Anthropic Head of Design on Claude Code's Evolution from an Internal Feature into the Fastest-Growing Revenue Product in History | Meaghan Choi | E298

Anthropic just closed a $65 billion Series H round at a valuation approaching one trillion dollars — and has crossed $30 billion in annualized revenue, driven largely by enterprise demand. Claude Code alone became generally available in May 2025 and reached $2.5 billion in annualized revenue in February 2026, with that figure more than doubling since the beginning of 2026. 

Meaghan Choi, Head of Design for Claude Code and Cowork at Anthropic, was in that room. This conversation goes inside the operating model behind that growth.

What you'll learn:

• Claude Code's evolution from an internal feature into one of the fastest-growing revenue products in history
• Anthropic's secret sauce to shipping products at an incredibly high cadence while ensuring quality
• How product teams get structured into small pods of 5 AI Builders and a fleet of agents, where non-engineers ship code into production
• Driving enterprise adoption through PLG from technical teams
• How organizations can measure AI ROI beyond AI adoption and token usage
• Designing user interfaces for agentic capabilities, including CLI

Key takeaways:

• Titles and role boundaries matter less than contribution. At Anthropic, designers ship code and engineers design, and the pod owns the output collectively.
• Quality gates have moved downstream. The richest product learnings come from working software, not from reviewing mocks or PRDs.
• Managing a team now means managing both people and a fleet of AI agents. The skills are more similar than they appear.

Credits:
Host: Carlos Gonzalez de Villaumbrosia
Guest: Meaghan Choi

Social Links:

• Find out more about Product School here
• Follow our Podcast on TikTok here
• Follow Product School on LinkedIn here

Transcript

Introduction

Meaghan Choi | Anthropic 00:00:00 In its first year, Claude Code made $2.5 billion from nothing, which was unprecedented for us as a team. And I believe at this point we're about 51% of the coding market. We used to gate product quality decisions at a discussion level or at a mock. You'd look at a mock or a Figma, and that's when you'd make a decision on whether this was the right quality. We've now pushed that decision-making into live working code. Titles don't really matter at all anymore. I am pushing code to production. My engineers are designing. I think we're very fluid, and it depends mostly on what we're building. I'm a very big believer that anyone should be able to ship to production, and anyone should be able to build. That's the whole point of Claude Code.

Carlos González de Villaumbrosia | Product School 00:01:00 Hey, this is Carlos, CEO at Product School and your host on The Product Podcast. Today's guest is Meaghan Choi, Head of Design for Claude Code and CoWork at Anthropic. This episode is special. It was recorded live on stage at ProductCon New York, the AI conference that we host three times a year in San Francisco, New York, and London.

Anthropic just closed a $65 billion Series H round at a valuation approaching one trillion dollars, and has crossed $30 billion in annualized revenue, driven largely by enterprise demand. Claude Code alone became generally available in May 2025 and reached $2.5 billion in annualized revenue in February 2026, with that figure more than doubling since the beginning of 2026.

In our conversation, we cover Claude Code's evolution from an internal feature into one of the fastest-growing revenue products in history, Anthropic's secret sauce to shipping products at an incredibly high cadence while ensuring quality, how product teams get structured into small pods of five AI builders and a fleet of agents where non-engineers ship code into production, driving enterprise adoption through PLG from technical teams, how organizations can measure AI ROI beyond AI adoption and token usage, and designing user interfaces for agentic capabilities, including CLI.

Let's get into it.

From Living Room to Live Stage: Welcome to ProductCon New York

Carlos González de Villaumbrosia | Product School 00:02:00 I usually do these interviews at home. Pretty cool. What a turnout. Welcome to The Product Podcast at ProductCon New York, Meaghan.

Meaghan Choi | Anthropic 00:02:08 It feels amazing. It's incredible to be here right now. I'm so excited.

Carlos González de Villaumbrosia | Product School 00:02:13 So, Head of Design at Claude Code and Claude CoWork.

Meaghan Choi | Anthropic 00:02:16 I lead design for Claude Code and CoWork, which is crazy to say and crazy to be here even now. I'm always shocked from where we started about a year and a half ago, twelve people in a room wondering if this product would even be a thing, wondering if people would use CLI, and now it's one of the most popular dev tools in the world. It's insane.

Carlos González de Villaumbrosia | Product School 00:02:38 I have a couple of fun facts for you. So you have the most female-led product org in tech right now. Your Chief Product Officer, your Head of Product, your Head of Engineering, your Head of Platform Product, your Head of Platform Engineering, and your President.

