Introducing Eternis

Eternis builds new coordination systems for a world with powerful AI. Our work splits into two directions: forecasting (understanding where the world is headed) and multi-agent systems (enabling different kinds of intelligence to act together).

We imagine a world where every human has digital twins and delegated agents operating autonomously. A human deliberates. A digital twin aggregates preferences and votes on thousands of issues. An autonomous agent executes, monitors, and adapts in real time. Together, they form layers of a single coordination system. Information markets aggregate signals across these layers to produce metrics that inform future decision-making: prices in financial markets, probabilities in prediction markets, premiums in insurance markets.

Today, markets and democracy are the two dominant forms of signal aggregation that civilization runs on. Both are about to change fundamentally. Our research, experiments, and product efforts are oriented toward building the infrastructure for this transition.

Forecasting as a science

In Isaac Asimov's Foundation series, the Second Foundation is built on psychohistory: a mathematical science of predicting the behavior of large populations across long time horizons. It is more important than ever to develop this into a real discipline, since most actors generating new information in the world are about to be digital. Forecasting and decision-making must become a science that incorporates the properties of the agents participating in information exchange.

There is reason to believe long range forecasting is achievable, because prediction at scale does not require modeling every component. You do not need to know the behavior of every cell in the body to know you will be hungry in several hours. Moore's law was not predictive of the specific rules along the path, but it was eerily predictive of the rise in digital compute over 40 years. Someone who anchored all their decisions around it would have captured something fundamental about the trajectory of human civilization. Emergent phenomena can capture the dynamics of societies — of cells, of humans, of agents, and of higher-order structures — into the future. With the rapid increase in information-producing agents that operate at far higher bitrate and bandwidth, the number of emergent properties multiplies over time.

Until now, the constraint has been the lack of sufficiently powerful intelligence applied toward millions of questions at once. For the first time in history, this is possible. We can reason over millions of statements every second, maintaining epistemic status in a continually changing world.

The range of questions is enormous:

• Will liquid biopsy detect more than 10 cancers at over 90% sensitivity by 2030?
• Will there be a Russia-Ukraine ceasefire by end of 2026?
• Will SpaceX achieve fully reusable Starship before 2027?
• When will weakly general AI arrive?
• What is the predicted increase in scientific breakthroughs if global compute increases by 10x?

These questions demand bottom-up analysis, geopolitical reasoning, and careful navigation of deep uncertainty. Answering them well requires world models that compress the relevant structure of reality, calibrated systems that are honest about what they don't know, and robustness testing that ensures the answers hold up under pressure.

Why coordination needs an upgrade

As AI automates cognitive work, coordination systems designed for decision-making at human speed will crumble. The current infrastructure was built for a world where humans were the only agents making decisions. That world is ending.

Without new coordination mechanisms, we will not get a society where intelligences of different bandwidth and throughput can meaningfully participate together in shaping the world around us. A digital twin that can process a thousand policy proposals per day is useless if there is no governance infrastructure to receive its input. An autonomous agent that can create new software and spin up capital structures in microseconds will continue building forever, eventually with increasing divergence from human preferences, unless there is repeated calibration from the digital twins — and the digital twins will diverge from the humans unless there is calibration from the humans.

The explosion of useful computation will make coordination a central bottleneck. We have the intelligence. We lack the systems to channel it.

What we are working on

• Forecasting systems that train models to predict better, maintain cheap, continuously updating world models, and take calibration seriously
• The shape of knowledge and metascience: investigating the platonic hypothesis more deeply, exploring different formulations of information complexity, and understanding how world models compress reality into useful representations
• Decision products: Axion, our product for making deeper, better-informed decisions, grounded in structured evidence rather than intuition, and Lume, our agent-focused prediction market platform
• Multi-agent coordination: researching emergent properties at scale when agents interact, running multi-agent RL experiments, and building environments where diverse agents (human, twin, autonomous) delegate and verify each other's work

The end goal

The ultimate ambition is a society where science is increasingly automated, human preferences are aggregated efficiently, governance decisions are tracked and verifiable, and the allocation of compute toward shared societal goals is made using forecasting systems that enable good interventions today.

We believe this is achievable. The components exist. What is missing is the engineering: building systems that tie forecasting, world models, and coordination infrastructure into something that actually works at scale.

If contributing towards solving societal coordination in a world with superhuman intelligence is exciting to you, please reach out: contact@eternis.ai