A technique of [[Context management]]: [^1]
> 1. When we **fold context**, we **diverge**, exploring separate paths in separate [[Context window|context windows]].
> 2. After exploring a solution space, we **converge**, comparing what we have learned, synthetizing everything together.
It is actually [[Double diamond]] model, which usually is divided into 2 steps:
- research
- development
Also folding is similar to compression, and [[Compression is cognition]]. It is especially useful in terms of [[Multi-agent second brain]] where by deciding what to forget (and save) and when to recall, an [[AI agent]] demonstrates what it truly understands. [^2]
## Examples
### Learning
This approach is generally great for learning[^1], you can create a process like:
1. Go to [[Organism earth]] and get pages for three different philosophers (e.g. [[Marcus Aurelius]], [[Alan Watts]] and [[Richard Feynman]])
2. Open three terminal windows, have separate [[Claude Code]] read each page
3. Ask some questions about each philosopher
4. Save a summary of each conversation to an `.md` file.
5. `/clear`
6. Have [[Claude Code]] read all three summaries, synthesize, compare and contrast.
7. Save to `synthesis.md`.
8. `/clear`
9. Include `synthesis.md` and have ay other conversation
[^1]: https://newsletter.squishy.computer/p/folding-context
[^2]: https://timkellogg.me/blog/2025/06/15/compression