Cognition Infrastructure
Memory, planning, perception, reasoning. The substrate that compounds beneath every embodied system.
— Experiments / Vol. I / Operations room
Engineering The Future, Via Experiments.
An operations room, not a publication. The institute's work organised into three structural layers — the engineering domains we operate inside, the active experiments in motion within them, and the field instruments being built beneath both to engineer the physical AI era through real-world conditions.
What we are running now · Engineering Domains · Active Experiments · Field Instruments
Live operations / what we are running now
The Experiments Under Active Operation This Quarter.
A subset of the programme — the experiments currently absorbing the most field input, and the typed signals that are shifting engineering positions in real time.
-
E.04 · Live Active · this quarter Operator Class For Physical AI
The integrator gravity shift is reshaping who actually mediates operator trust. Position under revision.
-
E.06 · Live Active · this quarter Sovereign Procurement Frameworks
Three early-draft frameworks observed. Likely to constrain operator selection by 2027 — earlier than the field models.
-
E.01 · Live Watching · cross-corroboration Humanoid Integration Economics
Integration cycles averaging 4.7 months across an 11-operator sample. Position holding; corroboration deepening.
Framework / Vol. I
The Physical AI Stack.
An interactive five-layer framework — embodiment, perception, world model & cognition, action & control, and the learning loop — with every component clickable.
Layer A / Engineering Domains
Six Domains The Institute Holds Positions On.
Domains are not topics. They are operating frames — the way the institute decides what counts as an answer.
Programme anchor
Cognition-adjacent
Sovereignty-adjacent
Embodied Intelligence
Humanoids and embodied platforms as policy objects — geometry, dexterity, labor profile, supply chains.
Deployment Systems
Integration, certification, telemetry, retraining, recall. The unglamorous half of intelligence.
Strategic Autonomy
Autonomy as a strategic surface — the control hierarchies, accountability structures, and political weight of self-operating systems.
Intelligence Architecture
The institutional shape of knowing. What states, operators and labs actually need to perceive the field clearly.
Civilizational AI Dynamics
The long-horizon trajectory: capital, governance, labor, and the metabolism of institutions inside the physical AI era.
Layer B / Active Experiments
The Standing Portfolio — Eight Long-Horizon Experiments In Operation.
Each experiment is a standing operational track — refined as field evidence accumulates, not republished as a new opinion every quarter. Status reflects current activity, not importance.
-
E.01 Watching · cross-corroboration Industrialization Of Humanoids
From credible demonstrations to ten million units per year: actuators, training data, regulation, capital intensity, and the labor profiles humanoid platforms will compete with first.
-
E.02 Active Post-LLM Cognitive Architectures
The full vertical between sensor and intent — perception, world models, memory, planning — and the deference policies that decide when a machine should yield to a human.
-
E.03 Watching · continuous Autonomy Liability Stack
The legal, insurance and political architecture around autonomous systems in public space — what permits scale, what halts it, and where the breakpoints will be.
-
E.04 Active · this quarter Multi-Agent Deployment Systems
Coordinating fleets of agents in the physical world — supervision, fallback hierarchies, telemetry, and the operating discipline that decides where multi-agent autonomy is real.
-
E.05 Watching · continuous Compute, Capital & Geopolitics
The physical and political topology of where intelligence is manufactured — chips, energy, fabs, labs, capital — and the strategic positions that follow.
-
E.06 Active · this quarter Sovereign Intelligence Architectures
How states are positioning across the stack — compute, models, operators, regulation — and which configurations actually constitute sovereignty in the physical AI era.
-
E.07 Active · primary watch Operator Doctrine For Physical AI
Who runs the fleets — the new operator class for humanoid and autonomous systems, their internal structure, accountability surface, and economics.
-
E.08 Draft · meta-inquiry Civilizational Time Constants
At what speed do institutions, infrastructure, labor markets and political systems actually metabolize the substrate being built? The honest answer reshapes every other position.
Layer C / Field Instruments
The Instruments Being Built To Perceive The Field.
Experiments compound into instruments — maps, atlases, taxonomies, doctrine systems. Released only when the underlying structure is stable enough to draw honestly. Until then, we resist the false map.
- 2026 · Q1 I.02 Cognition Stack Atlas
- 2026 · Q2 I.03 Deployment Failure Taxonomy · I.05 Humanoid Industrialization Matrix
- 2026 · Q3 I.01 Battlefield Map · live cartography update
- 2026 · Q4 I.06 Operator Manual for Physical AI
- Restricted I.04 Compute Sovereignty Map · ongoing
Scheduled release
Restricted instruments held until structure stabilises
Physical AI Battlefield Map
A live cartography of the institutions, capabilities and dependencies competing across the physical AI substrate — at a resolution that supports real decisions.
Cognition Stack Atlas
The full vertical of embodied cognition — perception, world models, memory, planning, deference — mapped layer by layer, with the open frontiers marked.
Deployment Failure Taxonomy
A structured catalogue of how physical AI systems actually fail inside real institutions — integration, regulation, trust, accountability, supervision.
Compute Sovereignty Map
Where intelligence is being manufactured — chips, energy, fabs, talent, capital — and which sovereign positions are real, which are aspirational, which are quietly lost.
Humanoid Industrialization Matrix
The path from credible demonstration to industrial scale — actuators, supply chain, training data, regulation, operator class — held together as one decision surface.
Operator Manual For Physical AI
A field manual being assembled in public: how autonomous and embodied systems should be staged, supervised, audited and retired inside real institutions.
Engage / 04
Access To The Programme Is Structured, Not Open.
Engagements, dossier access, and instrument previews are made available to a small number of institutions operating directly on the physical AI horizon.
Let's Talk