Authority stays explicit.
An agent’s role is separate from its permission to act. Tools and delegation stay within the organization’s granted scope.
Build an AI-native company.
Or transform the company you already run. Bring together agency, a controlled organisational harness and a knowledge system—one business unit at a time.
Agents own responsibilities, not just tasks. The work becomes a system that can learn and improve.
For invited pilot teams. Setup and qualification happen within each organization.
One organisation.
Three essential primitives.
- Agency
- Agents, workflows and the loops that improve them. Give work an owner, a repeatable process and an outcome to pursue.
- Organisational harness
- A well-formed tool layer over the organisation. Let agents access its systems and take action through secure, controlled capabilities.
- Knowledge system
- Context built from the organisation’s existing information and its interactions—within the company and with the outside world.
Agency uses the harness to act. Knowledge informs the work. What happens becomes evidence for the next decision.
Close the loop around
the KPI that matters.
Work produces evidence. Evaluation finds what to improve. Approved changes return to the work, and their results are measured again.
A support system that learns from each run.
Illustrative walkthrough—not a live workflow or measured customer result. Select a stage to explore it.
Measured outcomes feed the next run
Rollout follows human approval: test two versions, or deploy the approved change directly.
Build a system for the next question, too.
Ask the CEO to build a support-reply system. Its specialists and workflows use the organisational harness to act and the knowledge system to understand the case.
- Responsibility
- Help the customer get a clear, accurate answer.
- What starts a run
- A customer event, a scheduled time or a person’s request, through a configured trigger.
- When a person steps in
- Uncertainty or a protected action needs human input.
Judge the outcome, with evidence.
An evaluation examines the completed work and its recorded outcome. Quality, compliance and business results are different questions.
- Illustrative checks
- Is the reply grounded in the approved facts? Did it respect the customer’s policy?
- Evidence
- The workflow version, reply, escalation and delivery record.
- A KPI to choose
- Resolution quality or time to a useful answer, once a real baseline exists.
Make a proposal, then make a decision.
The evaluator records findings. A separate improvement process proposes a change to an agent’s instructions, its tools or its workflow—within existing permissions.
- Possible change
- Improve the answer source, handoff rule or workflow.
- What stays attached
- The reason for the change and the evidence behind it.
- Who decides
- A person reviews the proposal before the initial rollout. Evaluation does not change a running workflow.
Choose how the change reaches the work.
The rollout follows the organisation’s policy and initial human approval. Compare two versions, or apply an approved change directly.
Compare before expanding.
Run the original and proposed workflows in an approved test scope. Compare the same KPI before deciding what to keep.
These are rollout options in the product vision, not a live experiment on this page.
Measure again. Keep learning.
A rollout is the start of the next observation. Track the chosen outcome against its baseline and feed what you learn into the next run.
- Selected path
- Compare the A/B test’s two versions against the same KPI.
- What counts
- Observed customer outcomes, including regressions and unresolved cases.
- Next decision
- Keep, adjust or revert the change. Then run the loop again.
Start a new company.
Or build on what exists.
The same three primitives support both paths. Build a new organisation around them, or map an existing company’s responsibilities, systems and knowledge into them.
Choose the responsibility.
Start with a focused job such as answering recurring support questions. Define the people, information and decisions it touches.
Connect the three primitives.
Assign agents and workflows, define controlled access to the organisation, and connect the information and interactions that give the work context.
Expand with evidence.
Learn from the work before adding the next responsibility. Support, retention and operations need their own context and measures.
Responsibility needs
clear boundaries.
People remain part of the system.
Requests for information, credentials and protected approvals belong in the right interface, with one recorded response.
Improvement stays inspectable.
Keep the work, evaluation and proposed change connected. A favorable evaluation is a reason to review a change, not permission to deploy it.
An invited pilot.
A longer-term vision.
The first product is being built end to end around organisation workspaces, agent execution, controlled tools and human requests. Access and connections are configured per organisation.
The three primitives and the loop describe the full vision. The pilot does not yet provide the complete knowledge system or every workflow, evaluation and rollout capability shown here. No customer KPI improvement is claimed.
Open workspace
Workspace access is for invited teams.
An organization account is required.