Building Has Become Cheaper; Choosing Has Become More Expensive

Building has become cheaper; choosing has become more expensive. This is not an investment quote—it is Andrew Ng's prescription for AI engineering teams.

DeepLearning.AI's The Batch newsletter published Andrew Ng's latest letter. He offered a framework for VP Engineering and engineering managers, with four Ps: People, Projects, Platforms, Processes. The goal is not to create a 'traditional team that knows how to use AI,' but to redesign a native team around AI capabilities.

The core judgment is in this sentence: building is now fast, so choosing what to build is the bottleneck. When code can be written quickly by agents, the truly scarce capability becomes 'deciding what to write.'

People. Individuals need to catch up first. Ng says you cannot learn to ride a bike by watching others; you have to get on and ride yourself. Engineering leaders need to write something and deploy something themselves, even if only occasionally. This is not just about experiencing life—after getting hands-on again, both technical judgment and business judgment become more accurate. Coding agents now make this path feasible again; many managers who have been away from the front line can consider returning.

Projects. The project portfolio needs a growth mindset. Bottom-up innovation should be invested in, and senior leaders also need to pick a few key projects and push them forward. Ng makes a point here—too many companies treat AI mainly as a cost-saving tool. Reducing costs and increasing efficiency is fine, but the opportunities for new products, new users, and new growth are much larger. If the project list contains only cost cutting and no growth experiments, that is wasting AI.

Platforms. Data needs to be turned into a form agents can use. Repositories, documents, tickets, design decisions—whatever can be opened to agents should be opened. Internal tools, CI pipelines, sandbox environments—these infrastructure pieces need to let agents run. Ng puts it more concretely: for some companies, the first step is to convince the CIO to stop making developers write code purely by hand. Technical obstacles are often not in the technology itself.

Processes. Processes should be reset around the new bottleneck. The cost of writing code to clear specs has plummeted, and the remaining bottleneck is almost entirely 'defining the specs.' So engineers need to participate in product decisions; teams need sufficient information and real decision-making authority, with guardrails only for risks such as leaks, outages, and data loss.

After this framework was published, someone in the comments asked: In meetings that decide what to do, is AI part of People or Processes? Others asked: When prototypes are almost free, what is the mechanism for screening ideas? These questions have no standard answers, but they all point to the same fact—choosing is becoming a new management challenge.

Ng added at the end: If you have a boss, forward this email to them. This is probably the common wish of all engineering leaders.

Original link