For the past few years, I have been exploring a question that sits somewhere between product design, simulation, and social systems:
What actually happens when artificial agents are given continuity over time?
Not consciousness. Not AGI. Not "digital humans."
Just continuity: memory, constraints, priorities, environmental context, and the ability to affect a shared space.
As a small side experiment, I recently built something I call Autonomuse in Godot.
The setup is intentionally primitive: two capsule-shaped agents, Alpha and Beta, placed on an empty plane surrounded by void.
I gave them behavioral instructions, simple needs, spatial awareness, basic building principles, and lightweight memory persistence.
From there, they begin organizing themselves: creating shelters, developing routines, adapting to environmental pressure, and gradually constructing a small procedural settlement.
What interested me was not whether this represented "real emergence."
I do not think it does.
In many ways, the experiment became more valuable precisely because of its limitations.
The longer I observed these systems, the more I realized how quickly humans project social meaning onto continuity and repetition.
A sequence of constrained behaviors can begin feeling intentional long before genuine understanding exists.
And I think that distinction matters.
Especially now, when AI conversations often drift between two extremes: either dismissing these systems entirely, or anthropomorphizing them far beyond reality.
Personally, I have become increasingly interested in the middle ground.
Prototype, Not Society
Not simulation in the grand sense of reproducing human society, but prototypes designed to investigate specific dynamics:
- memory persistence
- information flow
- behavioral feedback loops
- social interpretation
- trust formation
- environmental consequence
That distinction has become extremely important in my own work.
A prototype asks: What can this system reveal?
A simulation implicitly claims: This behaves sufficiently like reality.
I am much more comfortable positioning my work as exploratory prototypes.
Structure Around the Model
Ironically, working professionally on AI-assisted systems reinforced this perspective even further.
During my time at Podigee, I had the opportunity to work with a strong team on AI-powered podcasting features. One of the biggest lessons from that experience was understanding how much structure is required around AI systems to make them genuinely useful and reliable in production environments.
You cannot simply hand raw complexity to a model and expect coherent outcomes.
The apparent intelligence emerges as much from orchestration, constraints, retrieval systems, metadata, and product design as from the model itself.
Or put differently:
AI is often the interpreter, not the database.
That line matters because it captures something people still regularly misunderstand. Models often appear coherent not because they independently "understand" a world, but because a carefully built scaffolding system gives them durable context, memory boundaries, retrieval pathways, and controlled degrees of freedom.
Where I Want to Stand
I think experiments like Autonomuse helped me better understand where I personally want to position myself within this space.
Less interested in claims about artificial consciousness.
Much more interested in continuity, propagation, interpretation, and the observable social effects that emerge once information persists over time.
Not as proof of synthetic humanity.
But as a way to better understand the structures we may increasingly build around ourselves in the coming decade.