Descriptions:
Recorded at AI Engineer Europe in London, this Latent Space podcast interview with Matt Pocock—creator of Total TypeScript and founder of AI Hero—covers what Pocock sees as the most important and underappreciated insight in AI-assisted development: engineering fundamentals matter more, not less, when AI is writing your code.
Pocock’s central argument is that treating generated code as an opaque compile target leads directly to unmaintainable codebases. When he went back to foundational software texts—Extreme Programming, The Pragmatic Programmer, Domain-Driven Design, A Philosophy of Software Design—he found that every principle designed to make code easy for humans to change also makes it dramatically easier for AI to change. A messy codebase that’s hard for a human to reason about is even harder for a model to modify correctly. This reframes AI coding not as a replacement for engineering discipline but as an amplifier of it.
He also describes a concrete workflow artifact: a domain glossary skill that scans a codebase and extracts project-specific terminology—the kind of context-laden jargon that defines every mature project. Keeping that generated markdown document accessible during prompting sessions substantially improves AI output quality because the model operates within the project’s actual domain model rather than generic language. Pocock ties this to the broader concept of a “narrow waist” in architecture: a stable, well-defined interface layer that insulates the rest of the system from churn, whether from human refactors or AI edits.
📺 Source: Latent Space · Published May 07, 2026
🏷️ Format: Interview
