How to Run and Create AI Agents

AI Agents Guide for Clients

A quick reference for using agentOS AI Agents in your agency

This guide walks you through how to find and run AI Agents, how to build your own automations, and a few things worth knowing along the way.

1. Open the AI Store from your agency dashboard. You can browse agents by category, and it helps to read an agent's description before you run it.
2. Try one of the example agents below. Choosing “Try” runs the master version exactly as it is, so there is nothing to set up first.
3. If you would like to adjust an agent for your own settings, you can install it first. See the note on this under “Good to Know” further down.
4. To build something from scratch, go to Settings and then Automation Builder.
5. Once you are happy with an agent, schedule it, either as a single run or on a recurring basis, right from its chat panel.
6. To check on a scheduled agent, go to Scheduled, find the item, and select “View Run” to see the results.

It's worth having one person in your agency own and manage AI Agents, and keep the rest of the team in the loop on what is being used. When trying a new agent for the first time, it's a good idea to start with one that doesn't change your data, so you can see how it behaves first.

These are tested and safe to run.

1. Write your prompt in plain language. The more specific you are up front, the less tweaking you will need later.
2. Generate it, then review the logic it generates in the ‘Automation Prompt’ field, including the controller, filters and fields, before saving. This is where any missing filters usually get caught.
3. Check the ‘Automation Prompt’ and “Output Hints” fields and add any fields you would like to see, even if you didn't mention them in your original prompt.
4. Give it a clear name, assign a category, and save.

If the results come back wrong, simply refine your wording and try again. This is normal and not a sign that something is broken. You're welcome to use Claude or ChatGPT to help tighten your prompt wording before you start. It also helps to refresh the page or clear the chat before starting a new build or schedule, so it doesn't pick up details from a previous request.

• Keep prompts simple, specific and to the point.
• Refresh the page, or clear the chat, between runs so requests don't get mixed up.
• Never trust the output blindly. Always cross check before any action that sends emails or changes data, since the agent will follow instructions exactly, including any that are based on incorrect information.
• Nominate one person, such as a manager, to champion and review how AI Agents are used in your agency, rather than everyone building freely. This also helps manage overall usage.

• The AI Agent is currently in the beta stage and is not yet available for clients to create their own agents. Please contact us if you require any bespoke automation to be built.
• At the moment, there isn't a way to keep an agent private while you're building it. Anyone else at your company can see and run it, so if you're testing something, it helps to label it clearly, for example “TEST do not use”.
• Results can vary slightly from one run to the next, depending on the data available at the time.
• Categories are assigned automatically and may occasionally need a manual correction.
• Adjusting an agent after installing it (Configure) doesn't always apply your changes when you run it. We're aware of this and working on it, so for now, building a fresh automation is a safe way to get the settings you want.

If you have any questions while using AI Agents, please reach out to your agentOS contact.