It's an inventor's time: what AI native engineering looks like

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 here with Andrew Lau, CEO of Jellyfish. Andrew, it's so nice to have you.

Andrew Lau: Thank you for having me here. This is super fun.

Emma Webb: Andrew, what the heck is Jellyfish?

Andrew Lau: What is a jellyfish? Well, it's a flexible animal that offers transparency. But Jellyfish the company is an engineering management platform focused on helping organizations navigate AI transformation. Companies today are using four, five, six different AI development tools. They're getting some value out of them, but the big questions become: How do we get more value? How do we understand what's working for which teams and codebases? How do we measure impact? How do we understand token consumption? How do we evolve our teams to use these tools more effectively? That's the problem we're solving.

Emma Webb: Are engineering leaders freaking out?

Andrew Lau: I think the whole world is freaking out. If you're an engineer, you're freaking out. If you're an executive, you're freaking out. It's not just you. There's so much change happening right now. Some companies are moving incredibly fast, but even those companies are constantly surprised by what's happening around them. The technology world is moving so quickly that there's no stable ground. What was true yesterday isn't true today. At the same time, companies are trying to change everything: Their processes. Their businesses. Their products. There's a lot of fear driving urgency. If you channel it correctly, it can be productive. But for many people, it's chaotic. It's messy.

Emma Webb: Is there any upside, or is it all fear?

Andrew Lau: There's definitely upside. From a technology perspective, what's happening right now is incredible. It's unlocking creativity. It's making it easier to build things. Instead of waiting for someone else to solve a problem, you can solve it yourself. Instead of waiting for a deep analysis, you can often just create the thing and learn from it. The challenge is change. The condition itself can feel either scary or exciting depending on how you interpret it. Most organizations were operating one way before AI. Now they're realizing they can do things differently, but that requires changing how they work. And change is hard.

Emma Webb: I think part of the anxiety is the speed. People think, "I'd like to do that, but do I really have to do it this fast?"

Andrew Lau: Exactly. The feeling is that if you don't move quickly, someone else will. There's definitely some fear and some greed involved. People don't want to miss the next wave. You see this especially with AI coding tools. Something that seemed impossible in October suddenly becomes reality by January. The world keeps racing ahead. I have a friend at a large company who described it as an avalanche. The technology is happening whether you want it to or not. You can't really control it. You can influence it at the margins, but mostly you have to learn how to move with it.

Emma Webb: I often think of it as surfing. I'm trying to stay on my board.

Andrew Lau: That's actually a great analogy. To stay on the wave, you have to pay attention. You need awareness.

You need to understand: What models are changing? What tools are emerging? What behaviors are working? Leaders can't afford to be heads down anymore. You need to be heads up. You need to be agile. You need to be willing to shift your weight and adjust constantly.

Emma Webb: I think there are people who are perfectly happy getting a task list, completing the tasks, and moving on. Some of the anxiety around AI may come from realizing that isn't enough anymore.

Andrew Lau: That's a hard truth. Some of those roles may not exist in the future. If a task can be written down clearly enough, there's a good chance an AI system can do it. I don't control that. You don't control that. It's just happening. That's the avalanche. Ignoring it doesn't make it go away.

Emma Webb: You've mentioned token budgets a few times. Are companies actually having conversations about which teams get access to which models?

Andrew Lau: Absolutely. Especially large organizations.

People are asking: Which teams get access to premium models? Which teams can use lower-cost models? How much should we spend? At a startup in San Francisco, the answer is often: "Let it rip." But in a large enterprise, someone eventually gets the bill. And then finance starts asking questions.

They want to know: Who's spending money? What value are we getting? Do we need the most expensive model for every use case? Those conversations are happening everywhere.

Emma Webb: What about engineering teams specifically?

Andrew Lau: It's less about one team versus another. The bigger issue is that most companies didn't budget for significant AI spend. Then suddenly the tools got dramatically better. Everyone started using them. Finance teams are now looking at unexpectedly large bills and trying to figure out what happened. They're scrambling to develop budgeting and forecasting models for token consumption.

Emma Webb: Has anyone figured out a reasonable approach yet?

