What does wearable AI do?

The framework fixes a conflation

The canonical question "what does wearable AI do" conflates two things: what wearable AI is marketed to do vs what wearable AI verifiably ships. Per DEPLOY's verified-vs-claimed framework, the framework's editorial work is naming this distinction explicitly; marketed capabilities that are not yet verified-shipped are different from capabilities verified at consumer-deployment scale. Trade-press coverage that surfaces marketed capabilities as shipped product is editorially imprecise; the framework's discipline applies the same verified-vs-claimed reading to wearable AI capability claims as to humanoid robot capability claims.

---

The six capability domains

Per cohort enumeration + verification posture analysis, wearable AI capability sorts across six domains. For each domain: what is marketed vs what is verifiably shipped + the cohort entries operating at each verification tier.

1. Visual Q&A on what the camera sees

Marketed: ask the device what it sees + receive AI-generated answers about objects, scenes, text, locations.

Verified shipped at commercial deployment: Meta Ray-Ban Look-and-Ask multimodal queries via Meta AI; Rokid GPT + Gemini routing for visual queries via camera; Brilliant Labs Frame with developer-accessible visual queries (SDK depth varies); Solos AirGo Vision GPT-4o multimodal processing camera input for "look-and-ask" queries in audio-first form factor.

The verified shipped capability is genuine cloud AI substance at consumer-deployment scale across multiple cohort entries. Specific feature parity + accuracy varies by entity + use case + the consumer evaluation produces verified-vs-marketing parity at structurally distinct depths.

2. Live translation across languages

Marketed: real-time spoken-language translation between participants.

Verified shipped at commercial deployment: Meta Ray-Ban (live translation feature shipped); Rokid with offline translation (cross-language conversation support without cloud roundtrip); Solos with GPT-4o translation. The offline-translation depth distinguishes Rokid in cohort; offline-capable translation is structurally distinct from cloud-only translation because connectivity dependency differs.

3. Capture & summarize voice + transcripts

Marketed: passive voice capture + AI-generated transcripts + meeting summaries + key-point extraction.

Verified shipped at commercial deployment (cap-flagged at DEPLOY registry anchor depth pending): Plaud (NotePin variants for capture + summarize use case at genuine cloud AI substance); Friend (pendant capture for ambient companion use case); Bee (adjacent ambient-companion entrant); Omi (emerging ambient-companion entrant; not yet anchored at registry depth).

Per cap-flag-as-trust-signal, Plaud + Friend + Bee + Omi are named in body content but not yet anchored at full DEPLOY-registry-entity depth; specific commercial-claim verification (pricing variants, unit counts, current launch state) should be verified against current maker public surfaces rather than aggregator-source propagation.

4. Ambient recall of conversations + context

Marketed: device passively listens to ambient context + remembers conversations + surfaces relevant context on demand or during interactions.

Verified shipped at commercial deployment (cap-flagged at DEPLOY registry anchor depth pending): Bee (ambient recall use case); Friend (ambient companion); Omi (emerging ambient companion). Ambient recall distinguishes from capture & summarize by use case: capture & summarize produces transcripts + summaries on demand; ambient recall continuously listens + maintains context across interactions.

Per DEPLOY's framework, ambient recall use case is structurally adjacent to capture & summarize at form-factor level (both pendant) but distinct at end-user-assumption level (continuous listening + memory vs on-demand capture + summarize).

5. Smartphone replacement (promised; verified-as-failed for cohort exemplars)

Marketed: device replaces smartphone for queries, communication, productivity, entertainment.

Verified shipped at commercial deployment: Humane AI Pin marketed full smartphone replacement; verified outcome: commercial discontinued February 2025 after HP acquisition + server shutdown; closed-ecosystem cloud dependency bricked hardware. Rabbit r1 marketed LAM (Large Action Model) as smartphone-replacement bridge; verified outcome: marketing-vs-shipped gap; pilot active with documented shipped capability significantly narrower than CES 2024 LAM framing.

Per DEPLOY's framework, smartphone-replacement claim is the cohort's most ambitious capability domain; both major commercial attempts demonstrated cap-flag warranted at the verification-vs-marketing layer. Anyone marketing a wearable as full smartphone replacement should be evaluated against the Humane + Rabbit verified-outcome record.

6. Health monitoring (uncovered in current DEPLOY cohort)

Marketed: biometric tracking + sleep + heart rate + activity + emerging AI-driven health insights.

Verified shipped at commercial deployment (uncovered in current DEPLOY registry): Oura Ring 4 (smart ring + comprehensive sleep + biometric); Apple Watch (smart watch + biometric + activity + emerging health features); Fitbit (smart watch + activity + biometric); Samsung Galaxy Watch (smart watch + biometric).

Per cap-flag-as-trust-signal, health monitoring is the cohort's most editorially significant gap in current DEPLOY registry. The smart ring + smart watch + AI earbuds sub-categories operate verified-commercial state with documented health-monitoring + emerging AI integration; DEPLOY's registry has not yet anchored these at framework depth. Per DEPLOY's framework, the gap is surfaced as editorial signal rather than fabrication; trade-press coverage that includes health-monitoring wearable AI typically draws from these unanchored entities.

---

The framework distinction: marketed vs verified-shipped

Per DEPLOY's verified-vs-claimed framework, the canonical wearable AI capability question conflates marketed and verified-shipped. The framework discipline distinguishes:

• Marketed capabilities: what makers claim the device does in product communications + advertising + demos.
• Verified-shipped capabilities: what the device actually does at consumer-deployment scale with verifiable depth.

The verified-vs-marketing gap is largest at the smartphone-replacement claim domain (Humane discontinued + Rabbit marketing-vs-shipped gap demonstrate the failure mode) and structurally distinct across visual Q&A + live translation + capture & summarize + ambient recall + health monitoring domains. Per DEPLOY's framework, naming the conflation explicitly + applying verified-vs-claimed discipline produces structurally different reading than trade-press capability surveys that propagate marketed-as-shipped framing.

---

Bottom line

Wearable AI verifiably does five things at commercial deployment scale + one things uncovered in current DEPLOY registry:

• Visual Q&A verified at genuine cloud AI substance: Meta + Rokid + Brilliant + Solos.
• Live translation verified including offline depth: Meta + Rokid + Solos.
• Capture & summarize verified at genuine cloud AI substance (cap-flagged registry anchor depth): Plaud + Friend + Bee + Omi.
• Ambient recall verified at genuine cloud AI substance (cap-flagged registry anchor depth): Bee + Friend + Omi.
• Smartphone replacement marketed + verified-as-failed: Humane discontinued + Rabbit marketing-vs-shipped gap.
• Health monitoring uncovered in current DEPLOY registry: Oura + Apple Watch + Fitbit + Samsung Galaxy Watch are canonical contenders DEPLOY hasn't yet anchored.

Per DEPLOY's framework, the editorial honesty across all six domains is the verified-vs-marketing discipline; specific capability claims should be evaluated against shipped-product verification depth rather than marketing-layer framing.

---

Where to go for context

For broader AI wearables category context + AI-substance spectrum framework, see what is an AI wearable. For wearable AI typology + form factor + substance axes, see what are the different types of wearable AI. For wearable AI selection framework, see best AI wearables 2026. For broader physical AI category context, see what is physical AI. For methodology canonical references applicable to wearable AI capability breakdown: verified-vs-claimed at within-entity granularity (use-case capability claims at within-product feature depth) + the 9-tier source-quality rubric.