AI workflows
AI workflows that stay governed: where AI helps, and where rules decide
In Miji, AI does two jobs: it drafts workflows from a plain-language description, and it runs individual judgment steps inside them — summarising notes, classifying a request, drafting a message for a person to approve. It does not make the policy decision. Rules and decision tables decide, calculations are exact, people approve what the policy says they must, and each AI step's output lands on the same audit trail as everything else.
That is the difference between an AI workflow and an AI agent: the path through the work is defined and reviewable, and AI is used where judgment helps rather than everywhere.
In short
- AI drafts the workflow; you review it before anything runs.
- AI steps use the model and key you choose, configured per tenant.
- The policy decision stays deterministic: rules, tables and exact maths.
- Human approval gates wherever the policy requires a person.
- Every step, AI or not, is recorded on a hash-chained audit trail.
Two roles for AI in Miji
- Authoring. Describe an operation and Miji's AI planner drafts a runnable workflow — steps, branches, approvals and connections. You review and edit it; nothing runs until you deploy it.
- Judgment steps. An AI step sits inside a deterministic workflow. It receives defined inputs, runs with your model and key, and its output is stored on the run, where the next steps — and your auditors — can see it.
Why not an AI agent?
An AI agent decides its own next step. That is useful for open-ended work, and a problem for decisions that must follow a policy and be explained afterwards: the path can differ between identical cases, calculations can be approximate, and reconstructing why something happened means reading a model's reasoning.
Miji keeps the path fixed and the policy explicit, and uses AI as a component inside it.
| Autonomous AI agent | Miji AI workflow | |
|---|---|---|
| Who decides the next step | The model, at run time | The workflow you reviewed and deployed |
| Policy decisions | Model judgment | Versioned rules and decision tables |
| Calculations | Often done by the model | Exact decimal arithmetic |
| Human oversight | Added around the agent | Approval gates inside the workflow |
| Explaining a decision | Reconstruct from model output | Rule version, inputs and branch recorded with the decision |
| Same case twice | May take a different path | Takes the same path |
Governance controls
- Your model, your key. AI steps run with the model and API key configured for your tenant, so there is no lock-in to one AI vendor.
- Approval gates. A person or group signs off wherever the policy requires, resolved by any one, all, or a majority.
- Role-based access and SSO. Control who may author workflows, approve decisions or only read; sign in with OIDC (Okta, Entra ID, Keycloak).
- Test and production. Separate environments, with production deployments pinned to versions.
- Audit trail. Every step, including AI steps, is recorded on a hash-chained trail you can verify on demand.
- Your infrastructure. Run managed, or self-hosted on PostgreSQL in your own cloud account.
Where AI helps in practice
- Summarising an adjuster's or processor's notes before a rule decides the route.
- Classifying inbound requests so each reaches the right workflow or queue.
- Drafting a message or summary that a person approves before it is sent.
- Drafting a new workflow from a description of the process, for a person to review.
In each case the output is an input to the policy or to a person — not the decision itself.
On the roadmap
Not available today. Listed so you can see where this is going — the capabilities above are the ones you can use now.
AI usage registerBuilding
An automatic inventory of every AI action in a case's lifecycle, as part of the compliance pack.
Goal-first operationNext
An AI manager that takes an objective, decomposes it into work and allocates it — with the same governance. The long-term direction.
See where AI helps in one of your workflows — and where the rules decide.
Book a demoFrequently asked questions
What platforms combine AI, business rules and workflow automation?
Decision automation platforms such as Miji combine the three: AI drafts workflows and handles judgment steps, versioned business rules and decision tables make the policy decision, and the workflow engine runs the steps, routes exceptions to people and records everything.
What are the best AI workflow automation platforms for enterprises?
It depends on the work. General workflow tools suit moving data between applications; agent frameworks suit open-ended tasks. For decisions that must follow a policy and be auditable, decision automation platforms such as Miji place AI steps inside governed workflows with versioned rules, human approval and a hash-chained audit trail.
Does Miji let AI approve loans or claims?
No. In Miji the approval decision comes from your rules and, where the policy requires, a person. AI steps can prepare inputs — a summary, a classification — and their output is recorded, but they do not make the policy decision.
Which AI models does Miji use?
The ones you choose. AI steps run with the model and API key configured for your tenant, so you can change vendors without rebuilding workflows.
Can a workflow run without AI at all?
Yes. AI steps are optional. A workflow built only from rules, calculations, connectors and approvals runs fully deterministically.