I was excited because during the period between Instagram and Anthropic, I did a lot of investing with my Instagram co-founder Kevin. We had a whole investment thesis, one part of which was "the future of work" – how work would get done in the future.

And Anthropic seemed very likely to unlock that thesis: what happens when you have a very smart assistant helping you work? I didn't even foresee how disruptive this would become.

Alex Heath: Back then you might have thought: this is an interesting little AI company, maybe it can help me understand some investment themes.

Mike Krieger: Yes, maybe it would help us understand some themes we were thinking about. But actually, it changed far more than I imagined.

That was phase one: a very small team, a handful of projects. Then fast forward to the end of last year, the product team had grown to several hundred people. We had a whole portfolio of projects. A lot of the work became deployment, understanding customer needs, customer-facing, management layers, and all the things that inevitably happen as a company grows.

I gradually realized that some people really like this kind of work and are very good at it. I respect them immensely. But for me, I had a great coach who described this state as the "competence zone" – things you're good at, you do well, you can handle, but it's not what truly ignites and drives you.

This is actually a dangerous position. Because you can stay there for a long, long time and perform well, but it's not where your flame and motivation are strongest.

So, in Q4 of last year, I started discussing this with Daniela. I said the company has grown. We've essentially compressed the typical five-year growth journey into a much shorter time. Even though it's only been about two years.

Alex Heath: Yes, I think you guys have grown quite well.

Mike Krieger: Yes, growth is okay. The team size and product portfolio expanded rapidly. So I said, I think I want to start a new company.

Daniela asked me: Is this because you want to leave Anthropic, or because you want to change what you do within the company? I said, I love this company. The people are great, I love the technology and mission.

Around that time, we were also restarting Labs. Because Labs v1 was so successful that all projects had graduated, leaving no one behind. So Labs was effectively put aside.

So we decided to reboot Labs, and I returned to the builder role. Everyone who saw me inside and outside of work said: "Mike, you look so happy."

Alex Heath: Some of your colleagues told me the same thing earlier today. They said Mike is in such a great place now, really enjoying it.

Mike Krieger: Yes. Of course, I'm still my own harshest critic. So every day, I think: how can I do better? What can we build? What are we validating?

So it's not an easy job. But it does align much better with what truly drives me.

Alex Heath: We don't need to dwell on this too long, but I'm fascinated by these "moonshot" or "zero-to-one" labs inside tech companies. Alphabet is probably the most classic example, but there are many such