GitHub CEO: Why Now Is the BEST Time to Be a Developer | Thomas Dohmke
Marina Mogilko (host), Thomas Dohmke (guest)
In this episode of Silicon Valley Girl, featuring Marina Mogilko and Thomas Dohmke, GitHub CEO: Why Now Is the BEST Time to Be a Developer | Thomas Dohmke explores gitHub CEO explains vibe coding, AI agents, and developer future Dohmke defines “vibe coding” as working in an IDE with AI agents (Copilot/Cursor-style) where you mostly direct the agent and run commands rather than constantly reading code.
GitHub CEO explains vibe coding, AI agents, and developer future
Dohmke defines “vibe coding” as working in an IDE with AI agents (Copilot/Cursor-style) where you mostly direct the agent and run commands rather than constantly reading code.
He argues you can build meaningful apps via prompting (auth, settings, basic web apps), but as complexity, performance, and scaling needs rise, understanding code becomes unavoidable.
He predicts more developers overall: AI lowers the barrier for newcomers and “consumer developers,” while elite teams use AI to build far more ambitious products—so the smartest companies will hire more, not fewer.
He frames job anxiety as an adoption problem: the best defense is to learn AI deeply, become the “orchestrator” of agents, and pair acceleration with responsible AI practices (security, guardrails, red-teaming).
Key Takeaways
Vibe coding is “directing agents,” not traditional coding.
Dohmke describes it as using agent mode in tools like Copilot/Cursor where the developer mainly interacts with the AI, approves actions, and executes commands—often without reviewing every line.
Get the full analysis with uListen AI
Prompting can take you far, but not all the way.
You can build real functionality (web pages, auth, settings), but when systems become slow, fragile, or need to scale (e. ...
Get the full analysis with uListen AI
No-code + AI won’t automatically create billion-dollar moats.
If anyone can replicate a product by prompting in minutes, differentiation collapses; durable businesses will still require deep technical decisions, unique execution, and increasingly complex products.
Get the full analysis with uListen AI
AI will expand the backlog, not erase it.
He rejects the idea that companies will “run out of work,” arguing AI enables more features and bigger ambitions—creating more total engineering work even if per-feature effort drops.
Get the full analysis with uListen AI
“90% of code by agents” can still mean steady (or more) developer demand.
His math: if total code produced grows ~10x, agents may write 9x and humans still write ~1x—the same absolute amount as before—while humans shift to higher-level direction and review.
Get the full analysis with uListen AI
The smartest companies will hire more developers to compound productivity.
If AI makes one developer 10x more productive, then adding more developers can scale output dramatically (e. ...
Get the full analysis with uListen AI
Job fear is best addressed by becoming the AI power-user/orchestrator.
Dohmke’s advice is to “learn with AI,” adopt it early, and position yourself as the person who can safely and effectively coordinate multiple agents—especially as firms mandate AI use to remain competitive.
Get the full analysis with uListen AI
Notable Quotes
“Vibe coding means you open your IDE… go into the agent mode… you’re mostly focused on interacting with the agent, and not so much of what the code actually is doing.”
— Thomas Dohmke
“You can get as far as you’re having the patience to keep prompting.”
— Thomas Dohmke
“The idea that AI without any coding skills lets you just build a billion-dollar business is mistaken, because if that would be the case, everyone would do it.”
— Thomas Dohmke
“The companies that are the smartest are going to hire more developers. Because if you 10X a single developer, then 10 developers can do 100X.”
— Thomas Dohmke
“I believe 90% of all code is going to be written by agents.”
— Thomas Dohmke
Questions Answered in This Episode
Where exactly is the “handoff point” where vibe coding stops working—debugging, architecture, security, or scaling—and how can beginners recognize it early?
Dohmke defines “vibe coding” as working in an IDE with AI agents (Copilot/Cursor-style) where you mostly direct the agent and run commands rather than constantly reading code.
