Area2 AI

Coding agents are steered by what we type, so their level of detail, questions and progress updates may not fit the person or the moment. Open Presence × Muse adds interaction context beyond the words in a prompt, so the agent can adapt how it collaborates at supported request boundaries. In an early, unblinded pilot with one known participant, effort was rated lower with Presence in 29 of 33 paired comparisons (88%). Area2 is building the Human Presence Layer for AI. This working prototype brings that layer to Meta's Muse Code. Message text and coding context go separately to Muse; they are not sent on the Open Presence route. Presence uses interaction metadata, such as timing and key categories, never message text or code, and turns it into short-lived guidance about response depth, optional updates and conversation style. The workspace makes this capability inspectable through two isolated native Muse Code sessions on the same starting task. Only one receives eligible Presence guidance at supported request boundaries. Developers can converse with either or both, steer ongoing work, review approvals and compare the results and their own experience. Prepared tasks include common code checks and a separate action to send each agent its own test feedback; developers can also bring their own task. Judges can run the side-by-side workspace using the judge setup in the repository. The workspace can hold optional updates when the policy calls for it. Native questions, approvals and final answers remain visible; held updates are revealed when the agent needs attention or completes its turn. During the hackathon, we built the Muse-specific integration, interactive comparison and evidence workflow around existing Open Presence inference, Cipher and selected Area2 components. The repository identifies the event-built contribution and reused foundations. The overall pilot assessment is promising for collaboration experience. In 33 paired ratings from one known participant in a repeated, unblinded exploratory pilot, effort was lower with Presence in 29 of 33 rated comparisons (88%) and noticed interruptions were lower in 25 (76%); four reported more interruptions. Mean effort rating fell 56% (from 3.82 to 1.70) and noticed interruptions fell 41% (from 4.82 to 2.85 per run). Across 51 paired resource records, output tokens fell 14%; the input profile differed and timing varied. These observations do not establish causal benefit, general productivity gains or equivalent technical quality. Muse Code is one application of Area2's broader ambition: AI that can adapt how it collaborates with people across tasks and tools. Presence is designed to slot into existing products without model changes: guidance arrives alongside each request, and message content never travels on the Presence route. Any surface where people type to an AI, such as Meta AI, Muse Code or AI chats in Meta's messaging apps, could use the same signal. We would welcome the chance to pilot Presence with Meta product and research teams: contact@area2.ai.