Qualcomm CEO Cristiano Amon told CNBC this month that AI agents are going to become "the new app." His example: instead of opening a banking app and hunting through menus, an AI agent simply retrieves your transaction details for you. Apps aren't dying, he said — but the way we interact with them is about to change completely.
This isn't just one executive's opinion. Qualcomm backed up the claim by revealing it has over 40 new agent-first devices in active development, spanning smart glasses, camera-equipped earbuds, wearable pins, and even jewelry — all designed to be "endpoints for agents" rather than app launchers.
What Counts as a Real AI Agent
Here's where most of the conversation gets muddy. A true AI agent doesn't just respond to one prompt it takes a goal, breaks it into steps, decides what actions it needs to take, executes them, and adjusts if something doesn't go as planned. The defining trait isn't intelligence. It's autonomy across a multi-step task.
A lot of products marketed as "AI agents" right now are really just a single call to a language model that returns one answer. That's not agentic behavior that's autocomplete with a nicer interface.
- •Interfaces are shifting from tap-and-scroll to stated goals. Users may stop thinking app-by-app and start asking for outcomes instead.
- •The competitive battleground is moving to hardware and on-device performance. Companies that win the "endpoint" the glasses, the earbuds, the wearable get to decide how their agent reaches your business.
- •Software businesses face a real distribution question. If agents become the primary interface, how do you make sure an agent recommends your product instead of a competitor's?
How to Tell If You Actually Need an Agent
At Strayks, when a client asks us to build "an AI agent," our first question is always: what decision should it make on its own, and what happens if it gets that decision wrong? That single question filters out most "agent" requests that are really just well-disguised automation which is fine, but it's a simpler and usually cheaper thing to build than a true autonomous agent.
The technology is genuinely powerful when matched to the right problem. Customer support triage, multi-step research, and cross-service task handling are strong fits right now. Fully autonomous decision-making in high-stakes areas like finance or healthcare still needs a human checkpoint, regardless of what the marketing copy promises.
The Practical Takeaway
You don't need to rebuild your product around agents tomorrow. But if you're planning a new app or a major feature this year, it's worth asking whether part of that experience could be agent-accessible — meaning a future AI agent could complete a task on a user's behalf without your product needing its own dedicated app screen for it. That's a very different design question than the one most teams are used to asking, and it's the one Qualcomm's bet suggests will matter more every year from here.
