The short answer.
Using an AI assistant at work has six levels: asking it questions; giving it your files and context; setting standing instructions it keeps; running reusable workflows the whole team shares; connecting those workflows to your business systems and running them on a schedule; and having several assistants work as a team. Most people with ChatGPT, Claude, Copilot or Gemini are on the first or second, only scratching the surface. Each level up takes a little more setup and hands the assistant more of the work, and the higher levels are easier to learn than advanced spreadsheet work, because you write them in plain language.
These levels describe one person. The stages of AI at work follows a whole company from switched on to managed. The makers rename and move these features often; the product details below were read on each maker’s own pages in November 2026.
Level one: asking it questions.
At level one, you use the assistant like a well-read colleague: you ask, it answers, and it knows nothing about your business.
- On a Tuesday.You ask how to word a difficult reply to a customer.
- The advantage.Nothing to set up.
- The trade-off.Every answer is general, so you rewrite it to fit your business.
- To move up.Next time, give the assistant what it would need to know.
Level two: giving it your files and context.
At level two, you give the assistant your own material before you ask, so the answer is about your business and not business in general.
- On a Tuesday.You upload last month’s sales export and ask which accounts went quiet.
- The advantage.Answers that use your numbers and your words.
- The trade-off.You supply the context every time, and each person does it a different way. Check your company allows a file in the assistant first.
- To move up.Notice what you paste in every time. That is your first standing instruction.
Level three: standing instructions and projects.
At level three, the assistant keeps your instructions and reference files between conversations, so you set them up once. ChatGPT, for example, has Projects: a place that keeps related chats, files and instructions together, where the instructions apply only inside that project and the project can be shared with teammates. The other three assistants have their own ways of keeping standing instructions and files.
- On a Tuesday.The office manager opens the supplier queries project, which already holds the price list, the returns policy and an instruction to answer in the company’s tone.
- The advantage.No setup each time, and answers that stay consistent from one day to the next.
- The trade-off.Instructions go stale when the price list changes, and a private project helps one person.
- To move up.Write down the steps of the work, not only the background.
Level four: reusable workflows the whole team runs.
At level four, one piece of work is written down as a workflow: the inputs it starts from, the steps, what good looks like and who checks the result. Everyone in the role runs it the same way. Microsoft 365 Copilot, for example, has Agent Builder, which builds what Microsoft calls agents, each with its own instructions and knowledge sources such as files on SharePoint, which you test and then share. Microsoft notes that the SharePoint files behind a shared agent may need permissions granted separately.
- On a Tuesday.Each of four account coordinators runs the same weekly client update workflow, and the account lead reads every update before it goes out.
- The advantage.The best person’s method becomes everyone’s method.
- The trade-off.It needs one owner, by name, who tests it, keeps it current and decides when it changes.
- To move up.Once the workflow is trusted, look at where its inputs come from.
Level five: workflows connected to your systems and run on a schedule.
At level five, a workflow reads from the business systems your company already runs and starts at a set time. Gemini, for example, lets you schedule a recurring request that is ready at the time you choose, and a schedule that uses another app, such as Google Workspace, needs that app connected first.
- On a Tuesday.A summary of overdue invoices from your accounting system is waiting for the finance lead at the start of the day, and the finance lead decides who to call.
- The advantage.The work is ready before anyone asks for it.
- The trade-off.A connection is a decision about access: what the assistant can read, and what it may change. Approve each connection on purpose, have a person approve any change it drafts, and name who notices when a scheduled run fails.
How AI workflows connect to the systems you already run sets out the steps from an exported file to a live connection.
Level six: several assistants working as a team.
At level six, one assistant splits a large job, hands each part to another assistant with its own instructions, and puts the results together. Today this lives mostly in developer and coding tools, not in the everyday chat product. Claude Code, Anthropic’s coding tool, for example, has subagents: specialised assistants, each with its own instructions, its own permitted tools and its own working memory, that return a summary to the main conversation.
- On a Tuesday.A developer asks for a release review: one assistant checks the code, another the tests, and the main one reports back.
- The advantage.Large jobs run in parallel, and each part gets focused attention.
- The trade-off.Anthropic says it plainly: each subagent makes its own requests and adds to usage, one that starts fresh takes time to gather what it needs, and work with a lot of back-and-forth is better kept in one conversation.
Assistants working as a team do best on work already written down as workflows, as AI agents or workflows: which first? explains.
How a person and a team move up a level.
A person moves up one level at a time, by noticing what they repeat: the question becomes a file, the file a standing instruction, the instruction a workflow.
Start from one piece of work you did last Tuesday and will do again next Tuesday, and test each step up on last month’s real cases.
A team moves up when the person who has climbed shares the method and one person owns it. Many teams have one person at level three or four and everyone else at level one. Share that person’s method as a workflow, and the whole team moves up together. How to hand a piece of work to AI covers the handover in detail.
Why the higher levels are easier than advanced spreadsheet work.
The gap between a basic and an advanced AI user looks like the gap between basic and advanced Excel, and it is easier to close. Advanced spreadsheet work means formula syntax, pivot tables and macros, where one wrong reference gives a wrong number with no warning. A higher level of AI use is written in plain language: instructions, steps and what good looks like. The assistant helps you write them and fixes them when you say what went wrong. The hard part is knowing the work well, and you already do.
What to check before you move work up a level.
Copy these questions and answer them for one piece of work.
- Does the work repeat the same way?Weekly or monthly, from the same kind of input.
- Can you list the inputs?The files, records or emails the work starts from.
- Who checks the result?One name. The assistant prepares; that person decides.
- Who owns the instructions?One name, who keeps them current.
- What will the assistant see or change?Which files, which systems, and whether a person approves every change.
- Where does the feature live?In the everyday chat product or in a developer tool, and whether your company’s account has it switched on. Ask whoever runs it.
- What happens when it is wrong?Who notices, and where misses are recorded.
Where FullyPowered AI fits.
Helping companies level up what they get from the AI they already pay for is what we do: we plan it with you, develop the workflows and deploy them to your teams. The plan lives in FullyPowered Planner, and How organizations plan their use of AI sets out the method.