Descriptions:
In this full two-hour workshop recorded at the AI Engineer conference, TypeScript educator Matt Pocock (@mattpocockuk) lays out a systematic framework for professional software engineers working with AI coding tools. The central thesis is that software engineering fundamentals—task sizing, structured planning, iterative delivery—transfer directly to AI-assisted development, and that engineers who ignore these principles end up frustrated with inconsistent results.
Pocock introduces the concept of the LLM “smart zone” and “dumb zone,” explaining that context windows degrade in quality around the 100K token mark regardless of advertised window size—a constraint that fundamentally shapes how work should be structured. He advocates for decomposing large tasks into small, clearly scoped units using kanban-style boards with explicit blocking relationships, treating AI agents like junior developers who need well-defined tickets rather than open-ended briefs. He distinguishes “AFK” tasks that can be fully delegated from tasks requiring human-in-the-loop oversight, and covers how to write Product Requirements Documents before prompting to anchor agent behavior.
Drawing on Martin Fowler’s Refactoring and the Pragmatic Programmer, Pocock demonstrates that principled engineering discipline—not just better prompting—is what separates productive AI-assisted development from the dumb-zone spiral. The workshop includes live exercises, a public Q&A board, and a walkthrough of multi-phase planning using Claude, making it a practical reference for engineers at any experience level with AI tooling.
📺 Source: AI Engineer · Published April 24, 2026
🏷️ Format: Hands On Build
