For integrators and teams deploying AI agents
DUBSAR is developing a governance foundation to keep mission state, tie actions to explicit authorisations, and retrieve execution evidence.
The aim: keep the work understandable when a workflow stops, an authorisation changes, or another team takes over.
Product in development.
An illustrative teaching diagram. This is not a product screenshot, and not an available connector.
Find the objective, the progress, the decisions and the points still open in a mission history that is separate from the conversation with the model.
The team taking over then has retained elements to understand the situation and continue the work.
Tie the actions being handled to an explicit authorisation scope and, when necessary, to a human decision.
Teams can see what is allowed, what is waiting for validation, and what must be reviewed again when rights change.
Identify what was requested, what was authorised, and the executions confirmed by the available evidence.
Keep observed failures and still-uncertain results visible, so an incident can be investigated and the next step decided.
These situations illustrate the needs DUBSAR aims to address, depending on the systems and actions involved. The CRM and ERP examples are not a catalogue of available connectors.
The consultant leaves the project. The client team needs to recover the state of the missions, the decisions already made, and the actions still pending.
What is sought: a usable history to organise the handover and continue the work.
A change in a CRM that was previously allowed now requires human validation.
The point: take this new frame into account before continuing, and keep the decision linked to the action concerned.
A request to create an order is sent to an ERP, then the workflow stops before confirmation.
The team must check the result before deciding what comes next. The aim: avoid a resumption that implicitly restarts an operation whose outcome is still uncertain.
An action produces an unexpected result. The team tries to match the initial request, the authorisation, any validation, and the available execution elements.
It can then separate established facts from what still needs to be checked.
Models provide the reasoning. The software carries the durable state, the authorisations, and the evidence.
Define the expected result, the permitted actions, and the human validations required.
Link the work steps, the authorisations, the human decisions, and the execution elements in a durable mission state.
Use that state to hand over, investigate, or resume, taking failures and still-uncertain results into account.
Each scope needs to define the systems involved, the actions to govern and the results to verify.
Targeted components and journeys have already been tested. Some trials covered real exchanges, others a simulated provider.
Full B2B integration and validation are still underway. The next step is to verify an end-to-end workflow with a defined scope and observable results. Pilots are not currently open.
Handover, human approval and uncertain outcomes: the DUBSAR guides help you define the decisions and records your workflows need.
