What do assistants need to use AI really well?

My insightAI looks different at every client. One organisation may be experimenting enthusiastically, while another lacks access, clear agreements or time to practise. For me, the question is therefore not only what the technology can do. More important is what assistants need to use it safely, practically and professionally.

A licence is not yet a working method

An organisation can make Copilot or another AI tool available. That alone does not create a better working method. An assistant needs to know which information may be used, where it is stored, what must be checked and who remains accountable.

Executive support also depends on context that is not always recorded: organisational sensitivities, relationships, confidentiality, timing and the history behind a decision. AI provides support; the assistant determines what is appropriate.

We know Copilot and ChatGPT. But do we need more?

Many assistants have now worked with Copilot, ChatGPT or both. We ask a question, improve a piece of writing or request a summary of a document. That is useful. But we often still collect the information ourselves, copy it into the chat and then move the result back into our working environment.

The next step may therefore not be another chatbot. Perhaps what we need most is a secure connection between AI and the tools in which our work already takes place.

Plugins can help provide that connection. OpenAI describes plugins as reusable bundles that can combine instructions with connections to external services. Depending on what an organisation makes available, this may include email, calendars, SharePoint, Slack or Google Drive. Learn more about plugins in ChatGPT.

For an assistant, that could mean no longer having to gather every relevant piece of information manually. A well-designed workflow could consult permitted sources, structure information and prepare a first briefing. The assistant then reviews the context, sensitivities, completeness and next steps.

This does not mean that every organisation immediately needs plugins. A connection only works well when the task, access, accountability and review process are clear in advance. Existing account permissions and approvals still apply; a plugin should not bypass them. OpenAI likewise explains that external connections retain their own authentication and access controls. See how ChatGPT Work uses files, tools and plugins.

1. Access to the right environment

AI can only help with information that is available and accessible. When documents are scattered across mailboxes, local drives, network folders and different collaboration environments, the result remains limited. The same applies when an assistant has a tool but not the appropriate licence or permissions.

Good use therefore starts with clarity: which sources are authoritative, which information may be used and where should that information be stored?

2. Clear and understandable rules

Assistants need more than a general announcement that AI may or may not be used. We need concrete answers. May a meeting document be summarised? May a draft email be improved? Which information is too sensitive? Which tool is approved? And when is human review mandatory?

Clear boundaries do not make people cautious out of fear; they give people confidence in how to act. They prevent everyone from having to guess where the line is.

3. Time to practise with real work

A one-hour training session can inspire, but skill develops in day-to-day practice. Assistants need room to test small applications, compare outcomes and discover where AI genuinely saves time.

It does not need to start on a large scale. Consider structuring your own notes, preparing an initial outline or checking whether a long text covers every agreed topic. Small, safe experiments build experience that relates to real work.

4. Trust in professional judgement

A strong assistant does not simply execute but assesses. Is the information correct? Is context missing? Does the tone suit the recipient? Is this the right time to send something? These questions do not disappear because of AI; they become more important.

Assistants must therefore retain the freedom to adapt a proposal, request further information or decide that a task should not be carried out with AI. Human review is not a final tick in a box. It is part of professional judgement.

5. A seat at the table when the approach is designed

Assistants see every day where information gets stuck, which questions keep returning and where actions risk being lost. That practical knowledge is valuable when choosing and configuring AI tools. Assistants should not be involved only after a system has been introduced, but also when the surrounding working method is designed.

The best application often begins not with the tool that is available, but with the point at which the work is unnecessarily complicated today.

The question we should ask more often

We talk a great deal about what assistants could do with AI. It is just as important to discuss what they need to do it well: secure access, understandable boundaries, time to practise, cooperation with IT and information security, and trust in their professional judgement.

Perhaps that is the best starting point for every AI initiative: do not simply ask assistants to use the tool. Ask what they need to work with it genuinely better.