Get the full analysis with uListen AI
You argue AI-only builders won’t create moats because everyone can prompt—what are concrete examples of moats that still matter in an agentic software world (data, distribution, compliance, infra, UX)?
He argues you can build meaningful apps via prompting (auth, settings, basic web apps), but as complexity, performance, and scaling needs rise, understanding code becomes unavoidable.
Get the full analysis with uListen AI
If companies mandate AI usage, what new hiring signals should candidates show (agent workflows, evals, secure prompting, code review discipline) beyond classic algorithm interviews?
He predicts more developers overall: AI lowers the barrier for newcomers and “consumer developers,” while elite teams use AI to build far more ambitious products—so the smartest companies will hire more, not fewer.
Get the full analysis with uListen AI
How should developers change their workflow when they “don’t review the code all the time” without increasing security and reliability risk?
He frames job anxiety as an adoption problem: the best defense is to learn AI deeply, become the “orchestrator” of agents, and pair acceleration with responsible AI practices (security, guardrails, red-teaming).
Get the full analysis with uListen AI
Your 90%-agents prediction assumes total code grows ~10x—what evidence suggests demand will scale that much, and in which product categories will it grow most?
Get the full analysis with uListen AI
Transcript Preview
What would you say to coders who are learning how to code right now? This is Thomas, CEO of GitHub, the world's largest platform for developers with over 100 million users. Under his leadership, GitHub Copilot became the most widely adopted AI coding tool in history. We see big companies put a stop on hiring. In two years, do you think I would need a developer?
The idea that AI without any coding skills, lets you just build a billion-dollar business is mistaken, because if, if that would be the case, everyone would do it.
He led GitHub's $7.5 billion integration with Microsoft. Now he's shaping the future of coding itself. So you're not scared?
The dream of software development was always that I can take the idea that I have in my head on a Sunday morning, and by the evening, I have the app in the- up and running on my phone.
Thanks to HubSpot for sponsoring this video. Hey, guys, welcome to Silicon Valley Girl. We're here at Viva Technology in Paris, and I have Thomas, the CEO of GitHub. I am so excited to talk to you about what's going on in coding. First of all, let's define vibe coding. For everyone who's heard this term, and they're like, "What's, what's going on? What's happening?"
Ooh, that's a tough question to start with. I think the, you know, the loosest interpretation from my side is that vibe coding means you open your IDE, you know, like, like Copilot or Cursor, Windsurf, any of these, and you go into the agent mode, and you give it a task to do, and then you're just following along of what the agent proposes to you, and you run the commands. And you're mostly focused on interacting with the agent, and not so much of what the code actually is doing.
And you don't have to-
Like, you're not reviewing the code all the time.
Yeah, and you don't have to learn how to code. 'Cause I tried-
Yeah
... GitHub Copilot.
Yeah.
I was just chatting with it.
Yeah.
I'm like, "Create this website, do this," and then it just tells me where to put the code.
Yeah.
So it starts working.
Yeah.
So how complicated can the website get with vibe coding? Can I build something that has a database?
Yeah.
Or is it just, like, a landing page or a very simple app?
My rule of thumb would be you can get as far as you're having the patience to keep prompting, because, you know, as you said, if you don't understand what the agent is actually writing, what the code looks like, well, then your only way of modifying, you know, the functionality is by figuring out how to prompt. It almost becomes a, your, your, a, a quiz-
Art
... or like a game-
Yeah
... right? Like, where you're trying to-
Yeah.
"Okay, so, hmm, let me try a different approach." I like to compare this to image models, all right? You start with a simple prompt, and you render an image of Paris, and then almost certainly you get something which isn't exactly what you expected. And then you start, you know, rewriting it, and for some time there were, like, tricks to do that in, in Stable Diffusion and Midjourney, uh, by th- doing things like training on Artstation. Uh, and then you got closer to what you wanted, but at some point-
Install uListen to search the full transcript and get AI-powered insights
Get Full TranscriptGet more from every podcast
AI summaries, searchable transcripts, and fact-checking. Free forever.
Add to Chrome