GitHub CEO: Future Devs Manage AI Agents, Not Code Directly
The landscape of software development is undergoing a profound transformation, with artificial intelligence increasingly reshaping the daily work of coders. Thomas Dohmke, CEO of GitHub, a leading platform for developers, posits a future where the traditional act of writing code becomes largely automated. Instead, he envisions developers transitioning into roles focused on managing and overseeing AI agents that handle the bulk of the coding process. This shift, according to Dohmke, is not a distant fantasy but an accelerating reality, with most developers anticipating that AI will generate as much as 90 percent of their code within the next few years.
This evolving relationship between human developers and AI tools has been a gradual but decisive one, particularly since the widespread emergence of advanced AI chatbots like ChatGPT and Gemini. Dohmke elaborated on this progression in a recent blog post, detailing insights gleaned from interviews with 22 developers who have deeply integrated AI into their workflows. Initially, many developers approached these AI tools with a degree of skepticism, limiting their use to minor tasks and simple queries, largely due to the early models’ propensity for errors.
However, as large language models—the powerful AI systems behind these tools—grew more sophisticated and reliable, developers began to expand their application. They moved beyond trivial tasks, leveraging AI for more practical functions such as debugging code, generating boilerplate (standard, repetitive code structures), and creating useful code snippets. This increased familiarity paved the way for more complex interactions. Developers started using AI tools as brainstorming partners, tackling intricate problems and refining their instructions through iterative prompting—a process of continually refining commands to achieve desired outcomes.
The journey continued into a phase of deeper collaboration, where developers actively sought to understand the AI’s “thought process.” This involved engaging in discussions and conducting internal demonstrations to share effective prompts and innovative use cases. Ultimately, many developers reached a stage where they treated AI tools as indispensable partners. In this advanced phase, human developers increasingly rely on AI systems to write and curate significant portions of code. Their focus then shifts from direct code generation to the higher-level tasks of refining AI prompts and rigorously verifying that the generated code functions precisely as intended.
The pace of this change is striking. When Dohmke directly questioned developers about the prospect of AI writing approximately 90 percent of their code, their responses underscored a strong belief in its imminent arrival. Half of the developers interviewed expressed confidence that such a scenario is not only feasible but will materialize within the next five years. The other half held an even more accelerated view, predicting this level of AI integration within just two years. These projections highlight a profound paradigm shift, suggesting that the very definition of “coding” is on the cusp of a radical redefinition, transforming developers from code writers into sophisticated AI orchestrators.