Notes from the Stack
A production line is only worth its speed because it can halt itself. This series is about building an agentic coding system that works the same way.
Two lanes run through it. The top lane is the harness, carrying work from an idea to something delivered, through an AI-driven pipeline that covers the whole lifecycle. The bottom lane decides what an agent is allowed to believe. I call it the agentic knowledge foundation.
The stack in the title is the technology and the organisation both. They do not come apart, so the series does not treat them as two subjects.
It runs in several parts.
Part I, methodologies and mental frameworks. Fifty years of software engineering did not stop counting when agents arrived. It started counting for more, because an agent runs the fundamentals faster in whichever direction you point it. This part works through the methodologies people are using in agentic engineering now, sets them against each other, and keeps what holds. Ways of thinking about the work, before any of it touches a tool. By the end you should be able to build your own AI-driven development ecosystem.
Part II, tools and technical capabilities. The bridge from methodology to daily practice. Tools for putting domain knowledge and engineering workflows into agents. Building code knowledge foundations. Reading and parsing a codebase an agent wrote. Agentic coding methods for code written alongside a model. Sharpie-style toolchains built for this way of working. Automations for code review and remote verification, and integrations that tie the whole ecosystem together.
Part III, system architecture, design and implementation. The coding agent, two lanes. The top lane runs the delivery loops from an idea to deploy. The bottom lane holds knowledge, retrieval and state. Harness design, context engineering, repository layout, control loops, running it day to day, and real source.
Part IV, human and AI coordination. Still taking shape. Self-evolving coding agents, making the work visible, how a team actually uses this, what an organisation has to change, and what happens when you scale past one person. I will say which of it I have done and which I am guessing at.
Written for people who have already got past the chat window and now have a setup that mostly works, without being able to say why it fails when it fails.