How AI Engineers Decide What to Build: Four Skills for Shaping the Build

A Structural Shift: Roles Are Blurring

Andrew Ng describes a fundamental change in how technology companies operate. Traditionally, product managers and designers defined what to build, developers implemented it, and project managers drove timelines. With AI tools amplifying individual building capacity, these boundaries are thinning — not because one role replaces another, but because each is expanding into the other's territory. Developers now participate in product decisions; product managers and designers gain engineering skills. This convergence accelerates development speed because engineers no longer need to wait for a PM to clarify every next step.

When you truly master AI engineering, your best work is no longer just implementing a product someone else defined, but actively participating in and shaping the entire build process.

Ng identifies the final, most upstream skill in the AI Engineering Skills Map: Shaping the Build — the ability to decide what deserves to be built and to what degree.

Skill 1: Driving the Build Loop — Bias for Action at AI Speed

Software is built through a loop: write code → get feedback → decide the next step. A skilled AI engineer drives this loop by repeatedly deciding what to do next at the high velocity AI enables. Ng emphasizes a bias for action : because AI drastically lowers the cost of each iteration, the frequency and quality of "what's next" decisions directly determine project velocity. Engineers who proactively drive the loop gather feedback and adjust direction faster than those who wait for instructions, even if the AI agent itself runs quickly.

Skill 2: Making Product Decisions — You Don't Need to Be a PM, But You Must Decide Beyond the Spec

Ng reassures developers: you don't have to become a product manager. However, you must make the decisions that a product spec doesn't cover, and if no spec exists, you should be able to write one. He breaks product decisions into three interdependent "senses," all rooted in user empathy :

Product sense : Choose a direction that satisfies real user needs without waiting for a PM to decide everything.

Design sense : Produce experiences that are not merely functional but comfortable to use.

Business sense : Reason about go-to-market, market size, unit economics, and P&L to make economically sound trade-offs.

Ng provides a practical ladder for continuously sharpening empathy, from lightweight qualitative methods to large-scale quantitative validation:

Quick informal interviews (2–3 users): Lightest, fastest first-hand insight.

Surveys (hundreds of users): Broader coverage, still quantifiable.

Large-scale A/B tests (many users): Validate hypotheses with real behavior.

Behavioral data analysis (thousands to millions of users): Read truth from scaled behavior.

You don't need to wait for a full research report to "understand users." Talking to 2–3 users for 20 minutes often yields more useful insight than a week of second-hand data.

Product sense is not innate talent; it is a deliberate practice loop fueled by these methods.

Skill 3: Communicating and Leading — Your Stage Is Bigger Than Code

AI engineering skills let you engage in a wider scope of work — marketing, finance, legal, and beyond. This makes communication more critical than ever: aligning and coordinating across stakeholders becomes a key lever for moving projects forward. Communication is also the foundation for user conversations that refine empathy.

Because AI evolves rapidly, non-technical colleagues are struggling to grasp its implications. Your technical judgment lets you shape that understanding — explaining what is feasible or not — and thereby influence organizational direction. Ng defines leadership not as a title but as using your technical judgment to influence a group's direction . Continuous self-investment — tracking the frontier, adopting new tools, refining workflows — compounds this capability over time.

Skill 4: High-Agency Ownership — Proactively Find Problems and Responsibly Deliver

The capstone skill is high-agency ownership : AI gives you an unprecedented opportunity to take greater ownership and agency. You actively discover interesting problems and opportunities, then execute responsibly to capture them. This requires judgment about pacing:

Know when to ship a quick MVP for user validation, and when to slow down and build more carefully.

This "fast vs. slow" judgment is something an agent cannot provide. Agents execute defined tasks faster, but they don't decide which path is worth pursuing. High agency is not blind speed; it is proactive discovery, decisive action, and accountability for outcomes.

An agent can tell you "how to do it faster," but "whether to do it, whether it's worth it, and how polished it needs to be right now" still require human judgment.

This directly echoes Shaping the Build : as implementation grows cheaper, the scarce resource becomes the judgment of what to build and to what extent .

Series Recap: From "Build It" to "Decide What to Build" to "Own the Outcome"

The five-part AI Engineering Skills Map traces a consistent arc:

Overview: What capabilities does an AI engineer need?

Building & Deploying AI Applications: Turning AI capability into usable products.

Software Engineering Fundamentals: Why engineering rigor matters more as AI writes more code.

Using Coding Agents: Moving from writing code yourself to directing agents.

Shaping the Build: Deciding what's worth building when implementation is easy.

All five articles orbit one question: As AI becomes more capable, where should engineers spend their time and judgment? The answer is not merely to produce more code with AI, but to apply engineering rigor to understand problems, make sound decisions, drive the build, and validate the resulting value.

From "getting things built" to "deciding what to build" to "being accountable for what gets built" — this may be the most important capability for engineers to continuously cultivate in the AI era.