By James Abela
When Thomas Dohmke first saw GPT-3, he didn’t believe it. The syntax, the accuracy, the apparent magic of it all—surely a language model couldn’t write real, working code? But then it did. And that single moment, that blink of disbelief turned wonder, was the spark that has lit up the entire world of software development.
In a recent conversation with tech YouTuber Matthew Berman, Dohmke painted a bold, thrilling picture of what comes next for programming. Spoiler: it’s less about typing lines of code and more about collaborating with AI to bring ideas to life faster than ever before.
From Doubt to Deployment: The Rise of GitHub Copilot
Long before Copilot hit the public stage, it was being tested inside GitHub. “It was writing 25% of the code,” Dohmke recalls, still amazed. Developers were hesitant at first—but soon discovered they were more productive, staying in the all-important flow state. It felt like magic, yes, but it was also grounded in smart UX design and years of developer habits, like tab-completion and IntelliSense.
Copilot is now a daily tool for millions—and just recently, GitHub took the radical step of open-sourcing its client for Visual Studio Code. “We’re giving back to the community,” Dohmke says, with evident pride. This isn’t just a gesture; it’s a signal that the next wave of AI-powered development will be deeply collaborative.
Should We Still Learn to Code?
Dohmke’s answer is a resounding yes.
Even if AI can build an app for you, understanding how it works remains critical. “It’s like learning maths,” he explains. “You may not become a physicist, but you need to grasp the fundamentals.” Especially as coding agents grow in power, verifying what the machine produces becomes just as important as writing the code yourself.
From Vibe Coding to Agentic DevOps
One of the most fascinating ideas from the interview was “vibe coding” — a term that captures how modern developers increasingly vibe with AI tools. You describe a problem, and the system builds something close to your vision. This works brilliantly for prototyping. But as Dohmke is quick to point out, serious software still needs oversight, testing, and security.
Enter “agentic DevOps”—a term that sounds futuristic but is already underway. AI agents won’t just help you code; they’ll run tests, scan for vulnerabilities, and even suggest refactors. You stay in control, but with a co-pilot who never sleeps.
The App That Disappears
Dohmke envisions a world where apps aren’t installed but generated on demand. Think: your child wants to track their weekly allowance. You don’t download a generic app; you generate a mini-app, just for them, customised in minutes. It’s ephemeral. It’s personal. It’s the software equivalent of bespoke tailoring.
As AI improves, this vision of just-in-time applications could redefine what we think of as software.
Will AI Replace Developers?
It’s the question everyone’s asking—and Dohmke’s answer is refreshingly honest.
Some roles will be automated. Translation, for example, may soon be handled entirely by real-time AI. But rather than making jobs obsolete, AI will expand what’s possible. More people than ever will be able to become software developers, regardless of background, language, or location.
“The backlog is endless,” he laughs. “AI lets us finally tackle it.”
A Connected Agent Ecosystem
Dohmke also foresees a future not of one super-agent but many—personal, professional, task-based—each specialised, each interoperable. Your personal agent might help you pick dinner. Your work agent manages your codebase. And when you change jobs? The work agent stays behind. The personal agent comes with you.
That vision of personalised, ethical, compartmentalised AI could be the key to trust in an AI-first world.
Final Thoughts
Thomas Dohmke doesn’t see AI as the end of coding—but the beginning of something better. A world where coding is less about fighting compilers and more about expressing ideas. Where more people can build. Where more ideas can thrive.
The future of programming isn’t just AI-assisted—it’s human-guided, community-built, and, in many ways, just getting started.
