Back

thesis cycle + 100m fund

1 min readZi Wang

Z / Runner πŸƒπŸ»β€β™‚οΈ

πŸƒπŸ»β€β™‚οΈ thesis recycle Google's trash bin β†’ $10B

πŸƒπŸ»β€β™‚οΈ chatbot & mental health: headspace. πŸƒπŸ»β€β™‚οΈ Limitless voice journaling + ChatGPT entries made headspace less useful. BUT, i need alt from Limitless, Meta acquisition means all my daily conversation will be egressed to FB.

  • πŸƒπŸ»β€β™‚οΈ Venture funds focusing on longevity & healthspan.
  • πŸƒπŸ»β€β™‚οΈ Yes on DTC, we need to solve DISTRIBUTION.
  • $4.15b A16z (Levels, FX health), Khosla (circulate health), General Catalyst (Commure, Olive)
  • πŸƒπŸ»β€β™‚οΈ Category 1: No FDA Approval Needed ("General Wellness") Category 2: Deployed in Mass Market (2025 Standard), Category 3: Projected Next-Gen (2026 & Beyond)
  • πŸƒπŸ»β€β™‚οΈ 10 AI Supernova startups we surveyed reached ~$40M ARR in their first year of commercialization and ~$125M ARR in the second year of revenue generation. … On average, these AI Supernovas have only 25% gross margins, often trading distribution for profit in the short term. Despite these low margins, these AI Supernovas seem to demonstrate an incredible $1.13M ARR/FTE, which is 4-5x above a typical SaaS benchmark."
  • πŸƒπŸ»β€β™‚οΈ Shooting Stars reach the ~$3M ARR range within their first year of revenue while quadrupling in YoY growth with ~60% gross margins, and ~$164K ARR / FTE (full-time employees) in their first years.
  • πŸƒπŸ»β€β™‚οΈ we need ecosystems like FastMCP from Prefect (that make it much easier to build MCP servers) and tools like Arcade and Keycard (that facilitate agentic authorization and permissioning.).
  • πŸƒπŸ»β€β™‚οΈ In 2025, large context windows and retrieval-augmented generation (RAG) have enabled more coherent single-session interactions, but truly persistent, cross-session memory remains an open challenge. While the foundational model companies are working on memory, so too are startups like mem0, Zep, SuperMemory, and LangMem by Langchain.
  • πŸƒπŸ»β€β™‚οΈ combine the following: Short-term memory via expanded context windows (128k to 1M+ tokens, depending on model and architecture) Long-term memory via vector DBs, memory OSes (e.g., MemOS), and MCP-style orchestration Semantic memory via hybrid RAG and emerging episodic modules, designed for context-rich recall.

Stephen / Basketball πŸ€

  • πŸ€ aging dieases.
  • πŸ€ precision health, harmonize biomedical data.
  • πŸ€ how much do you still use it (now that you run 150 miles a week)?

πŸ€ $100m health + ai fund: maximize fitness performance for 10b people. whitepaper "body intelligence for optimal self".

  • continuous contexts: tiny wearables, streaming biomarkers.

  • persistent agents: composable memories, deductive advice.

  • blue socials: connected humans, atomic lifestyle.

  • πŸ€ more importantly, what our fund is not: healthcare (insurance, providers, treatments). hence, the words "fitness" and "performance" – for direct-to-consumers (DTC) and better-than-preventive. "if you have a body, you are an athlete."

  • πŸ€ via high-school athletes + startup founders.

  • Healthspan Capital: Focuses on early-stage longevity biotech (LongBio) and regenerative medicine, targeting fundamental aging drivers, LongVC: biotech, early stage, Apollo Health Ventures, life science, aging, Maximon, studio, company building, Continuumhealth VC.

πŸ€ "rank 100 biosensors by cost, energy, weigth - for healthspan wearables and fitness ai". for "wearables with no screen or button, but open source and development".

  • πŸ€ how (often) do you use mcp / agents?
  • πŸ€ yes, but show me your (good) session runs – i can hack db together to test different memory strategies / implementations.
  • πŸ€ OpenEvidence automates medical literature review and delivers instant answers at point of care.

Select any passage to copy a prompt Β· esc to return