- 1: Part 1, Understanding AI agentsPart 1, Understanding AI agents
- 2: Part 2, The six principles
- 3: Part 3, Introducing an agent
- A–E: Appendices
- 1.1: What an agent is
- 1.2: What an agent works with
- 1.3: How an agent carries out work
- 1.4: What agents are used for
- 1.5: The opening hours example
Part 1, Understanding AI agents
Part 3, Introducing an agent
Chapter 1.2
What an agent works with
Scope and limitations of this guide
This guide provides general information on organising work with AI agents, not advice tailored to a particular organisation, system or decision. It is not legal or regulatory advice.
Its examples, checklists and templates do not establish that a proposed use is safe, lawful or suitable. You need to assess and test the proposed use in your own circumstances. Following the guide does not by itself meet the requirements of any law or standard.
The guide is provided in good faith, using information available at the time of publication or revision. AI systems, provider terms and legal requirements can change. Before acting, check the current requirements that apply to your organisation and country, confirm what the system can do and test the proposed use. You are responsible for how you apply the guide and should obtain professional advice where appropriate.
Reading, downloading or using the guide does not by itself create an adviser–client relationship with Tom Clowry or Considered Ideas Ltd.
To the fullest extent permitted by law, Tom Clowry and Considered Ideas Ltd accept no liability for loss or damage arising from use of, or reliance on, this guide. Nothing in this notice excludes or limits liability for death or personal injury caused by negligence, fraud or fraudulent misrepresentation, or any other liability that cannot lawfully be excluded or limited.
In this chapter
This chapter explains the instructions an agent follows, the information it can use, its memory, and the tools and connections through which it acts.
Instructions#
An agent’s instructions can come from the AI service’s own rules, your organisation’s standing instructions, reusable guidance for a type of task, and the brief that describes the particular work delegated to it. How these instructions are prioritised depends on the service and how the agent is configured.
An agent can misinterpret or fail to follow an instruction. Where an error could have serious consequences or be difficult to reverse, restrict what the agent can do or require approval before it acts.
Context: what the agent can use#
A model draws on what it learned during training and the information supplied for the current task. The information supplied at a particular step is known as its context. It includes the instructions, the conversation so far, documents supplied, results returned by tools and notes from earlier work.
There is a limit to how much information a model can process at once. This limit is its context window. In a long task, information supplied at the start may not remain available to the agent. Some services summarise or remove earlier material as the window fills, so that the work can continue, but detail may be lost in the process.
Do not assume the agent has the information needed for this task.
Memory#
Memory allows an agent to use information beyond the current request. It takes three main forms.
Within a task
The information in its context while it works, together with any working notes it keeps, such as a record of work already done. During a long task, some of the context may be summarised or removed.
Between tasks
Information kept from earlier work. Depending on the service, this can include notes and preferences saved by the agent or by a person, and past conversations that the agent can refer back to. The agent can draw on this information as well as its instructions.
In support of tasks
Searching documents or records and adding what appears relevant to its context. This is known as retrieval.
What an agent keeps from earlier tasks can affect later ones, so it should be visible and open to review.
Tools and actions#
A tool is a defined action that an agent can choose to take, such as:
Being able to use a tool does not mean having permission to use it for every purpose.
How connections are made#
Connections allow an agent to retrieve information from other systems and, where permitted, make changes or act on your behalf.
An agent can be connected to another system in four main ways.
Ready-made connections offered by the AI service for widely used systems. You sign in to the other system and choose what to allow.
Many services support the Model Context Protocol (MCP), an open standard that defines how an agent can use external tools and data. Some built-in connectors use it.
A direct connection uses a system’s application programming interface (API) to retrieve information or carry out actions. It usually needs technical help to set up.
Some agents can operate a website or application by reading the screen, clicking and typing. This way of working is at an early stage. Changes to the page, or instructions written on it, can mislead the agent, and it acts with whatever access the account has.
What an agent can do through a connection depends on the permissions granted to that connection. Where possible, the agent should have its own account, with access limited to what the task needs and a record kept of what it does. These arrangements need to be set up or checked by someone with the technical knowledge and authority to do so before the agent is connected.
© 2026 Considered Ideas Ltd. Licensed under the Creative Commons Attribution 4.0 International licence (CC BY 4.0). This guide provides general information and is not legal or regulatory advice. The guide’s home page gives the suggested credit.
Scope and limitations of this guide
This guide provides general information on organising work with AI agents, not advice tailored to a particular organisation, system or decision. It is not legal or regulatory advice.
Its examples, checklists and templates do not establish that a proposed use is safe, lawful or suitable. You need to assess and test the proposed use in your own circumstances. Following the guide does not by itself meet the requirements of any law or standard.
The guide is provided in good faith, using information available at the time of publication or revision. AI systems, provider terms and legal requirements can change. Before acting, check the current requirements that apply to your organisation and country, confirm what the system can do and test the proposed use. You are responsible for how you apply the guide and should obtain professional advice where appropriate.
Reading, downloading or using the guide does not by itself create an adviser–client relationship with Tom Clowry or Considered Ideas Ltd.
To the fullest extent permitted by law, Tom Clowry and Considered Ideas Ltd accept no liability for loss or damage arising from use of, or reliance on, this guide. Nothing in this notice excludes or limits liability for death or personal injury caused by negligence, fraud or fraudulent misrepresentation, or any other liability that cannot lawfully be excluded or limited.

