Fiction_Co is the north star. All projects should ladder up to: Using AI to revolutionize publishing and storytelling at scale. Knowledge capture is the current critical path.
March 14, 2026 - Anthropic Announcement:
Also: "Claude Mythos" (Capybara tier) leaked - larger model in early access testing, "dramatically" better at coding/reasoning. Coming soon.
COS database crossed 100K messages - demonstrates heavy usage, excellent performance at scale, and system stability under production workload.
UFO research database, arXiv content channels, YouTube transcript pipeline, AI video production research, and Google integration - all delivered actionable insights.
DigitalOcean droplets powered down when not needed - reduced from ~$24/mo to ~$2/mo while maintaining capability.
Action: Schedule structured interviews to extract Michael's publishing expertise. This is the critical path to Fiction_Co success.
Timeline: Start within 2 weeks
Action: Fix double notifications and thinking indicator issues before adding new features.
Impact: System stability at 100K+ messages scale
Effort: ~4-6 hours focused debugging
Risk: All COS data in one SQLite file with no offsite backup
Action: Automate daily backups to cloud storage (S3, Dropbox, etc.)
Effort: ~2 hours to set up
Question: Which Skunkworks experiments support Fiction_Co?
Action: Define commercialization path for successful experiments
Decision Needed: Droplet has been offline since late March
Action: Review simulation outputs and decide next steps