As AI is changing coding, it’s also requiring the rethinking of team structures.
Andrew Ng — co-founder of Google Brain, founder of DeepLearning.AI, and one of the most influential figures in AI — has been making the case for a particular kind of team design that’s emerging from the realities of AI-accelerated software development. When you can build software ten or a hundred times faster, the old org chart doesn’t hold up.
The bottleneck, in his view, has simply moved. When engineering was the slow part, everyone else could more or less keep pace. Product managers could track what was being built. Marketers had time to figure out how to communicate it. Legal could review things before they shipped. But compress the build cycle from months to days — or days to hours — and suddenly every other function in the company is lagging behind. The product management bottleneck that people have been talking about is real, Ng says, but it’s just one of several that emerge when coding gets dramatically faster. Marketing scrambles to understand what engineers shipped. Legal, which might have comfortably signed off on something built over three months, now sits as a week-long wait on a product that took a single day to build. Design faces the same pressure.
His solution isn’t to throw more people at each of these bottlenecks. It’s to build very small teams — anywhere from one to ten engineers — and give them enough autonomy and context to handle things themselves.
The key phrase he uses is “high context, highly empowered generalists.” These aren’t engineers who dabble in other areas as a secondary concern. They’re people who are deeply technical but use AI to extend their reach across functions that would traditionally require dedicated specialists. An engineer on one of Ng’s teams might use AI to draft the first version of a terms of service, or write marketing copy for a feature they built. The work still gets reviewed — a lawyer polishes the terms of service before it goes out — but the loop from idea to something reviewable is dramatically shortened.
There’s a mathematical logic to this, one Ng spells out directly. If a product needs software engineering, product management, design, legal review, and marketing — five functions — and you have a team of two people, then by the pigeonhole principle, each person has to play more than one role. There’s no way around it. The question is whether each person is capable of doing that, and AI is what makes the answer yes.
Ng is candid that he’s not a marketer. Using AI doesn’t make him a good one. But it makes him less bad than he’d be without it — and in a small, fast-moving team, less bad is often good enough to ship and iterate. The goal isn’t to replace specialists permanently; it’s to avoid the situation where a two-person team is blocked for a week waiting for someone else to do something they could do adequately themselves.
This thinking connects to a broader pattern playing out across the industry. Broadcom CEO Hock Tan has described a similar dynamic: a senior engineer with AI can now accomplish in a week what previously required ten engineers over three months. Salesforce stopped hiring software engineers entirely in 2025, citing significant productivity gains from AI tooling. The underlying force in all of these cases is the same — the constraint has shifted from how fast you can write code to how fast everything around the code can move.
What Ng is describing is less a hiring strategy than an organizational philosophy. Wide guardrails, not micromanagement. Small pods with real ownership, not large teams with diffuse accountability. People who understand the product deeply enough to make decisions across functions, rather than people who escalate every cross-functional question up a chain. The teams he describes can “just run like crazy and build and ship code,” with the authority to make judgment calls on things — including marketing copy — that would traditionally require sign-off from another department.
It’s a model that Andrew Ng himself has argued will generate more jobs, not fewer — but different ones. The AI engineering jobs of the future, in his view, will look nothing like traditional software engineering, and many of them won’t sit inside traditional tech companies. The generalist pod is one version of what that future looks like in practice: technical people who are also, out of necessity and possibility, a little bit of everything else.