We thought we were AI native. We weren't.

Emma Webb: One of the things I think is so wild is that when I tell people we have an OKR around becoming AI native, they're surprised. They'll say, "You're an AI company. Why do you need an AI-native OKR?"

Vinay Perneti: I thought we were already AI native until I realized we weren't AI native enough. One of my favorite parts of my role is talking with people on the team and understanding how their workflows are evolving. Back in late December and early January—what we now call the "November Opus moment"—I started noticing two groups emerging. One group was using agents inside the IDE. Another group was pulling away from everyone else. That second group had independently come to the same conclusion that it no longer made sense for them to write code directly. Instead, they were orchestrating agents. Once you're orchestrating agents, you're running multiple agents simultaneously. You can do more things in parallel, and you're willing to make changes in parts of the codebase that you might never have touched before. Even at Augment, where we've always leaned into AI adoption, I realized there was a difference between thinking you're AI native and actually being AI native. That forced us to define what AI native means. For me, it's simple: the first tool you reach for when you need something done is an agent.

Emma Webb: And it sounds like AI native isn't a destination. It's an ongoing process.

Vinay Perneti: Absolutely. When we first coined the term internally, having multiple agents running in parallel felt cutting edge. Today, that no longer feels AI native enough. To me, AI native now means agents operating on events and triggers. You have a system that knows when to involve humans and when it can run independently. I interviewed Eric Schmidt recently and asked what future organizational structures might look like. His answer was, "Do you need more than one person?" At the time it sounded like a joke. Now it doesn't seem that far-fetched. His point was that in a fully agentic software delivery system, you could start with a hypothesis, hand it to the system, have it implement the idea, instrument it, launch it, collect feedback, and iterate until the hypothesis succeeds or fails. So yes, AI native is going to keep evolving.

Emma Webb: If I don't have four or five agents running all the time, should I feel bad about myself?

Vinay Perneti: No. It's more nuanced than that. I'm a big believer in the Theory of Constraints. For most of software history, coding was the constraint. We're rapidly removing coding as the constraint. Once that happens, a new constraint will emerge. The job is always to improve the current constraint. But once coding is no longer the thing limiting your ability to create value, you need to step back and ask: "What actually matters now?" Maybe your agents shouldn't just be writing code. Maybe they should be solving problems higher up the stack.

Emma Webb: You're listening to *We Built What?*, the podcast for engineering leaders building in the agentic era. I'm Emma Webb from Augment Code, and today I'm joined by my colleague, Vinay Perneti, VP of Engineering at Augment. Vinay, thank you for joining me.

Vinay Perneti: Super happy to be here, Emma.

Emma Webb: I'm really interested in what it's like to have your job right now. You're leading an AI engineering team in a field that changes incredibly quickly. How do you do it?

Vinay Perneti: There's a phrase I love: "The future is already here. It's just unevenly distributed." I spend a lot of time looking for where it's unevenly distributed and then trying to bring those ideas to the rest of the team. That mindset helps because it means I don't have to figure everything out myself. I can learn from the team. The second thing is recognizing that everyone is going to make mistakes. Creating psychological safety—for yourself and for the team—is critical. As long as you're the fastest-learning team, you can win.

Emma Webb: I like that. The way through AI is through people.

Vinay Perneti: One hundred percent.

Emma Webb: Looking back six months, what would you do differently?

Vinay Perneti: One thing that really resonates with me is something Boris Cherny said. I don't think I fully appreciated the pace at which models were improving. When you're building AI products, you need to build for where models will be three months from now. At Augment we've always tried to stay on the bleeding edge, but I wish I had adopted that mental model sooner. Today, my assumption is that models have effectively solved coding. Six or eight months ago, that wasn't obvious. People pointed to bugs and failures and questioned whether the technology was ready. Today, I think most people would agree that coding is largely a solved problem.

Emma Webb: One thing we've seen internally is that engineers are simultaneously excited and afraid. They're optimistic and anxious at the same time. We ran a survey to see whether that was unique to Augment.

Vinay Perneti: And it wasn't. One thing I've learned from meditation is that the way you make a feeling go away is by paying attention to it. When we surfaced those fears internally, the team actually became more liberated. We stopped avoiding the conversation and leaned into it. The survey confirmed that those emotions exist well beyond Augment. Leaders should bring those conversations into the open.

