Are AI Phones Just a Gimmick? Analyzing Their Real Value

Background

Since Apple Intelligence two years ago, followed by Doubao AI phones, Qianwen’s Alipay integration, and WeChat’s "Xiao Wei," manufacturers have chased a universal AI agent that can execute open‑ended commands on a phone, hoping to go beyond specialized AI functions and chat companions.

AI Coffee Ordering Evolution

The author demonstrates the current workflow: the AI locates the appropriate mini‑program page for ordering coffee, then hands the user over to the graphical interface where the user selects options. The AI merely finds the entry; the subsequent interaction is entirely GUI‑driven.

Is Conversational Interaction the Future of Software?

Some claim that apps and traditional GUIs will disappear in favor of dialogue‑based interfaces. The author counters that mature product interfaces result from extensive design iterations, pixel‑perfect mockups, and thousands of A/B tests that optimize user preference and conversion. A generative UI that randomly assembles elements cannot reliably outperform this process.

Furthermore, dialogue is less efficient for information transmission: a GUI can display images, prices, ratings, and parameters within a single card, allowing a user to absorb them in seconds, whereas a conversational exchange delivers data sentence by sentence, reducing density and browsing speed.

Conversational interaction also demands more cognitive effort (System 2 thinking) to articulate precise requests, which many users find difficult, as illustrated by a real‑world example of ambiguous housing listings.

No "Optimal" Solution for Many Tasks

The author notes that tasks like price comparison often lack a single optimal answer because users consider quality, reputation, and personal preferences. Consequently, AI struggles to provide efficiency gains in such subjective decisions.

Even when AI could speed up shopping, the author argues that many users enjoy the browsing experience itself; automating it would diminish the pleasure derived from exploring options.

When Mobile Tasks Aren’t Worth an Agent

Effective AI assistance requires tasks that are complex, have high operational cost, occur frequently, and inspire user trust in the AI’s judgment. Examples include programming, document creation, or PPT design—tasks typically performed on desktops or in B2B contexts.

Most mobile tasks are simple, low‑cost actions completed with a few taps. Even tasks that are frequent (e.g., ordering coffee) lack sufficient complexity, while infrequent complex tasks (e.g., filing taxes) do not occur often enough to justify the agent’s token cost.

Agents also incur high execution costs because each task consumes token resources, unlike traditional software whose marginal cost per user is near zero.

Potential Value: Finding the Entry Point

The author suggests a hybrid approach: combine conversational AI with traditional GUI. The AI can act as an advanced search tool, locating the appropriate mini‑program or feature within a crowded app ecosystem, after which the user completes the task via the familiar graphical interface.

WeChat’s aggregation of millions of mini‑programs exemplifies this use case, but the AI contribution remains limited to entry discovery rather than full task execution.

Mobile as a Communication and Entertainment Device

Phones are primarily used for communication, short videos, chat, gaming, news, and shopping—activities that are ends in themselves. Embedding a productivity‑oriented AI agent into these leisure contexts yields a heavily discounted value proposition.

While AI phones are not entirely useless, the author cautions against expecting disruptive products that revolutionize mobile productivity.