The short answer.
Advanced use of ChatGPT, Claude, Copilot or Gemini is everything past the chat box: workflows saved once and run again, projects that hold standing instructions, connections that let the assistant read from your business systems, and runs that happen on a schedule. Nearly everyone at work can type a prompt, so nearly everyone believes they are using AI, while most people have never opened any of those four features. Your company is paying for the advanced spreadsheet and using it as a calculator.
Companies have put money into AI in two ways.
Companies have spent on AI in two broad ways. The first is building something new with AI: a whole new function, a product feature, a service the company could not offer before. That spend is visible, has a budget line and usually has an owner.
The second is pairing every employee with an AI assistant. Most companies now give their people access to at least one of the big four: ChatGPT, Claude, Copilot or Gemini. That spend is quieter. It arrives as seats on a subscription and rarely has an owner beyond whoever approved the invoice.
The second investment is the one being left on the table. The seats are paid for, and the assistants have moved far past what most people do with them.
Chatting is basic Excel. Advanced use is the modeller’s spreadsheet.
Typing a prompt into an AI assistant is the equivalent of opening Excel and adding up a column. It works, it is useful, and it is where nearly everyone stops.
Everyone has worked next to both kinds of Excel user. One person opens a sheet, types some numbers and uses a sum. Another builds a model with linked tabs, lookups and macros, and the month’s reporting runs from it. Both are “using Excel”. Only one of them has turned the software into a piece of the business that runs the same way every month.
AI assistants have the same split, and it is invisible. Everyone who can chat assumes they know the tool. What most people do not realise is that the same assistant can hold a written-down way of doing a piece of work, keep the context for a whole account or project, read from the systems the company runs and do the work on a timetable.
The good news: the advanced side of an AI assistant is far less complicated than advanced Excel. There are no formulas: you write the instructions in plain language. The hard part is knowing they exist and deciding what to put in them.
What sits past the chat box, product by product.
Past the chat box, all four assistants offer the same four kinds of feature, each under the maker’s own name. The names below were read on each maker’s own help pages on 20 November 2026. The makers rename and move these features often, and several are renaming one right now, so search your assistant’s help for the kind of feature if a name has changed.
- Saved, reusable workflows.Instructions for one piece of work, written once and run again by anyone who has access. OpenAI calls these skills in ChatGPT, bundled into plugins, and is moving its older GPTs across to them. Anthropic calls them skills in Claude. Microsoft calls them agents in Copilot. Google calls them Gems, and its help pages say Gems are becoming skills in Gemini.
- Projects with standing instructions.A space for one client, one account or one recurring piece of work, holding the files and the instructions that apply every time, so nobody starts from nothing. ChatGPT and Claude both call these Projects. Microsoft calls them Copilot Workspaces, formerly notebooks. Gemini calls them notebooks.
- Connections to business systems.A link that lets the assistant read from the tools the company already runs, such as your file storage, your email or your ticket system, and in some cases write back with a person’s approval. ChatGPT calls these apps. Claude and Copilot both call them connectors. Gemini calls them Connected Apps.
- Scheduled runs.A piece of work the assistant does on a timetable, such as every Monday morning or on the first of the month, with no one typing anything. ChatGPT and Claude both call these scheduled tasks, Claude’s inside Claude Cowork. Microsoft has renamed its scheduled prompts to scheduled tasks in Copilot. Gemini calls them schedules.
Put the four together and an assistant prepares a whole piece of recurring work from your own records, the same way every time, before anyone asks. A person still reads the result and decides.
Why the advanced features stay unused.
The advanced features stay unused for reasons that have nothing to do with the software.
- The chat box looks like the whole product.Every assistant opens on an empty box. The features that matter sit one menu away, under names that change.
- Training stopped at prompting.Most introductions to AI teach people to ask a better question. Asking a better question is the basic Excel of AI: useful, and a ceiling. How do we train our team on AI? covers training built around the work each role does instead.
- Each feature needs a decision nobody owns.A saved workflow needs someone to write down how the work is done. A connection needs someone to agree which system the assistant may read. A scheduled run needs someone to own what lands each Monday. An individual employee cannot make those decisions alone, so the features wait.
- Feeling fluent hides the gap.People who chat with an assistant every day feel expert, and in the chat box they are. Nobody notices how far past it there is to go.
The result is a company where AI is everywhere and nothing has changed in how the work runs. Why isn’t AI saving us time? follows that thread into where the time in a business actually sits.
How to recognise the problem in your own company.
You can see where your company stands in one afternoon, by asking the questions below of three or four team leads. Copy the list.
- Saved workflows.Does any team have a piece of its repeating work saved as reusable instructions in the assistant, used by more than one person?
- Projects.Does any client, account or recurring report have its own project with standing instructions, or does each person paste in the background every time?
- Connections.Does the assistant read from any business system you run, or does every input arrive by copy and paste?
- Scheduled runs.Does anything get prepared by the assistant on a timetable, without someone starting it?
- Who knows.Name the people who could set up each of the four. Write “nobody” where nobody can.
- Who decides.Name the person who decides which work goes into these features. Write “nobody” where nobody does.
In most companies the answers are no, no, no, no, one or two enthusiasts, and nobody. That is not a failure of the people. Experimenting was the right way to start, and the chat box is what everyone was given.
Illustrative example.
Take a 45-person veterinary group with three clinics. Every vet, nurse and receptionist has an assistant and uses it daily for client emails and discharge notes, each in their own chat, from scratch. Ask the six questions and the answers are no saved workflows, no projects, no connection to the practice management system and nothing scheduled. One receptionist has built herself a set of instructions for appointment reminders and shared it with nobody. The discharge note every clinic writes, every day, is written three different ways.
Nothing in that picture is a technology problem. The features are in the accounts the group already pays for. What is missing is a decision about which work goes into them, and who writes it down. What is the difference between using AI and deploying it? names that gap.
The way out is a plan.
The way out is a plan: choose the work each team repeats, develop it into workflows that use those four features, and deploy them to the people who do the work. A person decides; a workflow prepares. FullyPowered Planner is where that plan lives: you set out your teams, their roles and the work between them, and choose what to build first. How organizations plan their use of AI sets out the three steps in full.