How AI is changing the engineering leader's job

Emma Webb: What are we going to talk about, Rob?

Rob Zuber: All things agentic, apparently.

Emma Webb: Rob and I go way back. We used to work together at CircleCI.

Rob Zuber: And before that.

Emma Webb: And before that.

Rob Zuber: Long before the agentic era.

Emma Webb: Long before that. That was the SaaS ecommerce era. The Facebook feed era.

Rob Zuber: As it transitioned into the mobile-first era.

Emma Webb: That's true. What has your role at CircleCI changed with AI?

Rob Zuber: That's a really good question. I've been at CircleCI almost 12 years. My role has changed a lot as the company grew from around 14 people to a much larger organization, but the rate of change now is much more significant. One thing I think about a lot is that engineering leaders live off their expertise. You're able to coach people because you've done the job. You've used the tools. You understand what matters. And then suddenly AI shows up and I have no expertise. I'm basically saying, "I heard online that AI is cool. You should use some of that." I understand the fundamentals of software because I've been doing this for a long time, but I don't know how to use all these new tools. There's a huge gap. As a leader, that pushed me back into a very hands-on mode. I had to learn alongside the rest of the organization because I couldn't rely on expertise I'd already accumulated. The ability to coach people through getting better outcomes becomes difficult when you've never gotten those outcomes yourself. So I think there was an instant credibility and expertise gap created for leaders, and my response was to get hands-on and start building.

Emma Webb: One thing I think about is that some leaders are deeply technical and love building. Others aren't. Does that change how they approach this moment?

Rob Zuber: Absolutely. I've always loved technology. I've always had side projects. Not everyone is like that. But one thing that's happened is that AI dramatically lowered the activation energy required to build things. For a lot of engineering leaders, there are all these little problems they've wanted solved for years, but they haven't built anything in a decade. Frameworks changed. Tooling changed. Everything feels unfamiliar. Before, you'd have to ask someone on your team: "Could you build this thing for me?" Now you type it into a chat box and the thing gets built. Even if your knowledge of modern frameworks isn't current, you still understand software fundamentals. That's enough. I think that's created a lot of excitement among engineering leaders because suddenly they're solving problems that have been sitting on their mental backlog for years.

Emma Webb: At the same time, there's a huge spectrum of AI adoption.

Rob Zuber: Exactly. On one side, you have people who sound almost fanatical. On the other side, you have people who are deeply skeptical. Helping people understand the opportunity, understand what AI means inside your organization, and understand the future you're trying to build—that's a huge part of the job right now.

Emma Webb: Even at Augment, where we build AI tools for engineers, we have company goals around becoming AI native. Partly because the definition of best-in-class keeps moving. If you're comfortable using AI autocomplete today, that's great, but next week there's something new and the bar has moved again. There's really no such thing as "done."

Rob Zuber: Exactly. You're AI native at noon today. By the time we all go home, you're behind the pack. That's what makes this difficult. The language around AI is also incredibly dramatic. Every announcement sounds like it's going to change the world forever. For a software engineering audience that's used to pragmatic, evidence-based language, that can be exhausting.

Emma Webb: How do you sift through that?

Rob Zuber: I was actually working on a writing project recently. I used Claude to help me organize my thinking for an internal strategy document. The process was great.

Then I read the final draft and thought: "This is incredibly melodramatic." Everything sounded world-changing. Maybe that's why all AI marketing sounds the way it does. People are writing with LLMs. I sent it anyway because the content was useful, but it definitely sounded more dramatic than I would have written myself.

Emma Webb: Let's talk about CI. I've seen people saying AI makes CI obsolete. Is the death of CI greatly exaggerated?

Rob Zuber: I actually see the opposite. I see people saying CI is more important than ever. The value of feedback and validation has gone up dramatically.

We're moving toward a world where: * Agents create code * Agents review code * Humans are involved less and less As we remove human review steps, we need stronger automated validation. Humans are inconsistent. One day they're focused. Another day they're distracted. So if we're taking humans out of the process, we need systems that provide confidence that what gets shipped is actually correct. That means stronger feedback loops.

