I’ve been building with coding agents for the last 18 months. My current setup today includes ~4 agents operating in their own git worktrees, following a strict process for ideation, planning, build and documentation. Plus codebase reviews every other day to keep things tight.
I’ve been working on a new codebase for the last 6 weeks pretty much full time and wanted to share how my agent context has changed and matured along-side the code.
One thing lacking in the discussions about context is recognition of the need for context to change and mature not only in content, but also in format and function.
An important note, each of my ~4 agents acts as an orchestrator. It holds the context and then seeds its sub-agents with the context they need to execute. This means they can hold a lot of context themselves as they don’t also have to maintain operating context. This allows me to have much LARGER codebase context ~30 – 50% of total available context.
A key process for me now is not just updating context with every change, but also re-engineering context. Given the pace at which my codebase grows and the supporting context grows, context re-engineering is happening every week.
To support this process, I had Chat GPT5.2 evaluate hoe my context is changing over time and identify key maturity phases and themes, and then look forward into the future in terms of how these themes and patters are likely to mature in the future.
The intent is to make my context re-engineering activity more intentional.
I’m sharing GPT5.2’s analysis of my context re-engineering patterns below for anyone who might find get value from meta analysis.
Leave a Reply