In a world where AI can generate entire applications with a simple prompt, the question of whether learning to code is still relevant has become increasingly urgent. GitHub CEO Thomas Dohmke, speaking at Microsoft Build 2025, offered a nuanced perspective on how programming will evolve—and why traditional coding skills remain crucial even as AI agents become more powerful.
The Magic Moment That Changed Everything
Dohmke's journey with AI-assisted coding began with skepticism. When he first encountered GPT models in 2020, his compiler background made him doubt that any AI system could properly handle the intricate syntax differences between programming languages. "I thought it didn't work, it wouldn't work, I doubted it," he recalled. "I couldn't believe that the model is able to keep the syntax of Python apart from Ruby or JavaScript."
But when GPT-3 and OpenAI's Codex proved capable of writing complete, syntactically correct methods for tasks like prime number detection, Dohmke experienced what he calls "magic." This skepticism-to-amazement transformation mirrors what millions of developers would later experience with GitHub Copilot, which now writes approximately 25% of code in files where it's enabled.
The numbers validated the impact: when GitHub internally tested Copilot before its public release, the Net Promoter Score hit 72—extraordinarily high for a developer tool that fundamentally changes workflow. "We thought that's going to be perceived more negative than it really was," Dohmke admitted. "In fact, the response was overwhelmingly positive."
Why Children Should Still Learn to Code
Despite AI's coding capabilities, Dohmke argues forcefully that children should absolutely learn programming. His reasoning goes beyond mere job security to fundamental literacy in an increasingly software-defined world.
"Software is everywhere," he explained. "Our cars today are mostly defined by software, our houses are dominated by software, travel is dominated by software." Just as we teach math and physics to understand the physical world, Dohmke believes coding literacy is essential for understanding the digital reality that surrounds us.
But there's a more immediate, practical reason: AI coding agents require human oversight. "The role of the engineer is now to verify what the agent has done," Dohmke emphasized. "How do I do that if we no longer have that understanding?" He painted a stark scenario where businesses could suffer security breaches or customer data loss because engineers couldn't properly evaluate AI-generated code.
This creates what Dohmke calls a new layer of "craft" in software development. Future programmers won't just need to understand traditional languages and logic—they'll need to master AI-assisted development as a core skill.
The Architecture of Tomorrow's Software
The boundary between deterministic code and AI-generated components is becoming increasingly blurred, but Dohmke sees this as evolution rather than revolution. He envisions developers working fluidly between two abstraction layers: traditional deterministic programming and the non-deterministic realm of natural language prompting.
"Part of the craft of a software engineer is going to be to be able to jump between the two abstraction layers," he explained. The process involves using AI for initial specification and ideation, then breaking complex problems into smaller, more manageable pieces that AI can implement reliably.
This hybrid approach addresses current AI limitations while leveraging its strengths. Complex requests like "build GitHub" will still overwhelm AI systems, but properly scoped tasks can be completed faster and more accurately than human-only development.
Beyond "Vibe Coding" to Professional Development
While acknowledging the appeal of "vibe coding"—the rapid prototyping enabled by AI tools—Dohmke distinguishes between hobbyist creation and professional software development. Real business applications require security reviews, code quality standards, testing, and compliance with team practices.
"In real software projects you still have to look for security vulnerabilities, code quality, the standards that your team has set," he noted. This creates space for what he terms "agentic DevOps," where AI handles routine tasks like bug fixes and test case generation while humans focus on architecture and creative problem-solving.
The goal isn't to remove human oversight but to eliminate tedious work. "Offloading that to an agent while I can then vibe with the agent mode on my local machine—that I think is the dream," Dohmke said.
The Open Source Gambit
In a significant strategic move, GitHub announced that Copilot's VS Code integration would become open source. This decision reflects both practical and philosophical considerations. Since VS Code itself is open source and Copilot's client-side code was already being reverse-engineered by curious developers, keeping it proprietary offered little benefit.
"The VS code is JavaScript, so it's not hard to reverse engineer co-pilot," Dohmke explained. "There have been blog posts since 2021 where people looked into what does it do, how does it build the prompt behind the scenes."
The open-sourcing enables community contributions while maintaining control over the valuable API layer. Developers can now fork Copilot, add new models, create migration tools for additional programming languages, or experiment with improved user interfaces—all while still using GitHub's backend services.
The Agent Ecosystem Vision
Rather than a single omnipotent AI assistant, Dohmke envisions an interconnected ecosystem of specialized agents. Personal agents would handle life management—photos, trips, entertainment recommendations—while work agents would manage company codebases and institutional knowledge. Task-specific agents from airlines or travel agencies would handle temporary needs.
The key innovation would be seamless integration between these agents, with smart governance determining what information belongs to which context. "When you leave the company, you disconnect them and then that knowledge that was part of your work goes with the company," Dohmke explained.
This distributed approach addresses both privacy concerns and practical limitations. Rather than training one massive model on all possible domains, specialized agents could excel in their specific areas while maintaining appropriate boundaries.
Addressing the Anxiety
To developers worried about AI replacement, Dohmke offers historical perspective. Just as the industrial revolution and personal computer adoption eliminated some jobs while creating others, AI will likely follow similar patterns. He points to software testing as an example—manual testing roles disappeared, but many testers transitioned into engineering or product management positions.
"Through the industrial revolution, through the personal computer, this has been true already," he noted. "When technology replaced roles, we have found new opportunities for those folks."
More optimistically, AI democratizes programming access. Language barriers disappear when you can code in German, Chinese, or any natural language. Children can now continue learning beyond their parents' technical knowledge by directly querying AI assistants.
The Endless Backlog Reality
Perhaps most importantly, Dohmke addresses a fundamental truth about software development: the work never ends. Every software company he knows maintains what he calls an "endless backlog" of features, technical debt, compliance requirements, and emerging standards.
"We're never running out of work," he emphasized. Rather than replacing developers, AI will help teams tackle previously impossible workloads while enabling entirely new categories of personalized software.
The future Dohmke describes isn't one where programmers become obsolete, but where they become more powerful—able to implement bigger ideas faster while maintaining the critical oversight that ensures software remains secure, reliable, and aligned with human needs. In this vision, learning to code isn't just career insurance; it's a fundamental literacy for participating in an increasingly programmable world.
Unlock the Future of Business with AI
Dive into our immersive workshops and equip your team with the tools and knowledge to lead in the AI era.
