# Shipping the first version of Wakefile I shipped wakefile over the weekend (!!) a shared operational layer for people and AI agents working together on long-running projects Built it because I was feeling the effects of context rot and noticed that tools were failing because they were trying to do too much all at once, rather than neatly approach one well-bounded problem at a time Wakefile is a markdown-based framework for agentic builds over long periods, multiple agents, and many context windows. It works with any AI tool that can read and write markdown files Clone the repo, run `⁠::init` and it scaffolds a shared operational context for whatever you're working on Ran a drag assessment on it to compare: - Wake's likely token costs compared to no persistent context - Wake's likely token costs compared to minimal spec-driven design files **Honest result:** it is not lightweight (yet). First open release was built with wakefile itself across eight cycles, and the overhead is real. I would not use it for one-off runs. I would use it for projects where continuity and decision tracking actually matter. As I was finishing this and sharing it with friends, I started seeing more people building toward (or who've already built) something similar [The GSD framework](https://github.com/gsd-build/get-shit-done) in particular feels like it has a headstart on solving a very similar problem, but there are a few areas where wake feels different, namely: ### Running narrative + lessons learned wake surfaces eductions (lessons learned via the build), build logs, and claims as it goes so the system as a whole can apply lessons across projects and tell stories about itself. This feels like it points in a similar direction to what [Andrew Ng is building with Context Hub](https://github.com/andrewyng/context-hub). ### Process agnostic wake's init sequence uses the Product Development Lifecycle (PDLC) and Software Development Lifecycle (SDLC) as helpful defaults, but the user is free to define their own process lifecycle that wake follows. For non-software builds, this feels important Glad I got to work through the problem on my own first, then through conversations with Jacob Mikesell. Now I want to see where these tools converge. If the answers are similar, I would rather contribute to a broader ecosystem for the network effects If there's an addressable gap, that'll probably shape the next version of wake Docs: ⁠[https://wakefile.com](https://wakefile.com) Repo: [https://github.com/jcalyon/wakefile](https://github.com/jcalyon/wakefile)