WAIC 2026 wrapped up in Shanghai last week. Every booth had hardware — AI glasses, brain-wave sleep devices, conversational labs. Specs everywhere: weight, brightness, battery.
But nobody answered the question I actually came with: what can these glasses do for me, today?
I went looking and found something interesting. The application layer that "agent glasses" promise — it's already buildable. Not next year, not after a firmware update. Right now, with a CLI tool and a phone camera standing in for the hardware.
The setup
# Install the Bailian CLI (requires Node.js >= 18)
npm install -g bailian-cli
# Authenticate
bl auth login
Grab a free API key from the Model Studio console — the free tier covers everything below. Setup takes about five minutes.
Two commands do all the heavy lifting:
| Command | What it does | When to use |
|---|---|---|
bl vision |
Upload an image + a question, get a visual analysis | Food, contracts, plants, gym equipment |
bl chat |
Send text (optionally with a file or persona), get a response | Planning, analysis, recommendations, chat |
Health: Eat, train, track
1. Food calorie recognition
Point your camera at a plate:
bl vision --image lunch.jpg --prompt "Identify the food, estimate calories and key macronutrients (protein, carbs, fat)"
I shot a plate of shredded pork with fish sauce over rice. Got ~650 kcal, protein/carb/fat ratios. Not lab-precise — but good enough to log a meal without manually searching a food database.
2. AI fitness coach
bl vision --image treadmill.jpg --prompt "Identify this equipment and generate a fat-loss plan: 3 sessions/week, 500 kcal per session"
Recognized the treadmill, produced a structured weekly plan — exercise type, duration, intensity, target heart-rate zone. Beats wandering between machines at the gym.
3. Body metrics analysis
bl chat --file health-7days.json --prompt "Analyze 7-day health data trends (heart rate, steps, sleep, blood oxygen, stress). Flag anomalies and suggest improvements."
Uploaded a week of smartwatch data. It caught a deep-sleep dip on days 3 and 5, correlated it with elevated stress scores, suggested an earlier wind-down. Whether that's medically actionable is a different question — but as a pattern-spotting tool, genuinely useful.
4. Smart diet manager
bl vision --image dinner-plate.jpg --prompt "Identify each dish, calculate total calories and macro ratios"
bl chat --message "Based on a fat-loss goal, suggest dietary adjustments for this meal"
Vision identifies the food, chat adjusts the plan. Turns "I should eat better" into a concrete, per-meal instruction.
Family: The third eye at home
5. Parenting assistant
bl vision --image baby-blocks.jpg --prompt "Describe the child's activity and assess developmental stage"
bl chat --message "Log today's feeding, nap, and play observations. Generate a weekly growth summary."
Photographed my one-year-old stacking blocks. The model identified it as fine motor skill development and logged the milestone. Not a pediatrician — but as a memory aid for sleep-deprived parents, it beats scribbling in a notebook.
6. Gift selection assistant
bl vision --image kids-room.jpg --prompt "Analyze this scene for gift-giving clues: interests, ages, existing items"
bl chat --message "Based on the scene analysis, recommend 3 gift options with reasoning and price range"
Shot my kid's room. It spotted a half-finished Lego castle and a nearly empty skincare bottle. Inferred: a Lego expansion set for the kid, skincare restock for my wife, a family board game as a wildcard. More thoughtful than my default ("just get a gift card").
Business: Scan, assess, decide
7. Contract risk scanner
bl vision --image lease-contract.jpg --prompt "OCR this contract page and extract the full text"
bl chat --message "Review this contract clause by clause. Flag any risky terms (excessive penalties, unusual terms, hidden fees). Assign risk levels and suggest mitigations."
Scanned a residential lease. It flagged a termination penalty clause as "high risk" (two months' rent — above local norms). My lawyer later confirmed it was worth negotiating.
8. Hotel search & booking
bl chat --message "Trip: May 20-22, 2 nights, near Shanghai Bund, budget hotel, total budget 1000 RMB. Find and recommend suitable hotels."
Three options with price, location, rating, amenities — plus a comparison table and a booking recommendation. No camera needed. What I'd normally spend 20 minutes on across three booking sites.
Companion: From tool to partner
9. Visual Q&A
bl vision --image plant.jpg --prompt "What plant is this? What's its name (scientific and common)? How do I care for it?"
Held up a succulent I'd been slowly killing for months. It identified it as an Echeveria, told me I was overwatering, and gave a care card: bright indirect light, water when soil is fully dry, every 2-3 weeks. The plant is now thriving.
10. Companion robot
bl chat --persona "You are XiaoZhi, a warm and cheerful companion robot. You pick up on emotional cues and proactively suggest activities or games." --message "I'm so happy today! Let's hang out!"
A chatbot with a persona. Not groundbreaking on its own — but in the context of agent glasses with a visual feed and calendar, the potential is a persistent, context-aware companion that knows what you're looking at and what you're doing.
What actually changed
| Task | Before | After |
|---|---|---|
| Log a meal's calories | Search food database, estimate portions | Snap a photo, get instant estimate |
| Start a gym routine | Google, cobble together a plan | Photograph equipment, state goal, get a structured plan |
| Review a contract | Read every line, miss the buried clause | Scan the page, get flagged risks with levels |
| Pick a gift | Wander a store, guess | Photograph the room, get targeted recommendations |
| Identify a plant | Reverse image search, then find a care guide | One command: name + care instructions |
| Book a hotel within budget | Open 3 tabs, filter, compare | State constraints, get a comparison table |
The pattern: input gets simpler (a photo or a sentence), output gets more structured (tables, plans, ranked recommendations). The AI doesn't make the decision for you — it does the tedious processing so you can decide faster.
Who is this for?
- Early adopters who bought AI glasses and are underwhelmed. The hardware is here; the application layer is what's missing.
- Developers building for the agent glasses ecosystem. Qwen opened the door to third-party Skills. The CLI is the fastest way to prototype what a Skill could do.
- Anyone who doesn't have glasses yet but wants a preview. Every scenario runs with a phone camera and a terminal.
-
Anyone whose daily routine includes repetitive information tasks. If you do it manually every week, there's probably a
blcommand for it.
WAIC 2026 made one thing clear: the AI hardware race is on. But hardware without applications is just expensive plastic. The real question isn't "which glasses are lightest" — it's "what can the glasses do for me that my phone can't?"
These 10 use cases don't answer that fully. But they show the application layer is already buildable today. The glasses just make the input frictionless — the intelligence is already there, waiting to be invoked.
What use case would you build first? Have you tried any CLI-based AI tools in your daily routine? Drop a comment — I'm curious what people come up with.
Top comments (0)