Phone AI Agents Tested: Can They Really Order Food and Book Flights for You?

In 2026, major tech companies are racing to build "AI agents that run on your phone." Bytedted's Coze, Zhipu's AutoGLM, Baidu's "Shengsuan"... Their shared goal: let AI operate apps on your phone to complete "handson" tasks like ordering food, booking flights, sending WeChat messages. Can they really do it? I tested 3 products—the results were more divided than expected.

Body Image

Why "Phone Agents" Are the Next Big Battlefield

AI assistants on PC are already crowded. But phones are just getting started.

Reason: Most of people's "actions" happen on phones.

If AI can only answer questions, not help you complete tasks, its value covers only 20% of real use cases. Phone agents solve "AI doing things for you on mobile."

Testing 3 Products: Who's Closest to a "Real Agent"?

1. Zhipu AutoGLM ⭐⭐⭐⭐

Task: Order me affordable takeout near the office

Process: 1. Opens Meituan app ✅ 2. Recognizes office address ✅ 3. Searches "under 10 yuan meals" ✅ 4. Lists 3 options for me to choose ✅ 5. I pick option 2 → Fills address automatically → Submits order ✅ 6. Stops at payment for human confirmation (safety design) ✅

Result: End-to-end completed without issues.

Verdict: AutoGLM is currently the closest to a "real agent" on mobile. Its core capability is "screen understanding"—it reads app interfaces, understands buttons, prices, and input fields.

2. Bytedance Coze ⭐⭐⭐

Task: Send Dragon Boat Festival greetings to 10 WeChat groups

Process: Works within the Bot framework, but limited for personal account automation.

Verdict: Coze excels at "Bot building" for enterprises—not at automating your personal WeChat.

3. Baidu "Shengsuan" ⭐⭐⭐

Task: Check Beijing weather, set a 7 AM alarm

Result: Completed basic tasks. Stops at anything involving login, payment, or sensitive actions.

The "Impossible Triangle" of Phone Agents

All 3 products share one fundamental constraint:

Can deeply automate
     /\
    /  \
   /    \
"Can't bind cards/pay"——"Can pay but security risk"——"Safe but too restricted"

Constraint	Explanation
App restrictions	WeChat, Alipay block third-party automation
Payment security	All payment tasks need human confirmation
Permission trust	Agents need broad phone permissions users don't trust

These constraints mean phone agents in 2026 can only handle "shallow tasks."

Kaihe's Unique Position

Phone agents face App ecosystem limitations and security boundaries. Local agents have unique advantages:

All data stays local—no App permissions needed
Runs 24/7—no phone screen required
Connects local hardware—printers, scanners, PLC devices

Phone Agent vs Kaihe Agent

Scenario	Phone Agent	Kaihe Agent
Order food	✅	❌
Manage local files	❌	✅
Connect printers/scanners	❌	✅
24/7 business monitoring	❌	✅
WeChat auto-replies	✅	✅

Bottom line: Phone agents in 2026 already handle the "search → confirm → execute" basic loop, but deep automation is still constrained by App ecosystems and security. For 24/7 local business monitoring, Kaihe's Agent solution remains the better choice.

AI Agent column tracks the latest Agent products.

Phone AI Agents Tested: Can They Really Order Food and Book Flights?