Talks and workshops about frontend architecture and delivery at scale.

Every time you start a new AI-assisted session, your agents start blind. No memory of last week's architectural decision. No knowledge of why you chose that pattern three projects ago. No awareness that the session ended before you finished. So you explain everything from scratch.

Context engineering is the discipline of fixing that: deciding what your agents know, when they know it, and how that knowledge survives session boundaries and tool switches. It's distinct from prompt engineering, and arguably the higher-leverage skill when building with AI agents.

This talk covers the techniques behind persistent AI memory: index-first loading, anchored iterative summarization, phase-based context loading. The architectural decisions that shaped them. And why the right answer turned out to be simpler than you'd expect.

Dependency Injection (DI) is a key pattern for building maintainable, scalable, and testable software, but it is often associated with heavyweight frameworks and complex runtime infrastructure. From a Lean Web perspective, this raises a simple question: can we keep the benefits of DI while staying close to the platform?

This talk explores how metaprogramming, specifically ECMAScript Decorators, enables a DI approach in pure Vanilla JavaScript. We will review the current TC39 Decorators proposal (Stage 3) and focus on Parameter Decorators (Stage 1).

We will demonstrate how to build a minimal Inversion of Control (IoC) container that aligns with Lean Web principles, defining injectable services with a custom decorator and annotating constructor parameters to collect the metadata required for dependency resolution.

The goal is to show how upcoming ECMAScript features make it possible to implement clean, declarative DI using the platform itself without relying on complex frameworks.