OpenCode's monthly active users grew from 650,000 to 6.5 million within months, heading toward 8 million—among the fastest growth rates in the AI coding tool wave. Yet in a recent podcast appearance, co-founder Dax Raad, someone directly benefiting from the AI boom, poured cold water on the industry, arguing that AI coding tools create three fatal illusions in real engineering teams.
1. Exploding output does not mean a better product
AI has made writing code 10x faster. That is a fact. But Raad's concern is simpler: engineers' instinct becomes "I used to build one feature; now I can build ten."
Products increasingly resemble Frankenstein's monster—every part installed, but the whole out of control. Every feature carries a long-term maintenance cost, and AI will not pay down your technical debt. As code volume grows 10x, architectural consistency, maintainability, and core user experience can actually deteriorate.
Raad says AI made "doing" cheap, but "figuring out what to do" remains expensive. His own time allocation barely changed: pre-AI, 95% thinking and 5% execution; now, 96% thinking and 4% execution. The denominator shrank; the ratio barely moved.
Speed is a property of tools; direction is a property of people. Conflating the two is a mistake many engineering teams are making.
2. AI is quietly removing your sense of guilt
This is the most insidious one.
In the pre-AI era, writing bad code made you uncomfortable. The second time, more so. The third time, you stopped and refactored. That sting was an engineer's self-correction mechanism—a quality sensor.
Now? Agents handle the dirty work for you. The landmines are still there, but they no longer hurt. Your judgment is being stolen incrementally.
Raad has seen it firsthand in his team: some engineers use AI to blast out output, caring only about finishing tasks; another group of principled engineers drowns in garbage PRs, exhausted from cleaning up technical debt, until burnout. This is not a tool problem—it is an engineering leadership problem. Most companies do not realize their code quality is silently collapsing.
When mistakes no longer hurt, you are not far from the mistakes themselves.
3. Engineers use AI to save time, not to produce more
This one is the harshest.
The story AI vendors sell CFOs: use our tool and team output doubles. What actually happens? Engineers finish the same work in the same hours, then leave early.
Raad says this is entirely rational for individuals—who does not want to go home earlier? But for companies, the ROI of AI tools is severely overstated. With incentives unchanged, most teams' output stays flat; engineers just clock out earlier.
You bought "10x productivity" and got "same output + less overtime." Between the information asymmetry between CFOs and AI vendors, engineers sit in the middle as silent beneficiaries.
Going deeper: why Raad dares to say this
He is himself a beneficiary of the AI boom—OpenCode is growing faster than anyone. But his underlying thesis: positioning beats intelligence.
OpenCode wins not because it is the smartest, but because it occupies the "open source" position. In developer tools, open source eventually becomes the default. Model competition is background noise—Claude is strong, but OpenAI and open-source models will catch up. In a market where model capabilities rapidly commoditize, neutrality and ecosystem niche matter more than model intelligence.
Raad also dislikes benchmarks. Academic benchmarks test "weird programming puzzles in three-file repos," unrelated to real engineering. Optimizing for them makes you mistake them for what matters. OpenCode instead tests agents on real PRs from real open-source projects—give the agent a real task and see whether it produces code that actually gets merged.
Real work, not academic puzzles.
Conclusion
Raad's interview offers a rare perspective: someone enjoying the AI dividend dismantling AI's hype. This is not contrarianism—it is honesty.
For engineers: do not be fooled by output volume or the feeling of speed. The things that truly matter—judgment, direction, quality sensitivity—are things AI cannot help with and may quietly steal.
For teams: once AI tools are adopted, incentives and management must change accordingly. Otherwise you get not a "productivity revolution," but "engineers who go home earlier."
OpenCode sprinted from 650,000 to 8 million monthly active users, yet Raad's message is: growth is fast, but do not trust the meaning of growth too quickly.
Sources:
- The Pragmatic Engineer Podcast: Building OpenCode with Dax Raad
- 36Kr: Popular AI Coding Tool: Co-founder Says Engineers Still Have Hope
- Codacy AI Giants Podcast: The Creator of OpenCode Thinks You're Fooling Yourself About AI Productivity