The Interview

embed-hidden

Key Takeaways from the Discussion

On-device AI is the Future, Putting Privacy and Reliability First

• Running AI locally on-device solves key issues like privacy, security, and control.
• AGI’s vision is to deliver “super intelligence on-device” with GPT-level capabilities without cloud dependency.

 
Agentic AI – App-less, Action-oriented Computing

• Moving beyond apps and building AI agents that can control any app with privacy in mind and execute tasks end-to-end completely on device.
• Core experience is to enable natural conversation, where AI takes action (not just provide answers).
• Long-term vision is to have a “no apps” ecosystem where users simply talk, and devices handle everything in the background.

 
Ecosystem Play: OEM Partnerships, Multi-device Expansion

• AGI is building custom AI models for OEMs such as Lenovo, tailored to specific apps and user needs.
• AGI also has deep collaboration with chip players like Qualcomm to optimize models for NPUs and device hardware.
• Expanding beyond smartphones, AGI is also looking at PCs, wearables, XR, and smart home devices.

Hybrid AI Approach with Scalable Model Strategy

• Models range from sub-1B to 8B parameters to fit different devices.
• Hybrid execution, including an on-device model for fast, private, everyday tasks, and a cloud model for complex queries or older hardware.
• Focus is on performance efficiency (including battery, memory, and latency).

 
UX Focusing on Accessibility as Key Adoption Driver

• AGI is also focusing on UX, designed for human-like interaction, with voice-first, contextual prompts, and confirmations.
• Real-time feedback (including visual cues, widgets) to improve trust and transparency.
• Strong accessibility angle, to help users with motor/vision limitations, enabling hands-free usage in driving scenarios.

Analyst Take

·      Agentic AI has become the next frontier for AI beyond Generative AI to automate tasks securely and meaningfully with the help of agents.

·      AGI has worked from the ground up in optimizing the LLM to the NPUs within the powerful flagship SoCs to leverage the best on-device AI experiences, which are privacy-first.

·      The Agentic automated AI experiences AGI showcased were carefully designed to solve specific problems, from reducing the number of steps of opening multiple apps to boosting accessibility for those who are technically challenged.

·      While AGI has customised or white-labelled its solution for OEMs, the company is also launching its agentic application featuring a personal proactive and context-aware co-worker that leverages its multimodal optimized visual LLM, which learns, plans, verifies and outputs reliably without the need for any system metadata or access.

·      The future is bright for AGI and can scale cross-device, and would become increasingly powerful with reinforcement learning across devices and use cases.