Emma Webb: What does that look like?

Rob Zuber: We're pushing validation earlier and earlier. For example, before an agent even tells me it's done, it's already running much of the same tooling that would traditionally run in CI. Some of it runs locally. Some of it runs in sandboxed environments. Instead of waiting 10 or 15 minutes for a full CI pipeline, we're getting targeted feedback in seconds.

The agent sees: "This broke." "This test failed." "This validation didn't pass." And it keeps iterating automatically. The agent doesn't stop. It just keeps fixing issues until everything passes.

Emma Webb: So CI becomes part of the agent loop?

Rob Zuber: Exactly. CI becomes a continuous source of feedback rather than a gate at the end. That feedback is far richer than simply saying: "Try again." We have a huge amount of context about why something failed and what needs to be fixed. That makes the system much more efficient.

Emma Webb: Speaking of efficiency, let's talk about cost.

Rob Zuber: That's a big one. There's this weird tension where people say: "Don't worry about efficiency. Costs will come down." At the same time, engineering leaders are saying: "We already blew through our AI budget." Those two conversations are happening simultaneously. Clearly efficiency matters. I look at usage leaderboards and see two engineers doing roughly the same amount of work, but one is spending ten times as much as the other. That tells me we need to understand how people are using these tools. They're incredibly powerful, but it's also very easy to generate runaway spend.

Emma Webb: That's interesting because not many people have brought up cost.

Rob Zuber: Leaders have to balance two competing pressures. On one side, everyone wants to move faster and become AI native. On the other side, budgets are real. You can't spend endlessly.

And you definitely can't tell everyone: "Congratulations, we've run out of AI budget. Go back to typing code by hand." That's like taking coffee away from engineers.

Emma Webb: Are you hiring differently?

Rob Zuber: Absolutely. We're much more interested in whether people are excited about this future and willing to work differently. Traditional technical interviews often focus on memorization. Can you hand-code a linked list? Honestly, I don't care. I care whether you can use these tools effectively. Can you evaluate what they produce? Can you tell the difference between good and bad implementations? Can you harness these tools to produce great outcomes? That's a much more interesting question than whether someone can reproduce a data structure from memory.

Emma Webb: Anything you've been thinking about that we haven't talked about?

Rob Zuber: One thing I've been thinking about a lot is constraints. People talk about bottlenecks and optimizing processes. But I think many organizations evolved around a world where coding was expensive. We built layers and layers of process to avoid making expensive mistakes. Requirements documents. Approvals. Sign-offs. Review processes. All of that existed because writing software was expensive. Now the cost of creating software has dropped dramatically.

So I don't think the right question is: "Which step in the existing process should we optimize?"

I think the better question is: "Why does this process exist at all?" Many of these processes were responses to constraints that no longer exist.

Emma Webb: That's a big shift.

Rob Zuber: Exactly. When software becomes dramatically cheaper to create, you can revisit the entire system. Instead of debating which of three ideas is best, maybe you build all three and see what customers actually want. Roles start changing too. Are you a designer? Are you an engineer? Maybe. Maybe not. If I have a teammate who's better at design than I am, I'll ask for feedback. But I'm not necessarily waiting for them to create a Figma file before I start building. We just build. We experiment. We learn.

Emma Webb: So roles become more fluid.

Rob Zuber: Much more fluid. Honestly, I'd love to put people together and say: Pretend you don't have a title. Pretend I don't have a title. Let's just work together and see what happens. Then we observe what works. Those patterns emerge naturally. That's what Agile was supposed to be. We weren't supposed to design the perfect process in advance. We were supposed to learn our way into it. I think that's what we're doing again. We're not going to invent the future way of working. We're going to live it first. Then we'll document it afterward.

Emma Webb: And then we'll make everyone else do it.

Rob Zuber: Exactly. Then we'll write it down. Then I'll become an AI ways-of-working consultant.

Emma Webb: I love that for you. Rob, it's been a treat.

Rob Zuber: Thanks for having me.