How to Write Instructions for the AI Agents
The instructions on the AI context screen are plain text that the AI agents read exactly as you type it. Each tab has its own: one set for the whole company, one for each call type, customer, engineer and office user. Open the menu, choose AI context under Administration, and type in the Instructions box, then press Save.
How do I write the company's instructions for the AI agents?
The Company tab of the AI context screen opens straight on its editor, headed Company-wide. Type in Instructions and press Save under the editor.
- Open the menu and choose AI context under Administration.
- Type in the Instructions box on the Company tab.
- Add any key terms in Key terms under it.
- Press Save.
The button reads Saving… meanwhile, and then the screen says "Saved. The agents read it the next time AgentDesk asks." The line under the heading says when the company's instructions were last saved, or "Nothing is saved for it yet."
The screen describes the company's instructions as "What every agent reads about the business before it answers anybody: who you are, how you work, and what matters when a call is logged." Things worth writing there are your trading names, the work you do and where, and what you always want asked on a call, such as which site the fault is at. Victoria and Fiona read it when they log a call by voice, and Fill in with AI reads it before anything else.
Cancel puts the box back to what is saved, after asking whether to discard your changes.
How do I write instructions for one call type, customer, engineer or person?
On the AI context screen, open the tab, find the line and press Add, or Open when something is already written; type in Instructions and press Save.
- Choose the Call types, Customers, Engineers or Internal users tab.
- Type part of the name in the search box at the top, or part of a customer's code on the Customers tab.
- Press Add or Open on the line.
- Type in Instructions, add any Key terms, and press Save.
The editor's heading is the name you opened. The line under it says, for example, "Read by the agents whenever" the name "is part of the conversation." For an engineer it says "Nothing reads the engineer layer yet." After Save the screen says "Saved the context for" the name, goes back to the list, and the line shows Has context and the start of what you wrote.
What each tab is for, in the screen's words:
- Call types — "What a call of each type means here, and what to ask for." For example, what to ask about a fault before it is logged.
- Customers — "What to know about one customer: its sites, its people, what it usually rings about."
- Internal users — "What to know about one person in the office."
- Engineers — no AI agent reads it yet, but people's connected AI assistants read each engineer's instructions when they suggest who could do a job.
How long can AI instructions be, and can I format them?
Each set of instructions on the AI context screen can be up to 2,000 characters, about a page. The counter under the Instructions box shows how many you have used out of 2,000, starting at "0/2000", and a screen reader reads it out as it changes.
Text pasted over the limit is kept whole so that you can shorten it yourself, and the counter says how far over it is, for example "2100/2000 — 100 over: shorten it before saving". Save with text still over the limit is refused with "The instructions are" the number of characters "long; the most one layer can hold is 2,000." When you save, spaces and blank lines at the very start and end are left out and do not count.
There is no formatting toolbar. The line under the box says: "Markdown is fine, and headings and lists help the agent find its way, but nothing renders it: the agent reads exactly what is typed here." So you may use simple headings and lists made of ordinary characters, such as a line starting with a dash, and they help the agent; they are not shown as formatting anywhere.
Write the instructions the way you would brief a new member of staff: short, plain sentences about what matters.
How do I change or remove AI instructions I have written?
To change instructions on the AI context screen, open the same tab and line, edit the Instructions box, and press Save: what you save replaces what was there.
To remove them, empty the Instructions box, remove the key terms with the × on each, and press Save. With nothing written, the line goes back to "No context yet" and the agents are given nothing for it. There is no history of earlier versions, so copy anything you may want back before you replace it.
If somebody else saved the same instructions while you were typing, your Save replaces theirs: the last one saved is what the agents read.
If you go on typing while a save is still being made, what you typed afterwards is kept in the box and is not saved yet; the screen says "Saved what was sent. What you typed since is not saved yet." Press Save again.