Give AI a world to reason from.
Models can reason, research, write, code, use tools and take action. But capability alone does not tell an agent what is true here, now—or how to coordinate when evidence, intent and authority conflict. Orient gives agents a maintained organisational reality in which to reason, coordinate and act.
Models provide intelligence. Agents provide agency. Orient provides shared reality.
MACHINE REQUEST
What should we believe about the enterprise AI rollout now?
Capability is not state. Before it can reason or act well, the machine needs the organisation’s current position, uncertainty, history and authority.
Inherited organisational state
LIVE
Current understanding
Developing
Evidence
12 linked items
Uncertainty
2 open questions
Recent change
18 min ago
Authority
Research + draft
Formed once. Shared across people and machines.
Context is not a world model.
Documents, search, memory, larger context windows and MCP can provide material. Orient’s live model forms that material into maintained organisational reality—the state agents inherit before they reason.
Current understanding
What the organisation believes now.
Evidence
Why that view is supported.
Provenance
Where the understanding came from.
Uncertainty
What remains thin, contested or unresolved.
Decisions
What was decided, and in what context.
Expectations
What the organisation thinks happens next.
Situations
What is under continuous attention.
History
What was believed before and why it changed.
Authority
What the machine may observe, propose or do.
Stop reconstructing the organisation on every run.
Most AI begins with a request, retrieves material, rebuilds the situation, reasons, acts—and starts again next time. Orient changes the architecture: the world persists between runs, independent of any single model or session.
Every run starts again
Context is rebuilt for each interaction.
01
Request arrives
02
Search and retrieve
03
Reconstruct the situation
04
Reason and act
05
Lose the working state
The world persists
Each run begins from maintained state.
01
Live organisational state
02
Request or event
03
Relevant understanding and authority
04
Reason, propose or act
05
Observe and update the state
The world persists between runs.
Agents do not automatically become organisations.
Agents can parallelise work brilliantly. But when work depends on shared state, trust, disagreement, conflicting objectives, ownership or recourse, coordination—not capability—becomes the bottleneck.
Shared reality has structure
A maintained institutional reality outside any single model gives people and agents a governed world in which to operate.
World model
What is true, believed, uncertain and changing.
Epistemics
Evidence, provenance, reliability, contradictions and confidence.
Intent
Objectives, constraints, priorities and commitments.
Authority
Who may decide, change, approve or act.
Coordination
Dependencies, ownership, responsibilities and other actors’ intentions.
Memory
What happened, what was decided, why and what followed.
Recourse
What happens when agents disagree or cannot establish reality.
STATE, NOT JUST DATA
What agents inherit from the live model.
A prompt describes the task. The live model supplies the organisation’s position, evidence, uncertainty, commitments, history and authority before an agent reasons or acts.
Position
The organisation’s current stance—including dissent and what would change it.
Uncertainty
What is open, provisional, contested or under pressure.
Time
When the view changed, what preceded it and which assumptions failed.
Authority
Who may investigate, propose, approve, escalate or act.
Situations turn state into watch conditions.
Because Situations are maintained against the live model, an agent can distinguish meaningful change from noise—and know when to investigate, propose or escalate.
Agents can watch, not just answer.
A Situation keeps an important part of the world current, rather than waiting for somebody to remember to search.
EU AI regulation
UNDER PRESSURE
ASSUMPTION
Governance remains secondary to capability
LATEST EVIDENCE
New guidance increases the expected burden
CHANGED
18 minutes ago
NEXT
Review the maintained view
Authority is part of the model.
Intelligence is not authority. An agent also needs explicit boundaries—and a path for recourse when actors disagree, detect conflict or cannot establish reality.
AI proposes. Humans ratify.
Machines can keep the model alive without silently becoming the authority on what the organisation believes.
01
Observe
02
Research
03
Recommend
04
Draft
05
Propose
06
Act
Recourse when coordination fails.
Conflict should be surfaced and governed, not hidden, silently resolved or turned into competing action.
01
Surface the conflict
02
Preserve competing evidence
03
Escalate for judgment
Humans and machines should not inhabit different realities.
Orient is where organisational reality is formed, contested, ratified and governed—before machines act. People and agents use different interfaces, but inhabit the same world.
Humans see the workspace
Add, think, create and inspect the organisation’s understanding.
Add
Think
Create
Memory
Activity
Situations
Impact
The shared world
One maintained organisational reality sits between every interface.
Understanding
Evidence
Uncertainty
History
Authority
Machines see the model
Reason from structured state with provenance and explicit boundaries.
Provenance
Expectations
Situations
Change
Authority
History
THE MISSING LAYER IN AI
What world is the AI reasoning from?
A larger context window, giant RAG store, Slack transcript, MCP connection or shared memory can all provide material. None, by itself, forms and governs organisational reality.
Orient provides shared reality.
