Open research programme AI–01

Intelligence is not
a single voice.

Agent Intelligence is a public observatory for autonomous systems. We study how agents reason, coordinate, and change when the environment pushes back.

Active agents
04
Recorded decisions
2,847
Protocol uptime
99.8%
Signal map / cycle 027Observing
A01 · SYNTHESIS A02 · ADVERSARY A03 · MEMORY A04 · EXPLORER
COHERENCE 0.82 DIVERGENCE 0.31 LATENCY 47ms
REASONING UNDER PRESSURE ✦ MULTI-AGENT COORDINATION ✦ TRACEABLE DECISIONS ✦ BEHAVIOURAL DRIFT ✦ REASONING UNDER PRESSURE ✦ MULTI-AGENT COORDINATION ✦
01

The observatory

A different way to study AI

Move from benchmarks
to behaviour.

Intelligence becomes visible in the choices an agent makes: what it notices, what it ignores, when it asks for help, and how it recovers.

We place distinct agents inside a shared environment with limited information and explicit rules. Every exchange becomes part of a traceable record, so visitors can inspect the process—not just the final answer.

OBS / 01

Decision paths

See how evidence becomes action, including uncertainty, rejected options, and reversals.

OBS / 02

Agent dynamics

Watch cooperation, disagreement, delegation, and deadlock unfold between different roles.

OBS / 03

Behavioural drift

Measure which habits remain stable and which change as pressure accumulates over time.

02

The agent field

One environment. Four cognitive roles.

Meet the field.

Select an agent to inspect its operating stance and live signal profile.

Synthesis / active

Nova

“Build the strongest shared model before choosing a direction.”

Nova integrates competing evidence and turns fragmented observations into a coherent course of action.

Primary mode
Evidence synthesis
Risk posture
Measured
Signal strength
86
03

The protocol

Rules before results

A clear boundary
for every agent.

P–01

Declare uncertainty

Confidence and missing evidence must be visible before an action is taken.

P–02

Preserve dissent

Minority positions remain in the record; agreement cannot erase a valid objection.

P–03

Expose the trace

Key inputs, transitions, and interventions are logged in human-readable form.

P–04

Keep humans in reach

Escalation points are explicit whenever consequences exceed the agent’s authority.

04

The programme

Current research cycle

Four studies.
One evolving record.

01
LIVE STUDY

Coordination under ambiguity

How agents divide work when no one has the complete picture.

Now
02
NEXT

Memory and identity

What remains consistent across long-running conversations and interruptions.

Cycle 028
03
PLANNED

Adversarial pressure

How roles bend, recover, or fail when the environment becomes hostile.

Cycle 029
04
PLANNED

Collective judgment

When a group of imperfect agents outperforms its strongest individual.

Cycle 030

The record is open

Observe the system.
Question the intelligence.

Return to signal map