Skip to content

Agents inside processes, with a mandate and an owner

A credit application waits for someone to gather the documents and check that they are complete. We build agents that do this work in the systems organisations already use, apply the rules of the Company Brain and stop whenever a person's decision is needed.

Agents · What it contains

Catalogue
Agents already set up for common processes, to be adapted to the company's rules
Studio
The environment where we assemble an agent, test it and version it
Reusable procedures
A task written once and used by several agents, such as checking a document
Levels of autonomy
For each agent, how much it does on its own and how much with a person's involvement
Test bench
Real cases on which we test the agent before release and at every change of model
Illustrative interface

What it contains

  • 01

    Catalogue

    Agents already set up for common processes, to be adapted to the company's rules

  • 02

    Studio

    The environment where we assemble an agent, test it and version it

  • 03

    Reusable procedures

    A task written once and used by several agents, such as checking a document

  • 04

    Levels of autonomy

    For each agent, how much it does on its own and how much with a person's involvement

  • 05

    Test bench

    Real cases on which we test the agent before release and at every change of model

The agent sheet

Before it goes into production, the agent has a sheet that the process owner and those who audit can consult at any time.

  1. Mandate

    What it has to do, and on which process

  2. Owner

    The person in the company who answers for it

  3. Data and tools

    Which systems it accesses, with read or write access

  4. Rules

    Which criteria it applies, in which version

  5. Approval point

    Which steps require a person's decision, and whose

  6. When it stops

    A threshold exceeded, a missing piece of data, a new exception: the matter passes to the named person

  7. Measures

    Outcomes, exceptions, cost per run

Three levels of autonomy, earned through testing

An agent arrives well prepared but without knowing the company, and nobody hands a new hire the keys on day one. The move up a level is decided by the process owner, looking at the results on the test bench.

LevelWhat the agent doesWhat the person does
ObserveWhat the agent doesReads, gathers and organises the informationWhat the person doesUses the picture to decide
PrepareWhat the agent doesProduces a complete piece of work, to be checkedWhat the person doesChecks, corrects, approves
Act within limitsWhat the agent doesCarries out defined actions, within set thresholdsWhat the person doesApproves the critical steps and spot-checks the rest

The agent prepares the replenishment, the merchandiser decides

The agent reads each shop's sales and stock levels, forecasts what will sell out and prepares the replenishment proposal, which the merchandiser checks and approves. If an order exceeds the planned budget or concerns an item with no sales history, it stops and hands the decision to the person named on its sheet.

  1. The agent reads sales and stock levels
  2. It forecasts stock-outs and prepares the proposal
  3. The merchandiser checks and approves the order

Illustrative example

Frequently asked questions

Do agents decide on their own?

No, when the choice cannot be undone or reaches the end customer, such as the buying plan for a collection: there the agent proposes and a person decides. On repetitive, low-risk tasks, such as routine replenishment, it acts within thresholds that you set.

Which models do they work with?

With those best suited to the task in terms of accuracy, cost and confidentiality. The model can be changed without rebuilding the agent, which we retest on the test bench at every change.

Can we use agents from other suppliers?

Yes. The Company Brain makes rules and context available through standard protocols, and Governance registers those agents like any others.

What happens if an agent gets an action wrong?

The action is logged and can be reconstructed; operations that cannot be undone go through a person's approval first.

Work out which tasks to entrust to an agent

We start from a process you know well and look together at where an agent would save time and where the decision must stay with a person.