Emma Webb: One thing we found was that 54% of engineering leaders are concerned or very concerned about losing shared understanding of the codebase. How do we solve that?

Vinay Perneti: A few months ago, I was one of those leaders. As we pushed harder toward agentic development, PR volume exploded. The temptation becomes rubber-stamping pull requests because reviewing everything becomes a bottleneck. That creates a dangerous feedback loop. As agents generate more code and agents review more code, humans lose visibility into how the system is evolving. Historically, code review served two purposes. First, it improved code quality. Second, it helped maintain shared understanding of the system. Once agents take over review, that second benefit disappears. So we had to evolve our workflow.

Emma Webb: What changed?

Vinay Perneti: Most code review products optimize for precision because they're designed for humans. They want to avoid false positives. But if the reviewer is an agent, the objective changes. You want to catch every possible issue. Internally we shifted toward what we call Deep Review. The goal is to catch as many potential bugs as possible. Then we run a separate risk analysis. The system decides whether a human actually needs to look at the PR. If it's low risk, we ship it. Today about 14% of our PRs are auto-approved. For the rest, we use something called Intent Review. Instead of reading every line of code, humans work with an agent to understand how the change affects the overall system. That preserves shared understanding while letting agents handle the detailed analysis.

Emma Webb: It's almost like the Socratic method with an agent.

Vinay Perneti: Exactly.

Emma Webb: Another thing we found is that engineers are worried about whether their skills will remain relevant. At the same time, most organizations haven't changed their hiring practices, promotion systems, or career ladders. How should leaders think about that?

Vinay Perneti: I think both sides will change. At Augment we've spent a lot of time thinking about the future talent profile. I don't believe the future is agents doing everything. At least not until we reach AGI. I think the future is small teams of humans working alongside large teams of agents. That means some skills become more important. Systems thinking. Architectural taste. Design judgment. Historically, only a handful of people focused deeply on architecture while others focused on implementation. Now is the perfect time to develop a stronger understanding of system design and engineering tradeoffs. Taste becomes increasingly important. Product taste. Architectural taste. Design taste. Those skills can be developed. You build them by exposing yourself to lots of examples and learning what good looks like.

Emma Webb: It's almost easier to develop taste now because you can generate multiple alternatives and compare them.

Vinay Perneti: Exactly. If I can look at ten possible solutions instead of one, I can learn much faster.

Emma Webb: You mentioned the agentic SDLC platform we're building. Why did we decide to build it?

Vinay Perneti: Because the current tooling is broken. Organizations were encouraging AI adoption, but most adoption happened at the individual level. Teams weren't actually becoming agentic. People were stitching workflows together with duct tape. That led us to realize teams need a platform designed for AI-native development. A few things matter. First, if your team is running hundreds of agents, how do you make sense of that complexity? Second, what does the equivalent of retrospectives and postmortems look like when agents are doing most of the work? Third, I strongly believe humans remain accountable. At least until we reach AGI. You need a platform that lets organizations define where humans should stay involved and where agents can operate autonomously.

Emma Webb: What parts of the platform are you most excited about?

Vinay Perneti: One is flexibility. You can run agents on your laptop, in a dev environment, in your own cloud, or in our cloud. The second is organizational intelligence. If someone on the team creates a valuable expert or workflow, everyone else should be able to discover it easily. And when humans interact with agents, the platform should learn from those interactions. The more the organization uses the system, the better the system becomes. That's exactly what we want from humans. So why shouldn't we expect the same thing from our agents?

Emma Webb: It sounds like the promise of AI.

Vinay Perneti: I think it is. Humans naturally fear the unknown. As leaders, our responsibility is to provide clarity wherever possible. Builders will always be builders. Maybe in the future we won't express ourselves primarily through code. Maybe we'll express ourselves through systems of agents. I'm optimistic. I think this world enables more people to build more things faster than ever before. And I'm excited to see what gets created.

Emma Webb: Thanks, Vinay. This was great.

Vinay Perneti: Yeah, super fun. Thanks, Emma.