The popular AI coding tool: The co-founder says engineers still have hope.

In the past year, AI coding tools have rapidly evolved from being "co - pilots" to "agents" capable of independently writing code and managing code repositories. OpenCode is one of the fastest - growing products in this wave. Its monthly active users soared from 650,000 to 6.5 million within a few months, and it is expected to exceed 8 million this month. However, its co - founder, Dax Raad, might be the least eager among all AI tool founders to hype up AI.

In this episode of "The Pragmatic Engineer Podcast", Dax, with a rare honesty, dissected three core illusions created by AI coding tools in real engineering teams:

First, the code output has skyrocketed, but "doing things faster" doesn't mean "doing things better". The more features are piled on, the more the product resembles a Frankenstein - like monster. Every piece is put together, but the whole has gotten out of control.

Second, AI is quietly eliminating engineers' "guilt". In the past, writing hacky code made you uncomfortable. The second time it was even more uncomfortable, and the third time you'd stop and refactor. That stinging feeling helped you correct your judgment. Now, the agent does the dirty work for you. The landmines are still there, you just don't feel them, and your judgment is being stolen.

Third, most engineers aren't using AI to increase output but to save time. They finish the same work and go home earlier. This is completely reasonable for individuals, but it has nothing to do with the "productivity doubling" narrative that AI vendors sell to CFOs.

The following is a compilation.

1

Dax's Growth Path: From Minecraft, Failed Entrepreneurship to the Open - source World

Host: Before talking about OpenCode, I'd like to start with you. You're now working on one of the hottest AI development tools globally, but you have a rather rare quality: you're not superstitious about AI. So let's go back a bit. How did you first enter the technology industry?

Dax: It's quite typical. I started programming when I was a kid. My dad is also a software engineer, so it was easier for me to get started than for many others. Later, I started working right after high school and also founded some companies. At that time, I thought I knew a lot, but looking back, I really knew nothing. Later, my company had a small - scale acqui - hire, and I officially entered the tech industry. After that, I did consulting and founded several more companies. Then, for about six years, I worked full - time on open - source projects.

Host: I also noticed that you were tinkering with Minecraft servers and mods very early on.

Dax: Yes. Almost everyone knows Minecraft. At that time, a mod framework came out, and I got directly involved in it and made many mods. What I'm really interested in is not "playing the game" but "creating a sandbox". I'd set up a server for about a hundred people and use these mods to create all sorts of strange scenarios and observe how people would behave in these environments. I find that extremely fascinating. Of course, it requires writing a lot of Java behind the scenes.

More importantly, there was that community. At that time, I was still a programming novice, and people mainly communicated on IRC. There were some very talented and experienced programmers in it, but they weren't overly career - driven. They seemed to be in a comfortable state in life, maybe working only two hours a day. They were very good at technology but not interested in the corporate rat - race, so they put their energy into Minecraft. This allowed me to learn a great deal in a very short time.

Host: Later, you went to a startup, founded your own company, and were an early - stage entrepreneur. Did the experience at Ride Health have a big impact on you?

Dax: It had a huge impact. It was a company at the intersection of transportation and healthcare. At first, it was just me and another co - founder, and later it grew to about twenty people. That was my second real attempt at entrepreneurship, and it went further than before, but it basically ended in a disaster. However, I met my wife there. She was the product lead, and I was the engineering lead. We got together after working together for a year. So if you ask me about the outcome, I'd say: although the company wasn't successful, this was more valuable than an entrepreneurial exit.

However, that experience taught me a profound lesson. We were all very young, in our early twenties, and the whole team was like a typical "young startup team". If I were to invest in a company in the future, I'd probably not invest in a team composed of very young people. Back then, our minds weren't fully developed. Our sense of security, desire for self - proof, way of handling relationships, and reaction to conflicts were all magnified infinitely in the high - pressure and intimate environment of entrepreneurship. Looking back now, I didn't really feel "stable" until I was about 26. Before that, I probably shouldn't have been in charge of running a company.

Host: So do you think that young people's entrepreneurial success is actually an exception rather than the norm?

