Skip to main content

The ideas AR Labs runs on

The vision and the ideas behind AR Labs' work with AI agents, voice interfaces and context engineering, explained for a newcomer.

A spine of numbered branches from cognitive ecology to the vision, each ending in a small emblem of one concept.A wide drawing sheet: a violet-threaded spine runs from a medallion of a head lit by a flashlight, cognitive ecology, to a striped setting sun, the vision. Eighteen numbered branches leave the spine to small medallions, each a tiny emblem of one concept: a prism, an orrery, a lamp, a rig, nested frames, a thermometer, gates, a queue, a building, a folder tree, memory slots, a gauge, a gantry, an arch, a computer and a compass.VOX01BELIEFS02MIND + MACHINE03THE LOOP04CONTEXT05HARNESS06LOGICAL LEVELS07HOTCOLDTEMPERATURE08CONVERSATION09123ORDERING10ORGANISATION11SECOND BRAIN12CHUNKING13ACCELERATION14FUTURE OF WORK15INTERFACE ARCHITECT16AESTHETIC17VALUES18COGNITIVE ECOLOGYTHE ROOTTHE VISIONTHE HORIZONAR LABSCONCEPTSINDEXTHE IDEAS, INDEXEDSCALE NTS

AR Labs builds AI agents, voice interfaces and the systems that coordinate them. The engineering on the other pages rests on a small set of ideas, and this section explains them for someone meeting them for the first time, whether a person or an AI agent reading the site.

Most of them start from cognitive science. Richard M. Thompson, who founded AR Labs, trained as a cognitive scientist, and our founding framework, cognitive ecology, describes a mind with a small conscious attention and a large unconscious process underneath it. Language models turned out to have a similar shape: a limited context window in front of a vast store of learned patterns. Much of what we do follows from taking the limits of both minds seriously.

The first pages set out where we are heading and why. The AR Labs vision and our mission and values say what we are building towards. What we believe about AI states our positions, and the future of work with AI agents gives our outlook on how work may change. An ecology of mind and machine explains the tagline, the AI interface architect describes the practitioner we think this work needs, and the lab aesthetic covers how the work looks. The Loop shows how AR Labs works, as a five-stage cycle from learning to sharing.

The idea pages follow. VOX, the lab's working system and its central idea, is a voice-first way of working in which people speak and agents answer with visible structure. Context engineering, the harness paradox and the temperature of specification cover how to give AI agents the right information and constraints. Structured conversation and cognitive chunking cover conversations between people and AI. Logical levels, the intelligence is in the ordering and the agentic organisation apply the same thinking to organisations of agents. The second brain shows how anyone can start, and the acceleration paradox explains why we choose carefully what to speed up.

Each page defines the idea, credits where it comes from, and says plainly which parts we have built and which are still design. The engineering pages show the same ideas as working practice.

  • Vision 02
  • Belief 05
  • Process 01
  • Concept 09
  • Framework 03
  1. 02.01Vision

    The AR Labs vision

    Vision

    The AR Labs vision for AI: voice in, concepts out, on an architecture that fits the limits of people and language models, held by individuals and small teams.

    Rev 2026-11-135 min read
  2. 02.02Vision

    Mission and values

    Vision

    AR Labs builds with machine intelligence for four values: connection and collaboration, education and progression, quality of life for all, sustainability.

    Rev 2026-11-134 min read
  3. 02.03Belief

    What we believe about AI

    Belief

    AR Labs' positions on AI: conversation as a main interface, AI that augments people, people keeping judgement, speed with direction, governed agents.

    Rev 2026-11-135 min read
  4. 02.04Belief

    The future of work with AI agents

    Belief

    AR Labs' outlook on the future of work with AI agents: agents carrying defined roles while people frame, supervise and judge, and what that asks of people.

    Rev 2026-11-134 min read
  5. 02.05Belief

    An ecology of mind and machine

    Belief

    The idea behind AR Labs' tagline: people, AI agents and shared records form one system, each mind doing what it is good at, actions meeting a wider check.

    Rev 2026-11-134 min read
  6. 02.06Process

    The Loop: how AR Labs works

    Process

    The Loop is the five-stage cycle by which AR Labs learns, experiments, produces, applies and shares, so that each piece of work feeds the next.

    Rev 2026-11-135 min read
  7. 02.07Belief

    The AI interface architect

    Belief

    The practitioner AR Labs thinks AI needs: a designer of people-to-agent interfaces, combining agent engineering, voice, visualisation and production skill.

    Rev 2026-11-134 min read
  8. 02.08Belief

    The lab aesthetic

    Belief

    AR Labs' visual principles: retro-futurist and vintage-computing inspired, readable first, calm for long thinking, people and AI drawn as collaborators.

    Rev 2026-11-133 min read
  9. 02.09Concept

    VOX: voice in, concepts out

    Concept

    VOX, AR Labs' central idea: speak to AI agents in ordinary conversation and get back a visual map of the ideas and the work, sized for people and models.

    Rev 2026-11-124 min read
  10. 02.10Framework

    Cognitive ecology for people and AI agents

    Framework

    Cognitive ecology, AR Labs' founding framework: small conscious attention frames a large unconscious process, and actions are judged by wider consequences.

    Rev 2026-11-126 min read
  11. 02.11Concept

    Context engineering for AI agents

    Concept

    Context engineering for AI agents as AR Labs practises it: choosing which instructions, knowledge, tools and memory enter a model's context window.

    Rev 2026-11-124 min read
  12. 02.12Concept

    The harness paradox

    Concept

    AR Labs' harness paradox: tightly constrained AI agents do more useful work than loosely configured ones, because each constraint is information.

    Rev 2026-11-124 min read
  13. 02.13Concept

    Structured conversation for voice agents

    Concept

    AR Labs' structured conversation pattern for voice agents: the language model talks freely while code tracks what has been collected and when to move on.

    Rev 2026-11-124 min read
  14. 02.14Framework

    Logical levels in agent design

    Framework

    Gregory Bateson's logical levels as AR Labs applies them to AI agent instructions: nested frames that each govern the level below, records kept outside.

    Rev 2026-11-125 min read
  15. 02.15Concept

    The temperature of specification

    Concept

    AR Labs' principle for planning AI coding agent work: start in design-level language and grow specific stage by stage, leaving code detail to later steps.

    Rev 2026-11-124 min read
  16. 02.16Concept

    The intelligence is in the ordering

    Concept

    AR Labs' view of AI agent orchestration: every actor, human or agent, needs a task list in the right order, so orchestration is ordering and routing.

    Rev 2026-11-124 min read
  17. 02.17Framework

    The agentic organisation

    Framework

    AR Labs' design for a company whose roles AI agents can fill: durable threads of work, seats agents occupy, handoffs, status up and direction down.

    Rev 2026-11-124 min read
  18. 02.18Concept

    The second brain

    Concept

    AR Labs' approach to an AI second brain: your goals, history and preferences in plain files that an AI agent such as Claude Code reads each session.

    Rev 2026-11-124 min read
  19. 02.19Concept

    Cognitive chunking for AI output

    Concept

    AR Labs' method for presenting AI responses one piece at a time, in chunks sized to working memory, so a long answer becomes a conversation.

    Rev 2026-11-122 min read
  20. 02.20Concept

    The acceleration paradox

    Concept

    AR Labs' view of AI productivity: feeling faster is not being faster, and some work, learning above all, should not be sped up.

    Rev 2026-11-122 min read