What is already known?
Earlier agreements about the task, including any corrections.
The starting point
When AI loses track of the work.
Taking over a case means knowing what has already been agreed. Without that background, the search starts again. We ran into this problem when working with AI.
We want AI to draw on what has already been settled when it takes on the next task.
01Find the documents again
02Explain agreements again
03Hand over open questions
AgentBrainBuild on existing knowledge
How we approach work
Saved information needs checking too.
An answer may rely on an old document. We need to know the source and whether its information still holds.
How do we know?
The document or observation behind a statement.
Is it still current?
New information can change an earlier assessment.
What still needs checking?
Missing information and decisions a person needs to make.
The shared foundation
What you need to check a result.
These questions guide our work on Employee and Analytics. They should help you judge a result for yourself.
Agreements
What has already been discussed or corrected?
Sources
Which document supplied the information?
Evidence
What supports the statement?
Responsibility
Who did the work, and who makes the decision?
Recency
When was the information recorded?
Uncertainty
What information is still missing?
Connections
Which earlier decisions are relevant?
Approval
What may the AI do, and when must it ask?
One brand · two products
Where our products can help.
Employee can prepare replies to recurring enquiries. Analytics helps you review the documents for a decision and identify questions that still need answers.
Employee
Recurring work
AgentBrain
Analytics
Decision preparation
Review results
What is the answer based on?
An assessment needs to point back to the documents it used. It should also show what remains unresolved or needs a person to approve it.
Our evidence page explains what we need to document a real deployment.
01Document
02Background
03Review
04Result
05Evidence
A new conversation should pick up where the work left off.
Our starting pointThe founder
Built from practical work with AI.
Theshoth Sritharan founded AgentBrain in Switzerland. Working with AI, he found that starting a new conversation often meant losing earlier agreements.
He wanted the AI to pick up its work again while being able to correct information that was wrong or out of date.
What follows from this
The right information for the task.
A recurring task depends on agreed responsibilities. A new decision may hinge on a recent source.
If a new source contradicts an earlier answer, the AI should take that into account.
We use the task at hand to assess which information helps and where human judgement is still needed.
Continue
Build on questions already resolved.
Keep track
Retain important agreements.
Show sources
Show where a statement comes from.
Assign responsibility
Make responsibilities clear.
Set limits
Agree on tasks and approvals.
Update
Correct outdated information.
From our work
The tasks we are working on.
Analytics in research
Research into antibiotic resistance draws on many documents. Analytics aims to help researchers assess the evidence and unresolved questions.
Employee in property management
Our work with Valtis concerns recurring property management tasks. Sensitive decisions stay with the person responsible.
Your use case
Together we can assess whether this fits your organisation, including the data available and the approvals each task needs.
Our principles
What guides our development.
Earlier answers must remain open to scrutiny.
A statement needs a source.
Each task needs a clear approval process.
Outdated information needs correcting.
People need to be able to check results.
What task do you have in mind? Tell us a little about it.