Building OpenCode with Dax Raad: A Conversation with Its Creator (Category AI4SDLC)
I watched the new episode of The Pragmatic Engineer, where Gergely Orosz talks with OpenCode creator Dax Raad. It is interesting beyond the product story: the creator of a rapidly growing AI coding tool consistently refuses to sell magic and speaks with noticeable scepticism.
OpenCode is an open-source coding agent that began in the terminal and later gained a GUI. Dax says it grew from roughly 650 thousand to nearly 8 million monthly active users in a few months, with around a million daily users. Impressive figures, but the story behind them is more interesting.
1️⃣ Anthropic’s block is the most revealing episode. When it stopped OpenCode working through Claude subscriptions, the move looked like an existential threat. Dax describes the opposite: the team turned to OpenAI and other model providers, and the episode accelerated growth. His lesson is almost a product manifesto: positioning matters more than execution speed. OpenCode did not win because its harness was best; he admits it was merely good enough in the early months. It won by claiming the open-source coding agent category first. “Get positioning right and the world just keeps handing you wins,” he says. Position and market share first, quality improvements afterwards.
2️⃣ He is equally direct about the business model. The company earns money through OpenCode Zen. GPUs are depreciating capital assets, and he estimates token margins of up to 90%, depending on the model. That is a useful perspective amid claims that AI loses money for everyone, although it comes from someone who makes money on inference.
3️⃣ The liveliest section concerns productivity. Dax says that before AI, 95% of his energy went into deciding what to do and 5% into execution. Now it is 96% thinking and 4% doing. Technically an improvement, but everyday work feels just as hard. AI removes doing, not thinking, and thinking was always the bottleneck. He calls confident predictions about AI’s future a form of self-reassurance: people tend to predict advantages for their own group.
4️⃣ I particularly liked the engineering culture discussion. Quality-conscious engineers drown in low-quality PRs from colleagues who care less and burn out cleaning them up. A subtler point: AI removes psychological friction — the guilt of cutting a corner — allowing technical debt to accumulate unnoticed. There is an upside too: agent-assisted refactoring is cheap, and we should use it more aggressively. Interestingly, “enterprise” patterns such as domain-driven design become useful again as safeguards for junior engineers and agents alike.
5️⃣ Dax’s career advice is simple: strong software engineering knowledge plus deep domain expertise is a combination likely to endure. Engineers systematically underestimate the second part.
When the creator of a successful AI coding tool says AI changes less than it seems, I pay closer attention than I do to vendor presentations. The constraint is thinking, quality and domain knowledge rather than code generation. That is where we should invest.
#AI #AI4SDLC #Engineering #Management #Product #Software #Architecture