Meaghan Choi | Anthropic 00:03:00 It's amazing. Honestly, a lot of us didn't even think too much about it until we were watching some of the Code with Claude talks and realizing the industry really recognized that. It happened so naturally, just based on the culture and the values that we have at Anthropic. We just have an incredible leadership team, and they all happen to be women. It's pretty nice to be a part of, I have to say.

How Claude Code Went from Side Project to Product Category

Carlos González de Villaumbrosia | Product School 00:03:22 I want to go back to the early days when Claude Code wasn't even considered a product. How did that happen, and what had to be true for Claude Code to actually graduate into a product category?

Meaghan Choi | Anthropic 00:03:38 Claude Code started as a side project with a few engineers on the team who were exploring, back in 2024, what it would look like to have Claude write code for you. At the time, we were still doing a lot of copying and pasting into a chat and then copying and pasting back into your editor, or maybe tab-to-autocomplete if you were using it. But Claude actually couldn't act on your behalf.

Then there was a breakthrough moment where someone on the team piloted an experience of doing this via CLI, which has access to everything on your computer. They posted a video of it on our internal Slack, and a few people looked at it, myself included. I was like, "Holy crap, this is it. This is incredible." But it took about an hour to set up. It was extremely difficult to use, didn't really work exactly the way you'd want it to, and was probably six months early for the models we wanted it to be powered by. But there were signs of life.

The team spent a lot of effort making it more usable and tuning the UX. The next three months were really about releasing it internally and trying to get as much internal usage as possible. We really believed that if we could build a tool that everyone at Anthropic would want to use, we'd build a tool that could be really useful for people in the industry as well. It was a tireless iteration of following people around, watching them use it, fixing bugs. My first interaction with the product was actually helping do some design iterations on it, but at the time CLI design was so new that it was all in Google Docs, because that was the best representation you could have of what a CLI would look like.

Building AI-Native: The Builder Mindset and Pod Model

Carlos González de Villaumbrosia | Product School 00:05:20 During my talk earlier, I was sharing some insights on top companies that had to transform into AI-native. Your company was born AI-native. I'm curious to pick your brain on the concept of "builder." How are you thinking about it?

Meaghan Choi | Anthropic 00:05:38 Transparently, we're all learning what this looks like together. We're growing as an industry as we grow this technology. But a few big shifts have happened within the team.

The first is that our titles don't really matter at all anymore, regardless of what role you come in as. I am pushing code to production. My engineers are designing. I'm not involved in every feature. I think we're very fluid, and it depends mostly on what we're building.

My ideal team size and shape is probably three to five people in a pod. That can be five engineers, four engineers and a designer, two designers and three engineers, two PMs and three engineers, or a PM and four engineers. Everyone is sprinting forward together to get something actually working in code, contributing in all the different ways and skillsets they bring. The goal from there is to use it yourselves and get other people to use it, and that's how you build conviction. I think of titles as a specialty you might bring to the team, but it doesn't delineate what you can contribute to the project.

Carlos González de Villaumbrosia | Product School 00:06:55 So you have a small pod of people. The general term could be "builder." Their background can be different: design, engineering, product. You can feed them with less than two pizzas. In terms of shipping cadence, you're mentioning that a non-engineer can ship. Tell me more about that.

Meaghan Choi | Anthropic 00:07:10 I'm a very big believer that anyone should be able to ship to production, and anyone should be able to build. That's the whole point of Claude Code. If you're nervous about that or if it's stressful, what you can actually invest in is good code review, good CI, good testing, to make sure code that goes into your production app is verified when reviewed.

The biggest thing I had to become more flexible with was the idea of product quality and polish. As a designer, it's sometimes really hard to let features go out that I haven't touched, or that don't have the most polished version of what I'd want. But once you start operating this way, you really get a sense that with the power of AI, everyone can do everything.

In the same way that engineers are inviting us into code right now, as a designer I invite other people to help design. Product managers should invite everyone to make PM decisions and product decisions. It's all a shared responsibility. When it comes to the decision of when something is ready to ship, the pod decides first: is this good enough that we're using it, that we want to release it to the rest of the team internally? Once we have enough people using it internally and we have feedback, we look for a specific number of internal daily active users who are truly adopting these products into their workflow. Once we see that growth, we release it externally.

Carlos González de Villaumbrosia | Product School 00:08:45 So you are customer zero. You release to your company first, and if there's enough traction, you open up the gates.

Meaghan Choi | Anthropic 00:08:51 Yeah, absolutely.

Shipping at Speed: Where Is the Bottleneck Now?

