Building AI-Native Startups [002]

Key takeaways

• See a working personal knowledge-graph 'second brain' that lets Claude Code answer 'what's the highest-leverage thing I should do today?' — including exactly which data sources to ingest (bookmarks, LinkedIn/Twitter exports, iMessage, Obsidian Web Clipper) and what it costs to run.
• Learn a log-centric way to build agents (ActiveGraph): treat the event log as the agent to get replay, fork, rollback, and first-class self-improvement — backed by real long-mem-eval benchmark results.
• Watch an agent run a controlled self-improvement loop (Regimes): categorize its failures, propose seam-scoped fixes, validate in a sandbox, and merge only the changes that raise the score.
• Get a concrete AI go-to-market playbook: mine competitor complaints on Reddit/X from the last 24 hours, auto-draft outreach DMs, and stand up daily competitor monitoring through a single agent-tool router (Monid).
• Hear three builders' candid take on where autonomy breaks down today — privacy/permissions for team-wide memory, keeping AI outreach useful instead of spammy, and what long-running agents still need beyond better models.