Audit agents
They allow you to record a situation at a particular moment (for example, how many people visit your website ?). As the interpretation of the results can only be carried out by people, the agent can only present facts as far as they are measurable.

Preparation agents
Working on certain established programmes, these agents can put forward recommendations and influence management decisions. As defining the work is the phase where human intelligence is most needed, for the moment the agents can only intervene marginally if the problems are complex. On the other hand, for simple tasks, agents can prove to be extremely efficient (in cases where the phase of analysis and consideration is easily standardised).

Production/execution/operation agents
The agent takes over the execution of a list of predefined tasks, to replace a human operator. The programmes which run automated production lines in car factories are this type of agent. They can carry out linear, parallel or recurrent tasks. The agent can be programmed to change or halt the list of tasks according to external events.

Control agents
The majority of the previous phase's results can be measured by certain variables. These could apply, for example, to an online advertising campaign : how many advertising panels were displayed, how many mouse clicks, how many sales took place, the average amount spent on each order, etc ...

In order to set up a system of agents to carry out a particular task, these four steps (audit, preparation, execution, control) must be identified and standardised (try to break down each step into a series of very simple actions linked together). Next, you must analyse the feasibility and cost of developing a specially adapted agent.

At present, however, agents can only be used for a limited number of tasks. A few crucial points should be taken into account when trying to evaluate whether an agent could be useful.

The idea of vital tasks should be considered : how serious would be the result of the agent system's breakdown ? The more vital the task, the higher should be the level of dependability. Agents belong to a group of programmes which are particularly susceptible to viruses, so a careful evaluation of the risks involved is essential.

The system must be able to adapt to cope with the a rising volume of work, anticipate new needs, change over to a new, more powerful system if necessary (in other words, avoid being limited to one type of technology which is incompatible with other agents' systems).

According to the tasks required of an agent, the result can vary considerably : the higher the amount of analysis required, the more essential a human presence is, the more difficult it will be to use an agent.

next