There Are Two Gaps. Agents Closed One. The Other Is Yours.

Short summary

AI agents shifted the software development bottleneck from execution to evaluation—the critical question changed from "how do I build this?" to "is this correct and aligned with our architecture?" Drawing on Don Norman's 1986 UX theory, the post explains how agents collapsed the Gulf of Execution into a simple prompt, but expanded the Gulf of Evaluation: human reviewers now face output faster than they can meaningfully evaluate. A practical framework emerges: low-risk reversible changes auto-approve; high-risk irreversible changes demand pre-review; ambiguous decisions require multiple domain experts. True judgment cannot be automated—it develops through years of shipping, failure, and real constraints.

• •Agents collapsed the Gulf of Execution (how to do it) but made the Gulf of Evaluation (is it right) the critical bottleneck in software development
• •Framework for human-in-the-loop: risk + reversibility determines review strategy—low-risk auto-approves; high-risk requires pre-approval; ambiguous needs multiple experts
• •Judgment, taste, and discernment cannot be automated—they accumulate slowly through years of shipping, failure, and learning from real-world constraints

Generated with AI, which can make mistakes.