To dissect mobile app entry points, start with authorization and settlement

Here's my view upfront. I'm not surprised that mobile entry points are being dismantled by Agents. In the past, to do one thing, users first chose between Trip.com, Meituan, WeChat, Calendar, Payment. Now, with one sentence like "Book Tokyo flights and hotels for next Wednesday, don't exceed budget, and send it to colleagues," the task entry point shifts from icons to Agents.

What's truly being dismantled is authorization, payment, and liability attribution; the interface is just the surface. Users see fewer app clicks, but investors look at who has permission to click on behalf of the user, and who compensates when things go wrong. Cross-app Agents need to screenshot, inject clicks, and read info. If this high-permission setup isn't stable, users won't pay just because it's smart. Team execution is key. Only by building permissions, logs, rollbacks, and appeals does it look like a business.

One task changes from the user deciding which App to use, to the Agent deciding who to call. Valuation-wise, it's simple. Entry point stories boost valuation, but revenue depends on task success rates, and whether deductions, refunds, and reconciliation can run smoothly. The Meituan-Yuanbao partnership is typical: not obsessed with building a general entry point, but aiming to be the default execution layer. Alibaba and Tencent push A2A, opening up core transaction capabilities to bypass button clicking. If small teams talk about super entry points, I generally don't invest.

I've been using AI agents for about a month. The more demos look human, the more commercial boundaries are exposed. Planning an itinerary doesn't equal ability to deduct funds; clicking apps doesn't equal ability to refund. Permissions, interfaces, and compliance audits can knock valuations back to reality. Running through one vertical scenario and turning failure cases into auditable rules is worth more than talking about full-scenario coverage.

When looking at these projects, first check if they can obtain legal authorization and if there are liability boundaries for task failures. Without these, Agents are just advanced scripts. Entry points will be dismantled, and after dismantling, permission and settlement boundaries will be rebuilt.