continuous context + sensor ai

· Zi Wang · 1 min read

Z / Runner 🏃🏻‍♂️

🏃🏻‍♂️ good list below (every sensory category is over-saturated), and all are optimizing the wrong variables, need more awareness, context, and memory. Accelerometer, gyroscope, gps, hrm, mics, speakers, all cheap commodity, including camera wearables, DJI, gopro, insta, meta ray-bans. None of them feel like the breakthrough ai-native hardware. 10/10 conviction, they focus on capturing beautiful pixels rather than understanding usable context, leave that for smartphones. Context perfect camera > pixel perfect camera.

  • Continuous streams of what is happening: motion, sound, pattern, behavioral patterns.
  • Don't record, remember.
  • In the context of healthspan: infer on stress, emotion, mood, training load, cognitive, peer-to-peer dynamics.
  • Next-gen sensors: Sub-millimeter Cameras, PPG (photoplethysmography), EDA (Electrodermal Activity), SCG (Seismocardiography), BIA. (Bioimpedance), Sweat Cortisol, Optical Glucose, more chemical sensors for lactate, ketones, hydration, fatigue markers.

🏃🏻‍♂️ my tl;dr is AI needs to see, we need continuous inference of the internal biological state with external environment awareness.

  • 🏃🏻‍♂️ 1). programmatic & persistent data storage access w/o using the clunky Files API. 2). Unix pipes to connect multiple prompts.

Stephen / Basketball 🏀

  • 🏀 "lab on a chip"; see "much later" below.
  • 🏀 "Mamba is a new state space model architecture showing promising performance on information-dense data such as language modeling, where previous subquadratic models fall short of Transformers. It is based on the line of progress on structured state space models, with an efficient hardware-aware design and implementation in the spirit of FlashAttention."
  • *🏀 "But situational awareness on AI isn't actually that hard, once you step back and look at the trends. If you keep being surprised by AI capabilities, just start counting the OOMs [order of magnitude]."*

🏀 #voice-ai #sensor-ai an always-on microphone + a 6-axis motion sensor on a bracelet – to log (all) mental / emotion / social states + to infer (most) health / fitness / sleep metrics.

  • minimal: 10g 2-nickel-heavy, 2mm 1-nickel-thick, 9mm usbc-wide.
  • no button / switch / bluetooth / wifi / screen / status.
  • 24-hour battery (10 mins to charge 90%), dangling usbc-male connector to upload and charge via iphone / macbook.
  • $10 bill of materials (bom) at 1m units: arm cortex m4, microphone, sensor, storage, battery, usbc, board, strap, waterproof coat.
  • limitless.ai: founded in 2020, raised 33m at 368m in 2023, sold 70k units, 5m annual revenue, 50 employees, acquired by meta in 2025 dec; 100-hour battery, $99 device + $29/month subscription, on-device compression, private cloud data; ex-googler, optimizely founder, working under ex-apple design chief alan dye.
  • whoop 5: 27 grams, 35mm length, 24mm width, 11mm think, 14-day battery.
  • fitbit flex 2: $60, 5-day battery, simply vera (wang), vibrating alarm / notify, 50m swim proof, launched in 2016, discontinued in 2019.
  • $368 apollo neuro, $199 hume band, $199 polar loop, $70 xiaomi smart band, nowatch, withings pulse hr, garmin $79 vivofit / $149 vivosmart, $114 pavlok, axivity ax3 (11g, 2-week battery), geneactiv (16g), apple watch (10mm), *amazon halo, mirai.
  • comparable size: yubikey 5c nano – 2.4mm-thick 9mm-wide tongue, shell-less pcb-edge usbc connector. comparable weight: airpod 4 (4.3g), airpod pro 2 (5.3g).
  • much later: gps, vibration, clock display, magnetic charger, camera; voice/sound activated recording; photoplethysmography (ppg), inertial measurement unit (imu), electrodermal activity (eda), bioimpedance analysis (bia), continuous skin temperature, pulse oximetry (spo2), sweat cortisol, piezoelectric sensor, seismocardiography (scg), non-invasive optical glucose.
  • bee.computer: founded in 2022, raised 8.5m in 2024, sold 5k units, $1m annual revenue, 15 employees, acquired by amazon in 2025 july; 160-hour battery (7 days), $50 device + $19/month subscription, always-on recording; ex-twitter/square founder maria zollo, working under amazon devices chief panos panay.

🏀 #ai-ux gemini 3 ultra api but an usable web app: previews from fast / thinking model in parallel while waiting for deep research, no click to confirm deep research, organized history with prompt previews, queue for 3+ research, reference link previews, more embedded images, chatgpt-like project document contexts, 1-sentence mentions of key names, keyboard-only access and shortcuts, single-page (not embedded panel) scrolling, fix page syncing with safari cloud tabs, fix auto/forced scroll-to-top at page reload..

  • 🏀 z, your 3 feature requests / bug fixes for ai backend / frontend? yes, trivial for us to build but i am spending / hating 90% time there.
  • 🏀 langchain = sqlite + awk. show me your .bashrc!