PRODUCT

GitHub's secret sauce for the future of coding

GitHub has 180 million users on its platform. A new user joins every second. But the company is not content being the world’s largest code repository. It wants to be the operating system for a new era of software development where AI agents do the heavy lifting and a billion people can build software using nothing more than natural language.

In a conversation with The Deep View, GitHub Chief Product Officer Mario Rodriguez laid out the company's vision that goes beyond code completion. He described a future in which developers wake up with an idea, delegate it to five AI agents, steer the work at a high level and push to production without ever assembling a traditional engineering team. He also explained how GitHub’s new “repository intelligence” layer could give those agents the memory and expertise they need to actually get the job done.

Here are the highlights from the conversation.

Jason Hiner: We are seeing so much buzz and interest around Claude Code right now. It’s even starting to transcend the tech industry. Are you seeing something similar? And when you think about GitHub Copilot and repository intelligence, is vibe coding part of the plan?

Mario Rodriguez: We have a narrative that we unveiled. We said the goal is to get to one billion developers. That’s 10 percent of the world population by 2030. There’s no amount of scaling that you could do to create a billion developers the traditional way. So then what does it mean? It means you have to redefine what a software developer is.

I know how to cook, but I didn’t open a Michelin-starred restaurant. I know how to bike, but you didn’t see me in the Tour de France. I know how to swim, and I wasn’t at the Olympics. [Similarly], a developer cannot be just the [experts]. The beauty of what is happening right now is the fact that we have technology that can make [coding] accessible. You call it vibe coding. Before it was called low code. But the big change is we have the power of natural language to describe these things.

Jason Hiner: I’ve seen people I know who previously had no idea about coding talking to me about how, in one weekend, they coded an app. And they said, "I made it available on GitHub if you want to see." I’m impressed that they even have a GitHub account. How do you get more of that?

Mario Rodriguez: That’s awesome. We need to make that super seamless for them. And mobile is the other big thing. If you think about a billion developers, you’re not going to do it with a Mac and a PC everywhere. But everyone has a phone. So if you go right now and download our mobile app, we’re trying to make it super seamless to [serve] them [on mobile]. You have an agent right there that you could actually create an app with and then publish it to GitHub.

TOGETHER WITH CODER

Secure environments where devs and agents work in parallel.

Coder Workspaces is an open-source platform where developers and AI agents collaborate securely in self-hosted environments on your infrastructure.

Whether you're a team of 1 or 10,000, Coder fits the stack you already use. Scale AI agents, bring any IDE, deploy anywhere, and stay in control, all in one governed workspace.

Get started with Coder open source today.

SECRET SAUCE

Repository intelligence: Giving agents a memory

Jason Hiner: This idea of repository intelligence is really where GitHub sees itself as a layer above even all of the different coding assistants, a way for you to provide added value, knowing that companies and coders are using multiple coding assistants. What does repository intelligence mean and what can it offer?

Mario Rodriguez: Today, all of the world’s code lives on GitHub. We see all of the commits that are happening. We see all of the pull requests. We see the issues. So there are all of these artifacts that come into play as you develop in a team context, and all of those go through GitHub.

Repo intelligence is the ability to say, look, it’s not just the code. It’s the entirety of the repo and all of the artifacts that go into creating a feature, and then being able to index that, provide midterm and long-term memory on it, and really start thinking through it as a graph of work for software development.

Imagine if I’m joining a new company and a new code base. I could go in and not only look at the code, but say: What were all the decisions that came into being? How did they develop it? What were the prompts they used? Did the quality get better or not, and when it didn’t, why not? What was super successful that ended up impacting the business? We did a migration to this library, then migrated away. Why? That’s what repo intelligence is all about. It’s that institutional memory plus expertise at any point in time. And then imagine every agent being able to query that and utilize it to get something done.

TOGETHER WITH CODER

OpenClaw meets Coder: One engineer's blueprint for running agents without the risk

AI agents with unrestricted system access are a liability. One Coder engineer fixed that by giving OpenClaw its own isolated, governed workspace using Coder Tasks, scoped tooling, and predefined repo access.

The result is a concrete, repeatable pattern for running agents where they can do real work without touching anything they shouldn't. If you're building agentic workflows, this is the blueprint.

Read how one engineer built a safe, repeatable pattern for running OpenClaw in governed workspaces and why it matters for agentic infrastructure.

THE PATH AHEAD

The idea-to-production future

Jason Hiner: Can you tell me about how the idea-to-production workflow is growing more valuable?

Mario Rodriguez: The way software development usually works is: you have an idea, you put it in a ticket, you discuss it, you design it, and then you get five developers assigned to it. The future is very different. You could wake up one day and say, I have an idea. I go in, I have five agents implement this with me. I iterate on it. And then I push it to production. That was just you. You didn’t need five more people. You just needed five expert agents collaborating with you.

If these agents are experts in the design system of that code base, experts in the testing, experts in deploying it, then why is it not possible for you to wake up, have an idea, collaborate with an agent for maybe three days with long-running tasks constantly happening? You do what we sometimes call macro-delegation and micro-steering. Here’s this task, and I’ve already put it in the right context. And oh, this agent needs me for a second [for clarification]. Let me help it out and then go back to the macro world.

Jason Hiner: What else needs to get better for the next generation of AI-powered coding? What are some of the friction points?

Mario Rodriguez: Quality. Validation and verification, and then quality of the output. That is definitely core. We released a universal protocol for code quality to be at the forefront of this. Imagine: in Copilot, the AI generates code, and prior to committing, it asks, does this have any security vulnerabilities? We do a scan, and if it does, it automatically tells Copilot to revise its output. We have this kind of judge, always with AI, making sure we’re creating the most high-quality and secure code. Because lines of code won't make a difference. What’s going to make a difference is high-quality code that achieves the intent of that human.

Jason Hiner: Are agents and the agentic experience going to be built on top of these coding copilot applications? Is that going to be one of the first ways people start having agents working for them?

Mario Rodriguez: Yeah, 100 percent. And what we’re missing there is this intelligence piece for it to really be unlocked. If every time you go and do something, you forgot how you did it previously, that’s not going to be really good. So repo intelligence is going to be critical in the future, for both software developers and information workers.