Build Your Personal AI Engineering System

Most developers now have access to powerful coding agents. The bigger challenge is building a system around them that actually fits the way you work. Skills get scattered across tools. Context grows until it becomes expensive and noisy. Useful preferences disappear between sessions. Work still needs to be triggered manually. And when you switch models or hit usage limits, workflows tied to a single coding harness can quickly break down. In this practical session, you'll see a real AI engineering setup built to solve these problems. You'll learn how to create a synced, Markdown-based skills and knowledge vault that travels across tools, write skills that meaningfully change agent behavior, capture recurring feedback so your system improves over time, and orchestrate Claude Code and Codex for different types of engineering work. You'll also see how scheduled tasks and an always-on setup can move parts of your workflow beyond manually starting every task. The goal is not to copy one person's workflow exactly. It is to give you a working architecture, practical patterns, and an open-source starting point you can adapt to build your own AI engineering system. What You'll Learn By the end of this session, you'll be able to: • Build a portable skills vault using plain Markdown so your instructions and workflows aren't locked into a single AI coding tool. • Write skills that change agent behavior, rather than simply storing prompts or documentation. • Create a feedback loop for your coding agents that identifies recurring preferences and turns them into reusable skills. • Keep context under control by deciding what should persist, what should be retrieved, and what should stay out of the active context. • Orchestrate multiple coding agents by using Claude Code for longer-running planning and Codex for focused, bounded tasks. • Choose reasoning effort based on the task, using higher thinking levels for architecture and lower levels when the execution path is already clear. • Design work around token and usage limits so hitting a limit in one tool doesn't stop your engineering workflow. • Set up scheduled and always-on tasks for work that shouldn't depend on you manually triggering it. • Adapt the complete setup to your own workflow using the open-source vault template shared during the session. Who Is This For? This session is designed for: • Software engineers already using AI coding assistants • AI engineers and developers building with coding agents • Developers using Claude Code, Codex, or similar coding tools • Engineers experimenting with skills, persistent context, or reusable agent instructions • Developers who regularly switch between models or AI coding tools • Technical leads looking for repeatable AI-assisted development workflows • Anyone who has moved beyond basic prompting and wants a more systematic way to work with coding agents