The CaseFlows Manifesto

AI is developing at a pace our organisations and public institutions can barely keep up with. Intelligence is becoming more powerful and more widely usable. At the same time, the means to develop it, offer it and steer it risk becoming concentrated in a small number of companies and countries. The choices we make now determine how dependent people and organisations will be later.

We believe the social value of AI only comes into its own when people and organisations can use intelligence independently and together. Access alone is not enough. A say, freedom of choice and the ability to work together have to be widely available too.

CaseFlows is our contribution to that larger task. We are not building a solution to every social question about AI. We are developing a way to organise work between autonomous parties, without them having to give up control or responsibility.

High-quality AI should become widely accessible and affordable, also for smaller organisations, communities and citizens. Open-source and open-weight models can widen that access. Free choice remains essential: no model provider should become a mandatory gatekeeper.

How you choose where your AI comes from: Control in an AI-powered organisation.

Intelligence gains value when people and organisations can use it together. Working towards shared results should not require everyone to use the same systems or hand control to one platform. We believe in shared cases and explicit case roles, with room for everyone’s own way of working.

How that works: Working across organisations.

Participants keep a say over their own data, processes, systems and technology choices. A collaboration should make its participants stronger, not more dependent.

What that means for your own systems: Your back office, your choice.

AI may analyse, propose and act within a granted mandate, but never on its own authority as a socially responsible party. Because AI is not always predictable or reliable, responsible use must not depend on a model behaving well: its powers are limited and can be revoked, what it does can be traced, and stepping in must be possible. Responsibility for decisions and their consequences remains traceable to people and organisations; how it is divided between the parties involved must be agreed explicitly. The systems in which AI works must enable them to actually carry that responsibility.

How mandates and limits are set: Control in an AI-powered organisation.

Not only the behaviour of AI, but the goals themselves deserve attention. Different parties legitimately have different interests. Good collaboration makes goals, limits and trade-offs visible, without forcing agreement. Feedback helps find out whether local optimisation really contributes to the shared result. Alignment is therefore not a one-off property of a model but ongoing work: making goals explicit, testing results and adjusting course.

Open specifications, interchangeable platforms, transferable data and freedom in choosing AI models prevent new dependencies. Openness has to be demonstrable in practice. The right to leave is as important as the right to take part.

What you can take with you today: Open specifications, portable processes.

Every case is experience. Experiences and feedback are sources of collective intelligence: together they become knowledge that can be tested and improved. Participants should be able to share knowledge and benefit from the insights that come from it, with respect for confidentiality, consent and their own interests.

How experience flows back into the way of working: An organisation that learns.

These beliefs have consequences for how CaseFlows is developed. We choose open specifications, freedom of choice in AI models and providers, and portability of data and processes. We want independent parties to be able to implement the specifications, and platforms eventually to work together directly.

We are clear about where things stand today: CaseFlows is currently the only provider of this particular model of working together. Independent implementations and full federated interoperability are development goals, not existing facilities. We want to make progress on them visible and testable.

We accept that customers must be able to leave CaseFlows without losing the work and knowledge they have built up. Our added value should show in what we make possible, not in the dependency we create.

We design for controllable autonomy. An AI agent gets only the powers its case role needs, in the phase the case is in. We do not assume that AI always acts correctly: the platform takes errors, unexpected behaviour and misuse into account. A caseflow limits what an agent can do, but it does not replace technical security, model safety or independent oversight.

What we claim above should be checkable. This is where each commitment stands today; we update this list when something changes.

• Open specifications: today. How you work is written down in open files, JSON for the structure and markdown for the reasoning, that you can read without CaseFlows and take with you.
• Your cases and data: on request, today. When you leave, we hand over your cases and data in an open format. An export you can run yourself will be available soon.
• Choice of AI model and provider: today, with one limit. CaseFlows runs on your own AI keys, and you choose the provider. Agents that use tools still work through one provider only.
• Shared cases between organisations: today. Organisations work together in one case through case roles they agree on; a partner can hold a role without a platform of its own.
• Powers per case role, enforced by the server: today. What a role may see and do in a phase is checked on every request, also for an AI agent.
• Agents only with explicit access: today. An agent reads or changes only data for which that access has been laid down explicitly; every refusal is recorded.
• Withdrawing an agent’s access: today, through the API. A withdrawn token no longer works from the next request; a button for it on screen is still to come.
• Stopping an agent in the middle of a task: not yet.
• A published meaning of the specification, independent of our runtime: not yet. Today that meaning is defined by our schemas and the CaseFlows runtime; publishing it on its own is our direction.
• Independent implementations: not yet. CaseFlows is the only runtime that executes these specifications.
• Platforms of different providers working together directly: not yet. This is a development goal.

Where each layer of portability stands is described in more detail in Open specifications, portable processes.

An open ecosystem does not come from a single supplier. We invite organisations, developers, implementation partners and service providers to think along, share experiences and contribute to open collaboration. Not to own collaboration, but to make it possible.

You can take part as an implementation partner, as an ecosystem partner, or as an organisation that wants to start with one shared case.