With all the AI hype sweeping through tech, GitHub CEO Thomas Dohmke is stepping in with a reality check: artificial intelligence isn’t killing programming—it’s transforming it.
Speaking on “The MAD Podcast with Matt Turck,” Dohmke tackled one of the biggest misconceptions in today’s developer world—that AI will make traditional coding obsolete. His answer? Not even close.
“The worst alternative is trying to figure out how to explain in natural language what I already know how to do in code,” Dohmke said. His point? Developers still need to know how to code. Period.
AI Can’t Replace Developer Instinct
Dohmke described an ideal workflow: AI generates pull requests, and human developers fine-tune the results. This hybrid approach, he argued, strikes the right balance between speed and accuracy. But blindly trusting AI to write flawless code? That’s a shortcut to chaos.
According to Deloitte research, over 50% of AI-generated code requires correction. Even Google, which uses AI for more than a quarter of its codebase, still runs every line through human review. The message is clear: AI can help, but it can’t think like a skilled engineer.
AI Is a Co-Pilot, Not the Driver
Today’s smartest dev teams are using AI to handle the boring stuff—like boilerplate code—so they can focus on high-impact decisions. It’s about working smarter, not replacing talent. Developers become more like AI orchestrators: directing tools, refining outcomes, and safeguarding software quality.
For junior devs, AI can be a launchpad. For seasoned pros, it unlocks time for architectural thinking and deeper innovation. And with talent shortages still hitting tech hard, AI isn’t replacing coders—it’s helping them scale their impact.
Watch Out for “Vibe Coding”
Dohmke also warned about the rise of what OpenAI’s Andrej Karpathy calls “vibe coding”—a trend where people lean too hard on AI-generated code without really understanding what it’s doing. It’s fast, sure, but risky. Especially for startups led by non-technical founders, vibe coding can lead to shaky foundations, mounting technical debt, and long-term headaches.
Big companies already know better: they use AI for speed but back it up with strong QA processes and engineering standards. That’s the model Dohmke recommends for everyone else.
About the Author
