Michael Ramos

I don't think I like the idea or architecture of wiki as memory; in which some external agent curates alongside development. I think your development process is flawed if you use surrounding context as a crutch and rely on an auxiliary process to combat things like context rot. In other words - the right process accounts for changes in decision making over time and using that as an advantage, not bloat that needs to be curated continuously. The better process also produces or removes context in a simple way alongside development. Context/development need to be tightly integrated. The default process treats context as ephemeral - master this before you handicap your project with bad process.

first time using GPT 5.6 Sol, I asked it to come up with good section headings for our next podcast episode it spun out 4 subagents to read one transcript file then conceded the subagents were just doing fake work because the parent did everything anyways the tokens must flow

A second strategy: use Fable 5 as an orchestrator. Fable 5 plans and delegates to workers (Sonnet 5). Most tokens are billed at the lower worker rate.

first time using GPT 5.6 Sol, I asked it to come up with good section headings for our next podcast episode it spun out 4 subagents to read one transcript file then conceded the subagents were just doing fake work because the parent did everything anyways the tokens must flow