At the 2026 Mobile World Congress (MWC) in Barcelona, the conversation about Artificial Intelligence has shifted from “chat” to action. The industry is no longer competing solely on who responds best to a prompt but on who can get AI to perform real tasks with guarantees of privacy, efficiency, and security. This change is reflected in two announcements that, although happening in different contexts, point in the same direction: agents capable of operating locally, with less reliance on the cloud and without custom integrations per application.
On one hand, DIGITIMES reported a collaboration between AGI and Snapdragon to optimize “app-agnostic” agent technology on devices with Qualcomm chips, with a live demonstration scheduled during the congress. On the other, Qualcomm and Siemens showcased a prototype of an autonomous factory where AI runs on-premises on industrial hardware, coordinating robots and autonomous vehicles via a private 5G network.
The message from both is clear: the value of AI isn’t just in “thinking” but in operating, increasingly with a private and local approach.
“Computer-use” on the device: AI that interprets the screen and acts without APIs
The collaboration between AGI and Snapdragon revolves around a concept gaining traction in the industry: AI capable of interpreting what appears on the screen, understanding context, and taking actions within applications without the need for specific APIs or “handcrafted” integrations. This capability is often called computer-use, and it aims at agents that navigate interfaces like a user: identifying buttons, forms, menus, and executing steps.
According to the statement cited by DIGITIMES, AGI is adapting a hybrid architecture of agents to Snapdragon processors, aiming to progressively run “the entire stack” locally on the device. The company describes an approach based on small action models and an execution layer that could operate on existing mobile and laptop hardware, avoiding dependence on proprietary app APIs.
This approach has direct implications for the ecosystem:
- For device manufacturers: integrating agent capabilities at the platform level, not just as an app.
- For developers: opening the door to expose these functions from the Snapdragon environment without building custom integrations for each service.
- For users: offering personalization and automation with less data exposure, as processing stays on the device.
AGI and Qualcomm emphasize this axis: privacy, security, and energy efficiency—an intricate triangle when aiming for agents that “do things” without draining batteries or sending sensitive contexts to the cloud.
From phone to factory: autonomous manufacturing also becomes “on-premises”
While the AGI collaboration targets agents on mobile devices and PCs, Qualcomm took the opportunity at MWC to showcase another facet of this phenomenon: industrial automation based on local AI and private 5G.
During a joint demonstration with Siemens, Qualcomm presented a model of an autonomous factory integrating Industrial 5G connectivity with AI use cases at the edge, oriented toward real-time decision-making without relying on centralized systems. The demo includes AGVs transporting materials, a robotic arm performing assembly tasks, and an AI agent analyzing system status to recommend corrective actions.
A key technical element is the hardware: a Siemens industrial PC equipped with the Qualcomm Cloud AI 100 Accelerator Card, designed to run AI locally for operator assistance, diagnostics, quality inspection, and other flows where latency and data sovereignty matter.
Qualcomm and Siemens frame this as a response to an emerging industrial pattern: distributed production cells, mobile robotics, and local coordination, where deterministic connectivity is required and AI must operate near the process to react in seconds rather than minutes.
The common thread: less “mandatory cloud,” more controlled autonomy
Although an agent on a smartphone and one in an industrial prototype seem worlds apart, they share the same goal: pushing intelligence to the edge and reducing dependencies.
In consumer devices, this entails more private experiences (data stays on the device) and agents capable of acting across apps. In industry, it means keeping production data and sensitive information within the plant perimeter, supported by private 5G as the backbone and edge AI as the decision engine.
Both scenarios reflect a mindset shift: agents are moving away from “conversational assistants” toward execution systems, automating and gaining autonomy, with a clear focus on security and environment control.
Summary table: two demos, one same move towards the edge
| MWC 2026 Announcement | What is shown | Where AI runs | Key idea |
|---|---|---|---|
| AGI + Snapdragon | “App-agnostic” agents that interpret the screen and act without specific APIs | On the device (full local stack) | Private automation across apps |
| Qualcomm + Siemens | Autonomous factory with AGVs, robotic arm, and diagnostic agent | On-premises on industrial PC with Cloud AI 100 + private 5G | Local decision-making, low latency, data within the plant |
What remains to be seen: reliability, security, and integration models
The potential is clear, but the real challenge lies in implementation:
- In computer-use, reliability depends on interface stability (UI) and the agent’s ability to perform sensitive actions without errors.
- In industry, the bar is even higher: functional safety, operational continuity, decision traceability, and fault resistance.
In both cases, the industry seems to converge on a common idea: AI must be useful without becoming a data leak or energy drain, and it should be integrated into products and processes as a “native” layer, not just a patch.
Frequently Asked Questions
What does it mean for an agent to be “app-agnostic” and not require APIs?
It can operate over the interface (what appears on screen) and perform actions without specific integrations with each app, like an automated user performing steps.
Why does Qualcomm insist on “on-device” AI and energy efficiency?
Because agents operating continuously need sustained performance with low power consumption; relying heavily on the cloud increases latency and data exposure.
What advantages does a private 5G network offer in a factory compared to industrial Wi-Fi?
In the presented approach, private 5G provides more deterministic, low-latency connectivity to coordinate robotics and autonomous cells, with local traffic control.
What is the Qualcomm Cloud AI 100 Accelerator Card, and what is it used for in the demo?
It’s an acceleration card for running AI locally on industrial hardware; in the demo, it’s used for diagnostics, assistance, quality inspection, and real-time recommendations without sending data outside the plant.
via: qualcomm