Carlos González de Villaumbrosia | Product School 00:08:54 If shipping is not the bottleneck anymore, given that other people can now contribute code, where is the bottleneck in terms of ensuring quality? There was an image of Claude going viral saying "this is all the stuff Anthropic released in the last few weeks," and it was insane.

Meaghan Choi | Anthropic 00:09:10 There are two things I think about there. The first is that working at an AI lab in particular, and if you've had any experience in emerging tech — AR, VR, spatial computing — the iteration of shipping very quickly and learning is by nature part of new tech development, and it's important to keep up with that speed. It should feel overwhelming, and we're not going to get everything right. But we like to build and test in the open. Our users often help us find what works fastest, and you don't necessarily need to adopt everything.

When it comes to maintaining product quality, this is the part that's a little uncomfortable right now. We used to gate product quality decisions at a discussion level or at a mock. You'd look at a PRD and make a decision. You'd look at a Figma and make a decision on whether this was the right quality. We've now pushed that decision-making into live working code. You need to be using it yourself, experiencing the product in its actual workflow. That's when you make a decision on quality, because it's so much more reflective of what the actual experience is going to be. The learnings are that much richer. But it's uncomfortable to push people to build first and then iterate on quality, rather than doing all the iterations beforehand. It's a complete flip on how we've worked.

Enterprise Adoption Through Developer PLG

Carlos González de Villaumbrosia | Product School 00:10:45 I'm sure there are a lot of people in the audience freaking out right now — CPOs, VPs of Product at large enterprises in financial services, healthcare, retail, saying "we're heavily regulated." Yet most of your customers actually come from large enterprises. How are you creating that level of adoption?

Meaghan Choi | Anthropic 00:11:05 Claude Code is primarily community-driven, which is pretty special for a developer product. The way we've been able to expand into enterprises is by having individual developers who use the product in their personal projects and side projects, and then they become advocates internally for expanding Claude Code usage within their organizations.

The really interesting thing about Claude Code, and about a lot of AI products if you build them as platforms, is that the more people who use them, the better the experience gets for everyone. As a team of engineers starts to spin up on using Claude Code or CoWork or any of these agentic tools, they'll start to build infrastructure around it. Their dev platform team will be highly motivated to support custom tooling, connectors to all of your databases, all of your internal infrastructure. Getting access to that makes using Claude Code even more effective. Then it's a cycle of everyone building things for themselves that help build things for each other.

As you expand into other functions, at Anthropic all the designers use Claude Code, all of product, even our finance team uses Claude Code. Once everyone is using it, you start to extend it into other skillsets that then multiply your ability to learn those skills as well.

Carlos González de Villaumbrosia | Product School 00:12:25 What you're describing sounds like the PLG motion that a lot of companies started, but as they went more enterprise they also had to develop a top-down approach. You mentioned developers tend to be the early adopters, then you expand from within. It must get to a point where someone raises their hand and says, "But we're using another tool and this is too risky for us." How do you provide peace of mind to people who are more risk-averse?

Meaghan Choi | Anthropic 00:12:50 There are a few things. The first is that by the design of the Claude Code package, it's actually extremely hackable and extremely safe. The bundle itself as a tool meets all of those requirements already, and that was an easy way to get a lot of enterprises already adopting Claude Code without needing to go through really intense review.

As we're growing into larger and larger enterprises, we are learning and adding more controls and more visibility into our platform. We're a young company, so we're learning by doing. We scaled faster than any of us imagined, but it hasn't really been a hurdle. It's more the opposite right now: everyone wants to use Claude Code, and we're trying as hard as possible to make it as easy as possible to access.

Designing for Agentic AI: UI, CLI, and the Minimalist Approach

Carlos González de Villaumbrosia | Product School 00:13:45 One of the things I notice is that you're trying to integrate with a lot of other tools and systems of record. It seems like a central spot for companies to come and collaborate. A lot of SaaS companies have had to give up on the UI as a moat to focus more on data. For you as a designer, how are you thinking about user interfaces?

Meaghan Choi | Anthropic 00:14:05 I generally feel that the user interface is just a medium to access the technology. By nature I'm a minimalist. What you actually want is the purest form of technology. That's why I loved the CLI, because I think it's truly the thinnest wrapper you could have around our models, and that's what people want access to when they're building.

I think we need to be more open-minded as an industry and as a function about what it really means to build a UI, and whether it's even valuable. At the end of the day, what you're really trying to get to is a work product or an output. That is a lot more important than the intermediary tooling. All the tools we build right now are really just ways for you to communicate to your devices or services what you want that output to be. If we could just focus on the final output and work based off of that, it's a much faster way to develop, and that's what a lot of AI tooling is focused on.

