Part 1, Understanding AI agents
Part 3, Introducing an agent
Principle 2
Context
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.
What information and instructions does the agent need, and who can resolve gaps or ambiguities?
What would an experienced colleague notice that the written instructions leave out?
Provide the information and guidance the agent needs to do the work. Arrange for a colleague to answer questions that the available material does not resolve.
- What to decide
- The permitted sources, what they cannot show and who resolves questions.
- Key point
- Explain relevant unwritten practices and exceptions in the brief.
Why it matters#
What the procedure and the records leave out#
Procedures and records may not contain all the knowledge that a task requires. An experienced colleague can:
That knowledge helps them decide what to do next. An agent can draw on its training and the information available to it, but it may not recognise an unwritten practice or understand why it matters.
Explain relevant practices and exceptions in the Brief The instructions given to the agent, including the purpose, permitted information and actions, checks and arrangements for seeking advice. you give it. Some cases will still need a person’s judgement, such as an unusual request. State which of these cases the agent must pass on, who will respond and what work can continue in the meantime.
Background#
From philosophy
Knowledge that is hard to put into words#
The philosopher Michael Polanyi argued that “we can know more than we can tell”. He called this tacit knowledge. An experienced colleague can recognise an unusual case without being able to list the signs that alerted them.
An agent can draw on its training and the information available to it, but it may not recognise an unwritten practice or understand why it matters. Principle 2 therefore asks you to work through real cases with experienced colleagues, to write down what you can and to arrange for a colleague to answer what cannot be written down.
Source Polanyi, M. (1966), The Tacit Dimension.
Information you can supply#
Personal and confidential information#
Do not give an agent personal or confidential information, or connect it to systems that hold such information, unless your organisation has approved that use. Your organisation remains responsible for that information, including when an outside AI service processes it. Approval therefore needs to cover both the AI service and the legal requirements that apply to the use.
This includes information the agent can reach through its connections, as well as what you give it directly. Check what each connection makes accessible.
Information about an identifiable person, such as a customer, colleague or service user.
Data protection law applies to it.
Information that you have agreed, or are obliged, to protect.
For example, client files, contracts and commercially sensitive material.
If you are unsure whether information can be used, do not use it until the person responsible for it has confirmed this use. This section is not legal advice. Its examples are drawn from UK law at the time of writing. The law can change, and the rules may differ in the EU and elsewhere. Ask the person who advises your organisation which rules apply to you. In the UK, the data protection regulator is the Information Commission (ICO). It publishes guidance on AI and data protection.
What approval involves#
Approval is a decision for those responsible for data protection and information security in your organisation, not for an individual user. Before approving any use, they need to know how the AI service stores, uses, shares and deletes the information, and whether the provider uses it to train its models.
Personal information brings further legal obligations. Information collected for one purpose cannot simply be used for another, and people must be told how their information is used. A use that is likely to create a high risk to people’s rights and freedoms needs a data protection impact assessment before it begins. The person responsible for data protection decides whether these conditions are met.
Source ICO, Guidance on AI and data protection, under review following the Data (Use and Access) Act 2025.
In practice#
4 steps
01Use only approved informationUse only information that your organisation has approved for this AI service.
List the information the task needs. Use only information that your organisation has approved for this AI service, and supply no more than the task requires. Use an account that your organisation provides or has approved, not a personal one.
Do not enter, upload or give an agent access to personal or confidential information unless your organisation has explicitly approved its use with that AI service for that purpose. If you cannot confirm that approval, do not proceed.
02Examine a case that required judgementAsk someone to show you a recent case that initially looked routine.
Ask someone to show you a recent case that initially looked routine. Working through it can reveal knowledge or checks that a written description of the process does not include.
Questions to ask:
What did they notice?
Why did it matter?
What did they check before acting?
What should the brief say?
Turn the answers into instructions.
03Improve the instructionsUse what you learn, and note who supplied each explanation.
Use what you learn to improve the instructions. For example:
If two policies appear to conflict, clarify which applies before the agent proceeds.If a record covers only part of the service, explain what it leaves out.Include source dates so the agent and the reviewer can distinguish current guidance from an earlier version.RecordsWhat is known, and whenInterpretationMeaning and exceptionsAuthorityWho can clarify or decide↓↓↓A shared briefSource links · assumptions · questions · agreed changesInformation and explanations kept together in the brief Add each explanation to the brief, noting who supplied it and whether it applies to future work.
04Allow time for questionsRepeated questions can increase a colleague’s workload.
Repeated questions can increase a colleague’s workload even when the agent works quickly.
Allow time to answer them
Group those that can wait
Pause work that depends on an unresolved answer
See also · 1.2 Context: what the agent can use · 1.2 Memory
Examples#
The opening hours example
How a colleague changes the analysis#
A local service is open from 9 am to 5 pm on weekdays. Some service users have asked for evening appointments, and the team must decide whether longer hours would improve access. An agent examines twelve months of request records and compares the costs of the options. See chapter 1.5.
Uses the number of recorded requests for evening appointments as an estimate of demand.
Explains that the phone line closes at 5 pm, so the records cannot show who might have called later.
Have heard requests for evening appointments, but these have not been counted consistently.
Revises its comparison to distinguish recorded demand from what remains unknown.
Decides what further evidence to gather before recommending an option.
The initial analysis treated few recorded requests as evidence of low demand. The service lead explained why records collected during opening hours could not establish demand after closing.
Illustrative example
Risks#
Check the interpretation before acting#
An agent can produce a plausible interpretation from incomplete information.
Exercise#
Check what the instructions leave unexplained#
Choose a case that required an experienced colleague to depart from the usual process. Identify what could have been explained in advance and what required advice from a colleague.
Write one instruction that would help an agent recognise when to ask.
List the permitted sources and who authorised their use. Note what each source does not show. Name who can answer questions, who keeps the guidance current and which questions must be resolved before the work continues.
Further reading · AI and meaning · clowryai.com/thinking
© 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.

