life levels + 10-min snippets

· Zi Wang · 1 min read

Z / Runner 🏃🏻‍♂️

🏃🏻‍♂️ "FSD for Life"; assumptions: (12-18 months) 1). 80% of tasks (work & life) will be automated by a swarm of agents & reasoned by the "hive queen" (i.e. next gen frontier models, AGI not a necessary condition). 2). Mobility framework analogy is >50% accurate (explanation below:)?

  • Level 1 - 5: cruise control → lane assist → hwy routing → driver-assisted (our tesla) → FSD (waymo). Deployment on well-known conditions (roads, weather, geo-fenced).

  • Elon built up supporting ecosystem (products, services, industries): the secondary business was more valuable than the primary (supercharger network, battery packs, manufacturing)

  • Level 0 = manual life

  • Level 1 = "cruise control life" browsers, software apps (passive)

  • Level 2 = "lane keeping", proactive (2x speed, notifications, summary)

  • 🏃🏻‍♂️ i only find it useful for synthesis & transformation, hence, notebookLM.

  • Level 3 = "copiolits", manus, open-claw (schedulers)

  • Level 4 = high-autonomy; proactive smart assistants

  • Level 5 = "full-autonomy"; you define intent, system handles all execution across contexts (cars move through geography, ghost moves through years)

  • 🏃🏻‍♂️ earn the right to capture all day; limitless being a good example. Too much friction, 60s as the start…

  • 🏃🏻‍♂️ yes, value in compression & clever on feedback loop (we should aim for delegation of monitoring as the v1 goal). Treat our step 1 as prototype or the sensing-layer, and using manus/claw as the continuous belief update engine.

  • 🏃🏻‍♂️ yeah, keep it simple. i don't see utility in search; i just it spoon feed.

  • 🏃🏻‍♂️ define?

  • 🏃🏻‍♂️ telegram is good but it maybe a bloat if we are talking about product.

🏃🏻‍♂️ above are useful in the context of compression, faster intake, better filtering (level 2) = personal intelligent curation + feedback loop. User benefits = optimizing content throughput. We should build it.

  • AND level 3 means we need to reduce decision load & drift over time. What's missing is AI helping me make better decision, such as shop better, reallocate capital, and relationships. We need to launch w/ context-aware alignment (your podcast-agent is needed), and we have to take advantage of the "panel 1 intent capture)
  • Level 4, means, we can start action-link. No longer just about content consumption. We auto-CRU(D) (create, read, update, deploy) at scale.
  • 🏃🏻‍♂️ i like telegram, hence, first and only wallet w/ native telegram client. But i worry it may bloat the product.

🏃🏻‍♂️ 10 min snippet /1000 channels → intent-coupled media ingestion; score on relevance to stated goals, so our agent can track and alert on drift

  • For me: Build a $100M ARR company, ship ghost, run 70 miles/wk, raise Ira w/ hybrid of east/west pedagogy, longevity ← (agent) → new macro data, ai tech & biz, parenting events, venture patterns

  • 🏃🏻‍♂️ events, activities, summer camp, spring break, music school, books to read …

  • If there is ER-100, then (ghost) → update longevity, proposes lab test, draft note to my PCP, check availability

Stephen / Basketball 🏀

  • 🏀 hate all summarizations, worse than slop. highlight clips instead. per my first #tgi talk and yc pitch: extractive (not abstractive) summary.

  • 🏀 longer voice journals or all-day screen captures seem to be the way to capture now? you liked macos's biome (system logs of apple ai)?

🏀 #podcast-agent: 10-minute snippets (highlight clips) from 1000 channels, 2-hour daily content while driving + basketball

  • 🏀 no need for manus or claws for ux for coming months in #podcast-agents. simply maximize content enjoyment, learning, search.. with 100k users before next value propositions of #audio-agent. that is, #her for life.

  • prototype stages: only forward button on airpod to skip; daily telegram voice messages for content feedback

  • not news or synthesis (yet): notebook, huxe, tl;dr, pocketpod.

  • feedback is key: vastly, firehose.

  • 🏀 via this telegram voice messages. minimal friction; maximal gradients.

  • 🏀 tdlib.

  • not contextual feedback (yet): snipd.

  • 🏀 yes, like manus have picked telegram, that may be the perfect client: emoji reactions, intent and actionable html5 mini app.

  • 🏀 what can agents help ira? Now that school selection is so over, what are the top tasks and goals in coming weeks that agents can help?

  • 🏀 #lean.

🏀 "10 business most valuable to turn around using ai agents"