A conceptual space where humans and agents can develop their ways of working across many tools and technologies
A team can have a conversation in one place, make a decision in another, record the resulting work somewhere else, and return to it a week later. The people involved understand that these are connected parts of the same undertaking. Maintaining those connections across tools takes effort.
Add agents to the team and the need becomes more explicit. An agent needs to understand which conversation concerns which work, whether a suggestion became an agreement, who is responsible, and what it may carry from one context into another.
The Audax Meta-Framework is our developing conceptual account of the collaboration environment we envision. It names the things that matter in collaboration and explores how they relate, so humans and agents can work with a shared understanding of what is happening.
This work is currently conceptual. We are shaping the vocabulary, relationships and principles, and using our practical experiments with agents to test them. The environment described here is a direction we are working towards, with questions still open about its design.
We call it a meta-framework because it provides a way to describe and connect different approaches to collaboration. Teams have different practices, organisational forms and technologies. The framework should help them express those differences within a common conceptual space.
A concept can have many implementations. A space could be a conversation, a shared document, a project board or a place where people meet. A commitment could be represented in a work tracker, a meeting record or a team’s agreed register. A process could span several of those places and include actions by people as well as agents.
One tool can also represent several concepts. A project board might hold proposed work, accepted commitments, responsibility and progress. These have different meanings even when the interface displays them together.
The conceptual questions come first: what is this record, what standing does it have, and how does it relate to the rest of the collaboration? Each team can then decide which tools will represent it.
For example, the following are possible mappings rather than prescribed software choices:
| Concept | Meaning to preserve | Possible representations |
|---|---|---|
| Space | A place of participation with a purpose and boundaries | Chat, document, board, repository, physical meeting place |
| Commitment | An accepted responsibility, attributable to someone | Work item, agreed action record, team register |
| Process | Connected activity working towards a shared purpose | Written practice, workflow system, several tools used together |
| Cycle | A recurring rhythm with phases and opportunities to renew | Calendar, planning board, recurring meetings, review practice |
We call the mapping between a concept and a team’s chosen tool a binding. It gives the agent a checked account of where something belongs and how to work with it. Another team can choose a different binding for the same concept. A team can also change its tools while preserving the meaning of its work and agreements.
This is where agents become especially useful. We envision them helping people act through the concepts of collaboration, while taking care of the corresponding operations in the tools.
A person could ask, “What have we agreed for this cycle?” or “Help us turn this idea into a process.” The agent would use the team’s map to find the relevant records, interpret their standing and bring the work back into the conversation. People would be able to focus on their undertaking without repeatedly translating it into each application’s terminology.
The tooling becomes more transparent in that experience. Its underlying records remain visible: people should be able to see where an answer came from, which agreement it refers to and what the agent changed. Where a binding or an interpretation is uncertain, the agent can bring that uncertainty to the people involved.
Consider an illustrative team developing a new shared practice. The idea starts in a conversation. A document holds the emerging proposal. A meeting brings the people together to discuss it. Some actions become accepted commitments, and a board records their progress.
In the environment we envision, an agent helps preserve the relationships between these things. It can connect an action to the decision behind it, identify which parts of a proposal are still open, and help participants revisit the practice after they have tried it. The team can understand and develop one collaboration across several tools.
The meaning of acceptance matters throughout. A possible action in a conversation can become a proposal. The person involved can accept, revise or decline it. An agent’s extraction of an action is one step in that progression; the person’s agreement is another. The environment should make both understandable.
Our working vocabulary also distinguishes participants from their roles. One person can contribute in several roles, and an agent can serve different purposes in different contexts. Responsibility and permission need to be connected to the relevant role and relationship. They cannot be inferred simply from someone’s presence in a conversation.
Spaces give participation a home, with a purpose and local expectations. A collaboration can span several spaces, and a single space can support several kinds of work. In our emerging vocabulary, a journey describes how a relationship can develop; a flow is a particular relationship moving through that journey, with its own progress and history. Those connections give the environment meaning beyond where its records happen to be stored.
We are exploring Process as a concept that connects multiple flows towards a shared purpose. Preparing a gathering, for example, might involve participant invitations, partner relationships and facilitator preparation. Each relationship can progress through its own journey. The process describes how their progress connects, where handoffs occur and which outcomes depend on one another.
People and agents could develop that process together: describing it, trying it, noticing where the description fails, and proposing changes. The shared semantic space would give them a way to discuss the practice itself, as well as carry out the current work.
Knowledge and memory belong in this environment too. Evidence supports an interpretation; a decision establishes an agreement; history records what happened. A remembered statement needs its origin and context if it is to inform future action. Those distinctions let a team review its own understanding and correct it as it learns.
The scope of knowledge also matters. Personal understanding, shared team knowledge and reusable experience have different owners and audiences. We envision agents helping carry appropriate understanding across contexts, with consent and boundaries remaining part of those connections. Useful learning can travel between teams while the private circumstances that produced it stay with their participants.
Time gives the environment another dimension. There are dates, deadlines, durations and histories. There are also cycles: daily check-ins, sprints, seasonal planning, annual reviews and longer organisational rhythms.
A cycle has phases and a purpose. Its particular occurrence has its own commitments and outcomes. An agent aware of that distinction could help a team prepare for reflection, gather the relevant experience and carry proposed learning into the next cycle. The rhythm can include pauses, rest and a change of direction when people decide it is needed.
Across these concepts, human agency is central. People need ways to understand, challenge and revise how their collaboration is represented. Agents should help make relationships and progress visible, support decisions within their authority, and ask for human judgment where the consequences require it.
The framework is intended to work across different scales of cooperation: an individual’s work, a team, an organisation and relationships between organisations. Its value would come from making those connections easier to understand while allowing each context to retain its purpose and boundaries.
Our current Living Agent work, using Hermes Agent, is one practical way to explore parts of this vision. Its skills, records and learning paths give us concrete experience to learn from. Other runtimes, document systems, work trackers and workflow technologies could represent the same concepts in different ways. Each implementation would need to make its mappings, capabilities and limits explicit.
What we hope people experience is continuity. They can move between conversations and tools, ask about the work in language that makes sense to them, see the relevant agreements and help shape what happens next. Agents help maintain those connections. Humans and agents can then build and improve collaborative processes within an environment whose meaning they can inspect and develop together.
← All posts