A quieter kind
of intelligence.

A relationship with continuity.

My work moves between buildings, spreadsheets, conversations and decisions. Context accumulates slowly. A rent comparison carries a date and a source. A client brief carries a preference that may change. A useful assistant needs to hold those distinctions over time.

LIFE began with a more speculative question about artificial life. The work now turns that curiosity into design questions I can test: what should persist between sessions, how should a system recognise a correction, and how much freedom should it have to act?

I use the image of a living system to explore continuity, adaptation and restraint. These are design metaphors. Whether an SI has subjective experience remains outside what this project can demonstrate.

A memory should carry its evidence.

The proposed architecture separates working context, longer-term memory and the rules that govern behaviour. Models may change. The history of a decision should remain readable.

Research on Generative Agents combines memory, reflection and planning to produce coherent behaviour in a simulated environment. That is a useful starting point for LIFE. My concern is what happens when the material is a real brief with conflicting versions, missing figures and consequences for another person.

Every retained fact should be traceable to its origin and date. A correction should update the record without hiding the change. Uncertainty needs to survive the summary. In office research, one unmarked assumption can travel through a workbook and become a confident recommendation.

Permission is part of the architecture.

I want LIFE to prepare useful work with fewer interruptions. I also want the boundary around that work to be visible: a draft remains a draft, an external action has an owner, and a stop control is always within reach.

Autonomy without a clear boundary can make a useful system difficult to trust. For LIFE, my design response is to attach permissions to the action itself. Access to a tool does not silently grant authority over everything that tool can reach.

“The system should make its next move understandable before it makes that move consequential.”

The direction is local first, with explicit access to files and services. Changes to important records should be reviewable. Recovery should be designed alongside action. These are requirements for the prototype, rather than claims about a finished product.

One ember. Room to think.

The visual identity follows the same idea. Washi, paper and ink form the surface. Alegreya gives documents their voice; Alegreya Sans carries the conversation. A single ember introduces warmth.

The flame explores six states: bloom, grow, balance, contract, dormant and freeze. Its movement makes a state legible without filling the screen with instruments. The preview above studies that visual language; it is not connected to a running SI.

I’m also exploring how continuity could move between a computer and a phone. That raises practical questions about handover, conflicting changes and limited battery life. A quiet system needs to understand the cost of staying present, including the cost to a person’s attention.

Small experiments. Honest results.

LIFE is currently in development. The identity prototype and architectural direction exist; the system has not been given a production identity, access keys or real working data. The next stage is to test a small, bounded assistant against tasks I know well.

I will judge the work by those outcomes: accuracy, recoverability and the effort it takes to stay in control. Any published result will need a stated task, a method and a limitation.

That is the ambition behind LIFE: personal intelligence that becomes more useful as the work grows, while keeping a clear place for human judgment.