How can an AI role persist when the model changes?
Separate the role, responsibilities and authority from the model used for any single execution.
04 / Research
From static assistants
to developmental AI systems.
Afluma’s long-term R&D explores how bounded AI systems can retain evidence, adapt workflows, coordinate specialised roles and improve without turning autonomy into an ungoverned black box.
Research questions
The hard part is not
making AI talk.
The hard part is building systems that know what they are responsible for, what evidence they can trust, when they should ask for help and how their behaviour can improve safely.
What should a digital coworker be allowed to remember?
Useful continuity needs evidence lineage, scoped access and deliberate promotion into reusable knowledge.
When has a system earned more authority?
Start with bounded tasks, observable acceptance criteria and a recovery path. Expand only after reliable evidence.
Applied research
Research has to meet
a real operating constraint.
Afluma uses its own products and internal workflows as test environments so architecture decisions can be challenged by practical tasks rather than judged only by demos.
Afluma runs on Afluma
Research, product work, operations and customer journeys become controlled settings for evaluating role behaviour, handoffs, memory and governance.
Measure before claiming.
A capability becomes proof only when the task, acceptance criteria, reviewer and outcome can be traced. A polished demo is not enough.
Afluma / Pilot