Andrew Lau: You hear everything. Some people argue token spend should exceed payroll costs. Others allocate a fixed monthly token budget per engineer. Some companies have no limits. Others have strict controls. The reality is that everyone is experimenting. The challenge is that AI became genuinely useful much faster than most organizations expected.

Emma Webb: Rob Zuber told me that once you teach engineers to work this way, it's difficult to tell them: "Sorry, you've hit your token budget. Go back to writing everything by hand."

Andrew Lau: Exactly. That's a difficult leadership problem. You want to encourage modern behaviors. At the same time, businesses need financial discipline. I think things will improve as tooling matures. Right now most people don't really understand how their token usage translates into cost. They don't have the visibility they need.

Emma Webb: You've talked about a framework you use: adoption, productivity, and outcomes.

Andrew Lau: It's simple, but it's effective.

First: adoption. Are people actually using the tools? Nine months ago the question was binary. Are you using AI or not? Now it's more nuanced. How often are you using it? How many tokens are you consuming? How much code is being generated?

Second: productivity. Are you moving the metrics that matter? Throughput. Cycle time. Deployments. Pull request velocity. Whatever your organization cares about. It's not enough to use the tools. The metrics have to improve.

Third: outcomes. Are you achieving the business result you intended? Maybe you're using long-running agents. Maybe deployments are increasing. But are customers actually using what you're building? Are you solving problems? Are you creating business value? That's the ultimate question.

Emma Webb: One thing I've been thinking about is that AI dramatically increases an organization's ability to build. That puts pressure on strategy. If everyone can build more things faster, do they know what direction to run in?

Andrew Lau: I think that's exactly right. This era requires more product thinking, not less. Just because you can build something doesn't mean you should. You need clarity around goals. You need clarity around priorities. You need clarity around what success actually looks like.

And here's a more controversial question: Do your customers even want more? Just because you can ship ten times more features doesn't mean customers want ten times more features. You can end up creating noise. More output doesn't automatically mean more value.

Emma Webb: That's such an important distinction.

Andrew Lau: And it connects directly to distribution. When everyone can build, everyone starts shipping. When everyone starts shipping, distribution matters more. Attention becomes scarce. Reaching the right audience becomes increasingly important.

Emma Webb: It's impossible to remove a major constraint like software development without changing everything else around it.

Andrew Lau: Exactly. And I think we also need to rethink roles. The distinction between product and engineering isn't some law of physics. It's something we created. Maybe we need to think more about jobs to be done. Who's listening to customers? Who's defining priorities? Who's building? Who's accountable? The roles themselves may evolve significantly.

Emma Webb: You spend all day talking to engineering leaders. You have data on who's succeeding and who's struggling. What do the organizations that are succeeding have in common?

Andrew Lau: There isn't one definition of success. But there are patterns. Some companies have broad adoption. The entire organization is using long-running agents and modern workflows. Others have strong pockets of excellence where a few teams are operating at the frontier. Some are seeing dramatic improvements in engineering metrics. One trend we've been tracking is agent-generated pull requests. Last year, even the most advanced companies might have had 3% of pull requests generated through agentic workflows. Today, the top 10% of organizations are around 20%. That's a massive shift in a very short period of time.

Emma Webb: That's incredible.

Andrew Lau: The companies leading this transition are changing behavior, changing process, and changing expectations. They're getting entire organizations to work differently. That's hard. Especially if you're an established company. A startup can build AI-native workflows from day one. A large enterprise has customers, legacy systems, existing processes, and organizational history. Transformation is much harder when you already have something valuable to protect.

Emma Webb: That's the paradox. You want the customers. You want the distribution.

Andrew Lau: Exactly. It's not a disadvantage. It's just a different challenge.

Emma Webb: Andrew, is there anything we didn't talk about that you think engineering leaders should hear?

Andrew Lau: I think we're living through a genuinely magical moment. It's incredibly dynamic. It's changing fast. And yes, some of what we've said today may be outdated in two weeks. But fundamentally, I think this is a builder's moment. It's a chance for people who haven't built something in years to start making things again. It's a chance to rethink how companies work. It's an inventor's time. It's a maker's time. I think it's a great time.

Emma Webb: Andrew, thank you so much for joining me.

Andrew Lau: Thank you for having me. This was a blast.