AI Engineer Agents¶

On this page you will learn about what is an AI Engineer Agent and how it fits into Kelvin.

What is an AI Engineer Agent?¶

An AI Engineer Agent is a digital assistant inside Kelvin that works for the Operator.

It is given a job to do, it understands the operation it is helping with, and it handles the parts of the day-to-day work the Operator hands over to it.

It is not a chatbot. It is not a model that only answers questions.

An Agent sits between the people running the operation and the automation layers underneath.

It is flexible and can be given a wide range of tasks:

• Watches for specific scenarios and respond to them
• Interprets complex operational situations and determine the next course of action
• Summarizes various data and unstructured information from across the platform
• Acts, when authorized, to carry out actions autonomously

It works through Kelvin, with access to the same live data, documents, and standard operating procedures the Operator works with.

Who AI Engineer Agents work for¶

An AI Engineer Agent works for the Operator. The Operator gives the Agent its job, provides the context it needs, and sets the limits it works within.

Day to day, the Operator decides how closely to work with the Agent. The Operator can review every recommendation, approve actions before they happen, or let the Agent handle agreed tasks on its own and check in on the results.

No machine-learning background is needed. An AI Engineer Agent is set up using the Operator's own knowledge of the asset and the operation, not by writing model code.

How an AI Engineer Agent operates¶

An AI Engineer Agent does not have to act on its own to be useful. The Operator decides how much room the Agent has, and that choice can be different for every task.

There are three modes:

• Agent-in-the-room -- the Agent works alongside the Operator, surfaces what it sees, and offers analysis. The Operator decides what to do.
• Human-in-the-loop -- the Agent prepares the action and the Operator approves it before anything happens.
• Closed loop -- the Agent acts on its own, inside the limits the Operator has set.

The same Agent can operate in different modes for different tasks. Many Operators start with full approval on every action and move to closed loop only on the tasks where they are happy to let the Agent run on its own.

How an AI Engineer Agent is set up¶

An AI Engineer Agent is built from a small set of clear pieces. Each piece answers a simple question about how the Operator wants the Agent to work.

• Objective -- what the Agent is responsible for helping with.
• Goals -- the specific outcomes the Operator wants the Agent to work toward.
• Context -- the information the Agent draws on, such as live data, documents, and standard operating procedures.
• Skills -- what the Agent knows how to do, such as reading a P&ID, an SOP, or an instruction manual.
• Tasks -- the work the Operator hands over for the Agent to carry out.
• Actions -- the changes the Agent is allowed to make on the operation.
• Impact -- the measurable result the Agent helps deliver.

Once these pieces are in place, the Agent is ready to start on the work it has been given.