AI Made Coding Cheap—Now Requirements Are the Moat

The hardest problem is no longer writing code—it’s figuring out what to build. The speaker’s hackathon found 17 of 21 AI agent ideas were abandoned because they created no business value. The real bottleneck is accessing stakeholders, reading the room, and eliciting requirements.

Implication: Engineering orgs must move their smartest people upstream, closer to customers and decisions, rather than optimizing code output. Old analysis toolkits—story mapping, business model canvas—become the differentiator. AI will give average answers; human analysis pushes beyond the common pattern.

Tradeoff: This is an organizational shift, not a tool upgrade. Metrics must change from features shipped to features used. It demands that engineers and analysts collaborate on value discovery, which may conflict with existing roles. The risk is over-process without real customer access.

Bottom line: The moat isn’t AI tool access; it’s the ability to ask the right questions, define value, and build the right thing—an old skill with new economics.