Europe 2026
Full Walkthrough: Workflow for AI Coding — Matt Pocock
Overview
This talk explores a structured workflow for AI coding, emphasizing that traditional software engineering fundamentals remain crucial when working with AI. The core thesis is that by understanding LLM constraints—specifically their "smart zone" and tendency to forget—developers can build effective AI-powered workflows. The approach focuses on breaking down large tasks into manageable chunks that fit within the LLM's optimal performance window, ensuring better decision-making and code quality.
Who should watch
- AI engineers and developers looking to improve their AI coding workflows.
- Product Managers and builders seeking to understand how AI can be integrated into the development process.
- Anyone frustrated with AI making "stupid decisions" or forgetting context.
- Individuals interested in practical techniques for aligning with AI agents.
Key takeaways
- LLMs operate within a "smart zone" and a "dumb zone." Tasks should be sized to remain within the smart zone (roughly the first 100k tokens) to avoid degraded performance and poor decision-making.
- LLMs have a tendency to forget, similar to the character in Memento. Each session resets to the system prompt, making context management critical. Compacting conversation history is an alternative to clearing context but is less preferred than a clean reset.
- The "Grill Me" skill is a powerful tool for achieving shared understanding with an AI agent. It involves relentlessly interviewing the AI about a plan or brief until alignment is reached, creating a valuable "design concept" asset.
- A structured workflow involves transforming an idea into a Product Requirements Document (PRD) as a "destination document," then breaking this down into a Kanban board of "journey" tasks. This planning phase requires human review.
- Implementation can be an "AFK" (Away From Keyboard) task for AI agents, allowing for parallelization. The Kanban board facilitates this by defining dependencies and enabling multiple agents to work concurrently.
- Test-Driven Development (TDD) is essential for AI coding, as it provides crucial feedback loops. AI can struggle with writing good tests, so TDD helps ensure code quality and makes it harder for the AI to "cheat."
- Codebase architecture matters. Deep modules with simple interfaces are preferable to shallow modules, as they are easier to test and manage, leading to better AI performance.
- Manual QA and code review remain critical, even with AI-generated code. Human judgment and taste are necessary to avoid producing "slop" and ensure high-quality software.
Notable quotes
*We forget that actually software engineering fundamentals, the stuff that's really crucial to working with humans, also works super well with AI.*
*When you're first kind of like working with an LLM, and it's like you've just started a new conversation, you start from nothing, that's when the LLM is going to do its best work.*
*The code is your battleground.*
Unofficial community note. Prefer the recording for nuance.