GitHub CEO Thomas Dohmke says the "key for winning" with AI coding tools is maintaining the ability to quickly modify AI-generated code manually, rather than relying entirely on automated agents. Speaking on "The MAD Podcast with Matt Turck," Dohmke emphasized that developers need flexibility to seamlessly transition between AI assistance and hands-on coding.
The CEO outlined an ideal workflow where AI agents can write code and submit pull requests, but developers retain the ability to make rapid changes using their existing programming skills . This approach prevents the productivity trap of spending minutes describing simple changes in natural language when the same task could be completed in seconds through direct coding.
"The worst alternative is trying to figure out how to provide feedback or prompt to describe in natural language what I already know how to do in programming language," Dohmke explained. Such scenarios could transform three-second tasks into three-minute ordeals, ultimately reducing rather than enhancing productivity.
Startups can't rely on " vibe coding " alone
Dohmke also addressed the trending "vibe coding" phenomenon, a term coined by OpenAI cofounder Andrej Karpathy to describe heavy reliance on AI for code generation. During a recent Q&A at Station F in Paris, the GitHub CEO cautioned that startups cannot scale on vibe coding alone.
Non-technical founders will struggle to build sustainable companies without skilled developers, he argued, noting they "can't build a complex system to justify the next round." The key is enabling developers to choose the best approach for each situation—whether that's leveraging AI agents for maximum ROI or handling tasks directly themselves.
The CEO outlined an ideal workflow where AI agents can write code and submit pull requests, but developers retain the ability to make rapid changes using their existing programming skills . This approach prevents the productivity trap of spending minutes describing simple changes in natural language when the same task could be completed in seconds through direct coding.
"The worst alternative is trying to figure out how to provide feedback or prompt to describe in natural language what I already know how to do in programming language," Dohmke explained. Such scenarios could transform three-second tasks into three-minute ordeals, ultimately reducing rather than enhancing productivity.
Startups can't rely on " vibe coding " alone
Dohmke also addressed the trending "vibe coding" phenomenon, a term coined by OpenAI cofounder Andrej Karpathy to describe heavy reliance on AI for code generation. During a recent Q&A at Station F in Paris, the GitHub CEO cautioned that startups cannot scale on vibe coding alone.
Non-technical founders will struggle to build sustainable companies without skilled developers, he argued, noting they "can't build a complex system to justify the next round." The key is enabling developers to choose the best approach for each situation—whether that's leveraging AI agents for maximum ROI or handling tasks directly themselves.
You may also like
1st Test: Duckett-Pope century stand takes England to 107/1, trail India by 364 runs
Desperate passengers trapped on Thameslink trains amid 34C highs evacuated on tracks
Marc Guehi 'decides between Arsenal and Liverpool transfer' with bid imminent
What's Behind DGCA's Demand for Air India Inspection Reports?
Five Premier League stars who went bankrupt including Man Utd icon and England regular