Dax: I think so. Of course, people always cite those famous cases, but on average, I don't think it's the norm. Entrepreneurship is very intimate and high - pressure. If a person isn't well - grounded as an individual, many problems will surface at work, especially in such a small and close - knit team. When you look back later, you'll realize that many of the conflicts, emotions, and judgments back then were magnified and completely unnecessary.

Host: In those early years, the tech circle really admired big companies and star unicorns. Did you ever want to go there?

Dax: I did. It was from the early to mid - 2010s. Big companies and popular unicorns were really attractive, and it seemed like everyone was desperately trying to get in. At that time, if you didn't go in that direction, there was an atmosphere that "you were going to fail in the tech industry". But to be honest, it wasn't that I "actively didn't go"; it's just that I never trained myself to fit into that system. Those companies have a very structured interview process that requires practice, preparation, and targeted training. I think I'm a decent programmer and have had several similar interviews, and I even did well in some of them. But that highly competitive, long - term preparation - style path isn't something I'd naturally invest my energy in. I prefer to directly create things, am more practical, and find it difficult to shift my attention elsewhere.

Host: So how did you actually transition into the open - source world?

Dax: After Ride Health ran into problems, I went to a startup that was in around the Series B stage. For me at that time, it was the biggest company I'd ever been in. Later, I was put in a director position and became a pure manager for the first time, with three or four hours of meetings every day. But I still really wanted to write code. So outside of meetings, I started working on open - source projects. I first wrote some really bad small things on my own, and then I came across SST, which had just been launched. It was a project by Frank and Jay. I started contributing code to it, and they were raising funds after YC. I even invested some money. A month later, I directly joined the team, and they gave me the money back as salary, but I still had to pay the tax myself. So I learned a lesson: if you might go to work for a company, it's better not to invest in it as an investor first, otherwise, you'll end up paying a tax for nothing, which is very unprofitable.

Host: Later, besides SST, you also worked on OpenNext.

Dax: OpenNext was actually one of the projects that first made us popular, and it wasn't something we originally wanted to do. When serving AWS developers, almost every day someone would ask: "You're doing a great job, but what we really need is: how to deploy Next.js on AWS?" This demand bombarded us for a whole year. We actually didn't want to do it because we weren't heavy users of Next.js ourselves, and the work was trivial and boring. Next.js is very complex. To recreate the correct infrastructure on AWS, we had to get into the annoying details of packaging and the internal mechanisms of the framework. In the end, we pushed this task to Frank.

Our goal was clear from the start: ideally, this project should "not need to exist" in the end. It was only because the Next.js team naturally leaned more towards Vercel, so there was a gap in the ecosystem. We wanted to fill this gap. Ideally, the official team would handle this in the future, and then OpenNext would no longer be needed. Later, it did gradually move in this direction: Cloudflare, Netlify, Microsoft, Google, etc. also joined in and created adapters for OpenNext. The Next.js team also had time to launch an official adapters API. In the past year or two, this has gradually changed from "confrontation" to "collaboration", and the necessity of OpenNext has been slowly decreasing.

2

How was OpenCode Born?

Host: So how did OpenCode come about? This is actually a very new story, starting around the summer of 2025.

Dax: A bit earlier. It was around February 2025. At that time, our company was fully focused on achieving profitability. SST already had a commercialization path, and for about three or four months, we were in a state where "the money was almost running out, but the revenue was increasing". By February, we only had one - month's worth of cash flow left, but at the same time, we also just achieved break - even. Strangely enough, our whole team was very calm at that time, as if we defaulted that "things would work out", and they really did.

That gave us some space to think: since theoretically, we could do anything next, what exactly did we want to do? We've long thought that in this decade, if you're still working on developer tools, you can't avoid AI. Every wave in history has been like this: when something new emerges, it may seem extremely absurd on the surface. Because as long as something has enough potential, it will attract a large amount of investment, and most of the investment is ultimately unreasonable. So the easiest mistake to make is to see a bunch of unreliable things and conclude that "the whole thing isn't worth getting involved in". But usually, the real situation is that some things are really stupid, but there are also a few things that are very real. If you completely stay out, you'll miss the truly valuable part. So we started trying out some AI - related ideas, many of which weren't even officially launched because we found they weren't feasible halfway through. Later, the whole team started using Claude Code. This was the first AI programming tool that we really used continuously, and it did solve some annoying problems in our workflow. While using it, there was a moment when I suddenly asked myself: why didn't we create this? We should have been the ones to do this kind of thing.

