Forgotten Time and Space
Derek Rosenzweig · Runtime Labs · July 22, 2026
Stateless Interfaces
It is fitting that our experience with technology is largely stateless. We live in time and space, but meet the world through interfaces that are not grounded in either.
The problem is not simply that these interfaces show us too much information. It is that they present information without event structure. A feed delivers images, video, and text as isolated fragments stripped of the time, place, and sequence in which they belonged.
The feed is grounded in our preferences, not in our time and location.
As information loses its setting, the boundaries of everyday life thins out too. Homes become offices, weekends bleed into workweeks, and the story of our days drifts apart from the times and places where our lives actually happen.
Event Structure and Memory
Human memory did not evolve to digest a feed. It evolved to organize continuous sensory life into event—bounded episodes with a before, a during, and an after.
We segment experience in time. We bind those segments to place. We order them so that sequence is legible. And we use that structure to learn consequence: what our actions caused, where, and when the outcome appeared—even across delay.
Coherence in time is not decoration. Event structure is how learning at inference becomes stable and metabolically cost-effective. Without temporal position, spatial setting, and causal relation, experience remains a pile of fragments rather than a path we can use.
Interfaces Without Events
Our dominant tools organize information in ways that are not native to this kind of memory.
Feeds rank fragments by relevance, not by where or when they belong in a life.
Search lifts facts from their original setting and returns a list of options.
Notifications pool contexts across platforms into a single undifferentiated surface.
Calendars often reduce time to slots for obligations rather than a visual record of what unfolded.
Chat interfaces preserve a transcript, but rarely represent where each statement belongs in a person’s changing life.
The same absence runs through language-model memory. The problem is not simply forgetting. A system could retain every sentence a person ever wrote and still fail to understand when something was true, where it applied, what changed afterward, or which future decision it should inform. Retention without temporal, spatial, and causal position is not understanding.
There is no law that requires a social media platform to center on a feed.
There is no law that prevents a calendar from becoming a visually informative timeline.
There is no law that time software must only schedule meetings we would rather skip.
Artificial Grounded Intelligence
This suggests an alternate and complementary north star: Artificial Grounded Intelligence.
Artificial Grounded Intelligence is intelligence situated in time, place, sequence, and consequence. It does not treat each prompt as an isolated request or each memory as a free-floating fragment. It understands information as part of an unfolding world—what has happened, what is happening now, what is expected next, where the person is, and how present choices relate to past experience and future plans.
Longer context windows and larger models are not the same as improved grounding.
But better grounding compounds as context and scale grow. Grounding is not preserving more text. It is giving information temporal, spatial, and causal position—so that what is remembered remains usable as the world moves.
Óra is an attempt to build an interface and form factor for intelligence grounded in time. It does not treat the timeline as an endless stream of content. It treats events as durable units of context, around which conversations, notes, photos, links, locations, plans, and model responses can accumulate.
The timeline becomes a model of a life and points to the latent space that connect between your moments: what happened, what changed, what should be stored, and how the past and present should inform what comes next.