The problem
An assistant can summarise a case, draft a proposal or retrieve a decision in seconds. Then the session closes, and everything is gone. Next time, the context, the choices already made, the people and the constraints have to be explained again.
This is not a comfort feature. It is what separates a tool you consult from a colleague that improves. And when several assistants work together, the problem grows: the one who knows something cannot pass it on.
Why it is hard
Over a month, it looks like this:
- the same context repeated, session after session;
- answers that contradict each other because the facts were forgotten;
- no trace of decisions, so no review is possible;
- and, as soon as several teams use it, the risk of mixing their information.
What we are building
Durable memory, built to be queried: by meaning, by words, and by the links between facts. Every element keeps its source, its date and its confidence level, so a response can be explained.
Assistants can also talk to each other, share what they just learned and coordinate, without one team’s memory crossing into another’s. What must be forgotten really is forgotten.
What changes
- a new assistant is operational in minutes, not weeks;
- decisions survive people and sessions;
- each workspace only sees what belongs to it.
What comes next
This is being built at Grantity R&D, the lab. We are working with a few pilot teams to test memory on real cases, not on demonstrations.
Do you live with this problem? Let’s talk.