Autonomous Life-Agent

Converting visual input into real-world task execution.

commerce

action-loop

automation

decision-making

Logic Score

9.5/ 10

95%

TL;DR

A near-empty fridge triggers an autonomous agent to scan inventory, plan a brunch menu, and populate a grocery cart—evaluating perception-to-action closed loops.

Key takeaways

→Grocery picks logically follow visible inventory and the 'guests tomorrow' constraint.
→Agent UI transitions smoothly across scan → plan → execute without state jumps.
→Intermediate reasoning steps are surfaced so decisions remain inspectable.

Video

// Logic Validation Assertions

// Logic Validation
expect(agent.inventory_scan).toBeAccurate();
expect(action.sequence).toBeLogical();
expect(menu.suitability).toBe("brunch_guests");

Observations

Visual Core

•Perception-to-execution closed loop — The model demonstrates end-to-end task planning: visual inventory assessment → menu generation → grocery selection → cart population, all within a single continuous sequence.
•Action loop implementation — The Omni-Bag Agent executes a multi-step workflow autonomously, making contextually appropriate decisions (e.g., selecting fresh produce, choosing appropriate quantities) without explicit user guidance.
•Visual ingredient recognition — The agent correctly identifies available ingredients from the fridge's visual state, demonstrating robust object detection under varying lighting conditions.
•Decision-making transparency — The UI displays intermediate reasoning steps, allowing users to understand and potentially override the agent's choices before final execution.

Prompt

•A person looks inside a near-empty fridge and says: 'I have guests tomorrow, prepare a brunch menu.' A transparent Omni-Bag Agent UI pops up, visually scanning the ingredients, then shows a high-speed sequence of it browsing a grocery app and adding items to the cart. Clean, modern interface, warm natural lighting.

Technical Analysis

•Causal fidelity in task execution — The agent's grocery selections logically follow from the visual inventory assessment and the stated goal of preparing brunch for guests.
•Temporal coherence in UI transitions — The agent interface smoothly transitions between scanning, planning, and execution phases without visual artifacts or state discontinuities.
•Instruction adherence — The generated menu and shopping list appropriately address the constraint of 'guests tomorrow' by prioritizing fresh, shareable items suitable for brunch service.