GitHub CEO Thomas Dohmke emphasized the need for developers to practice hands-on coding skills in addition to using AI tools. Engineers should be careful of relying too heavily on AI-generated code. Instead, they should promote a flexible approach that enables them to switch between using AI assistance and writing directly.
According to GitHub CEO Thomas Dohmke, the “key to winning” with AI coding tools is the ability to edit AI-generated code manually rather than depending solely on automated agents. Thomas Dohmke, speaking on “The MAD Podcast with Matt Turck,” underlined the importance of developer flexibility to switch between AI assistance and hands-on development.
Dohmke delineated that the ideal workflow should have AI tools that write code and submit merge requests. Developers should keep their ability to make faster changes in the code using their existing coding skills. This perspective is anticipated to prevent the productivity trap of investing minutes in describing simplified alterations in the natural language when the same activity can be completed in seconds via direct coding. He explained:
“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.”
These scenarios would transform a 5-second task into a 5-minute difficulty, simultaneously minimizing rather than improving productivity.
Why ‘Vibe Coding’ Won’t Make Your Startup Grow?
Dohmke, the CEO of GitHub, addressed the popular ‘Vibe Coding’ phenomenon. This term is coined by Andrej Karpathy, the OpenAI co-founder, to describe the approach to application development which includes relying heavily on the LLM (Large Language Model) to generate code. During the latest Q&A at Station F (Business Incubator for Startups), Dohmke warned that the startups cannot grow individually on ‘vibe coding’.
He argued that the non-technical startup founders may struggle to design sustainable companies without the involvement of skilled developers. He noted that they:
“can’t build a complex system to justify the next round.”
The prominent way to resolve this is to enable developers to select the most appropriate approach for every situation. It may include leveraging AI agents for maximum ROI or managing the tasks on their own.