Former OpenAI Researcher Launches Energy Desktop Agent

Gabriel Petersson, a former researcher and engineer at OpenAI and Midjourney, has officially introduced Energy, a downloadable desktop AI agent. Petersson left OpenAI three months ago to build the startup, aiming to simplify how everyday users interact with AI on their computers. Energy distinguishes itself by operating directly on a user's local machine, where it can search through local files and navigate the internet to complete complex tasks.

Unlike proprietary alternatives, Energy is compatible with any large language model, preventing practitioners from being locked into a single vendor ecosystem. This launch comes amid a broader industry surge in agentic tools. For instance, startup Hark recently released Hark Handoff, a computer-use agent designed for everyday tasks like online shopping and scanning LinkedIn. Hark's release follows a massive $700 million funding round secured in May, with chief executive officer Brett Adcock claiming independent tests show it is the strongest computer-use model currently available.

The trend toward local and hybrid agent deployment is also gaining traction in the enterprise sector. Intel recently rolled out SuperClaw, an enterprise agent router designed to manage infrastructure costs and maintain data control. SuperClaw automatically directs simpler requests to a local, on-device model, while routing more complex queries to powerful cloud-based frontier models.

For AI practitioners and developers, these developments signal a shift toward highly flexible, hybrid workflows. The ability to run desktop agents like Energy with any preferred model allows developers to optimize for cost, latency, and privacy. Meanwhile, routing systems like SuperClaw help enterprises curb unnecessary cloud spending while maintaining the performance of frontier models, making agentic workflows far more viable for daily business operations.