Then we started looking at this from the perspective of "market positioning" rather than "functional details". There were already a number of coding agents in the market, but for some reason, no one had really occupied the "open - source" position. In the world of developer tools, this is an extremely valuable position. Looking at databases, compilers, infrastructure, and various development tools in the long run, the open - source option often becomes the default option. At the same time, the competition among model providers is extremely fierce. Claude was very strong at that time, but you can't really believe that other players who have invested tens of billions of dollars will just sit back and let Anthropic win. OpenAI will push forward, open - source models will push forward, and other companies will too. So in a stage where model competition is extremely intense and the ecosystem is constantly changing, creating a neutral, open - source product that can connect to all models has great strategic value.

Host: So you didn't start from the idea of "creating the smartest agent" but from "occupying the most important position" first.

Dax: That's more or less what it means. When we started, there were only three of us, all co - founders. Later, we asked a friend to help us with the early version, making it four people. After that, we recruited an excellent designer who we'd always wanted to work with. That was after the launch, around autumn. The team was actually very small, but once we clarified the direction, many subsequent actions went smoothly.

3

From 650,000 to Nearly 8 Million Monthly Active Users

Host: How fast did you grow after the launch?

Dax: Extremely fast. Faster than anything we'd ever done before. By December 2025, our monthly active users had reached 650,000. At that time, we thought this was already amazing. In autumn, we were still telling people that our goal was to reach 1 million monthly active users by the beginning of next year, and everyone thought we were crazy. Then in January 2026, we directly reached 2.5 million monthly active users. The next month, it was 6.5 million. I estimated that we'd soon reach nearly 8 million. Our next goal is 10 million.

Host: What happened during that jump from 650,000 to 2.5 million?

Dax: Part of it was seasonal. Having worked on developer tools for a long time, we know that there's always a surge in January every year because people try new things during their December vacations, and the usage spikes in the first week back at work. Normally, other products usually see a decline before the holidays and then a sharp increase in the first week after. But OpenCode is special: it kept growing during the holidays, which has never happened with any of our previous products.

On the other hand, to be honest, Anthropic helped us a great deal. They wanted to ban users from using Anthropic's subscriptions in OpenCode. This decision detonated the situation. We didn't even say much ourselves; it was mainly the user community that exploded. Anthropic inadvertently put us in the same sentence as them. We're actually not worthy of this, as they're a much larger and more successful company. But that week, this suddenly pushed us into the spotlight.

Host: What do you think of the Anthropic incident itself?

Dax: I think it shows that they're not very used to dealing with developers. Of course, you can do what you have to do for the sustainability of your business, and there's nothing wrong with that. But if you suddenly impose a ban silently at 9 p.m. without pre - communication or a phased rollout, you're creating a moment where "everyone hates you". Even if you had said it clearly a month in advance and phased it in, people would still be unhappy, but there wouldn't be such a concentrated outburst of anger.

Host: Were you panicked at that time? Because Claude was still the strongest coding model at that time.

Dax: On the contrary, we weren't panicked. We actually knew this day would come sooner or later. For some reason, when it actually happened, everyone was even a bit excited. Because we're very clear that the public opinion circle we're in can distort reality. Especially on X, it seems like everyone has a $200 Claude Max subscription, but that can't represent the real world. At that time, we already had 650,000 monthly active users. It's impossible that all 650,000 of them are spending $200 a month. For most ordinary people, spending $200 a month on anything isn't a small thing. So we knew that only a part of the users would be affected, not all of them.

More importantly, a few weeks before this incident, we had already been talking to almost all other companies about "officially supporting OpenCode subscription access". Microsoft had already agreed to officially support OpenCode with GitHub Copilot; we were also in the