Carlos González de Villaumbrosia | Product School 00:15:10 I come from a technical background. I thought I would never use CLI again in my life, yet here I am. Is this a temporary interface that will evolve into something more elaborate, or is the CLI cool again and here to stay?

Meaghan Choi | Anthropic 00:15:28 The CLI is still an extremely important interface because it connects the most with all the operations and all the information on your computer already, and it's really integrated into the dev stack today. In that sense, it just works out of the box.

As we're expanding into more non-engineering audiences, I consider myself the semi-technical audience, where a lot of the work products and tooling aren't necessarily only on your computer. We're starting to introduce new products like CoWork, which are a little easier to use. If any of you have used the CLI, you know it's not easy. It's a whole new learning curve. I think it's a great place for engineering and coding work. I'm a CLI diehard user. But the desktop app for Claude Code and CoWork are really making it more accessible for more people.

People Management in the Age of AI Agents

Carlos González de Villaumbrosia | Product School 00:16:30 A lot of companies are not only removing management layers but also expecting that people managers act as player-coaches. To what degree do you expect your people managers to also be building?

Meaghan Choi | Anthropic 00:16:48 All of the people managers at Anthropic build. I build, I design. I personally have a principle that it's always helpful, if you're joining or at a company, to know what the shipping motion looks like. Now more than ever with AI, because the process looks so different, going through that experience yourself is really important so you can build the muscle and understand what tooling and training to invest in for your team.

I also see all of my team and a lot of individual contributors at the company as mini managers right now, because everyone is managing a fleet of Claudes that are also working. In the same way that you as a people manager might manage people, you're also managing Claude. It's a little bit of moving up the chain, I guess, is the analogy I would give.

Carlos González de Villaumbrosia | Product School 00:17:32 You're still managing agents anyway.

Measuring AI ROI: Beyond Adoption and Token Usage

Carlos González de Villaumbrosia | Product School 00:17:38 I want to talk about money. We're in New York. Adoption doesn't seem to be a strong enough indicator of value for the business. How are you thinking about ways to actually measure the value of AI?

Meaghan Choi | Anthropic 00:17:52 Transparently, we're all pretty early in this journey. The analogy I like to give is when engineering companies used to evaluate IC engineers by lines of code or PRs committed, and then over time we learned that's not necessarily the best measure of someone's contributions. We're in a similar era right now where we can see this technology is providing a lot of value, and we want to see more of our teams adopting it because we know it accelerates their work. But by how much should they be using it? That's hard.

The two tactical things I say are: token usage is primarily a good indicator of someone not using AI. If it's zero, that should be a concern. But no one should be competing over who's using the most tokens, because it's very easy to use an infinite number of tokens and not do any work at all. There are automations you can set up that do use a lot of tokens and do give a lot of value, so I don't know if tokens is the right measure of general ROI, but it's a good leading indicator right now.

I still go back to the original impact metrics: what is the adoption, what is the retention, what is the revenue? Those all still hold true no matter if you're using AI or not, and those are the things we should really be looking at.

$2.5B in Year One and What Comes Next

Carlos González de Villaumbrosia | Product School 00:19:30 Just to wrap it up — Claude Code is one of the fastest-growing products in terms of revenue and adoption, and you're at a scale where it's really hard to maintain that kind of percentage growth. How do you keep that momentum?

Meaghan Choi | Anthropic 00:19:52 To add some numbers that I always find a little bit fun: in its first year, Claude Code made $2.5 billion from nothing, which was unprecedented for us as a team. And I believe at this point we're about 51% of the coding market.

I think we're just so early in this journey. We're at 1%, and there's so much more to go. For us on the team, it doesn't feel like we've succeeded or that we're at the end. It feels like we've only just put our foot in the door of what's possible. Every day we're prototyping new things, trying new ways of working with AI that flip everything we know on its head and change all the assumptions we have about work.

The real skill right now is to be flexible and not hold anything too dear. Live curiously and build curiously, because it is going to keep changing. It's already changed so much over the past few months, and all you can do is be excited. Building right now is so fun, and finding ways to make building even more fun is what brings us joy on the team.

Carlos González de Villaumbrosia | Product School 00:20:48 Meaghan, it's been awesome to have you here. You call yourself a builder as well, but you're literally the only designer on the stage with us today. Let's give it up for Meaghan.

Meaghan Choi | Anthropic 00:21:00 Thank you so much.