How To Use Your Agent · AI Potential

Overview

What You Have Received

We deliver your agent as a zip file containing a small number of HTML files. HTML is the same format used by every web page on the internet. We use it because it is easy for you to read and also very effective for AI to read.

Each file has a specific job:

00 Start Here An overview of what the agent does and how to get going.

Process The step by step instructions the AI follows when it runs.

Context Background information about you, your role, and what matters to you.

Golden Example A completed sample of what the output should look like. AI produces much better results when it can see a great example.

Quality The standards the AI checks itself against before presenting its results to you.

You can double click on any of these files and they will open in your web browser, just like a normal web page. Everything is fully transparent. There is nothing hidden or technical. Have a look through them so you understand what the agent knows and what it is going to do.

Getting Started

Setting Up

Unzip your files. Your agent arrives as a single zip file. Right click on it and choose Extract All (on Windows) or double click it (on Mac). This will give you a folder containing all of the HTML files listed above.

Open your AI tool. This works in ChatGPT, Microsoft Copilot, or Claude. We are completely platform agnostic, so use whichever one you are most comfortable with. For the rest of this guide, we will use ChatGPT as the example, but the steps are almost identical in the others.

Upload the agent files. Select all of the HTML files in your folder (not the zip, the individual files you extracted). Then simply drag and drop them into the ChatGPT conversation window. You will see them appear as uploaded attachments. In ChatGPT, the send button changes colour once the files have finished uploading, so wait for that before moving on.

Running It

Give It Data and Go

Give it your data. If your agent needs data to work with (a spreadsheet, a report, a set of numbers), drag and drop that file into the same conversation as well. The agent will use it alongside the instructions you have already uploaded. You can provide data in whatever format you have it. Excel spreadsheets, CSVs, PDFs, or even just pasted text all work well.

Tell it to run. Type a simple instruction. Something like the example below. You do not need to be precise. You do not need to use any special language. Just tell it what you want in plain English.

"Read through these agent files. I have also given you a spreadsheet with data. Analyse that data and give me the final deliverable from the agent."

Review the output. The agent will take about a minute to do its work. When it is finished, it will present the final deliverable. In most cases, you can click to download it as an HTML file, then open it in your browser to see the finished result. Because the output is a web page, it is also very easy to share with other people.

You do not need to be precise. AI is very good at understanding what you mean, even if your instruction feels a bit rough around the edges. The important thing is to give it a nudge in the right direction. The agent files handle the rest.

Making It Yours

Changing the Output

This is one of the most important things to understand: you should be active in the process. Do not just accept whatever the AI gives you first time. If something is not quite right, say so.

Here are some examples of things you can ask:

"That is not our colour scheme. We use blues and oranges. Please update it."

"I do not want the data in a grid. What are my other options?" The AI will suggest alternatives like charts, gauges, or heat maps, and you pick the one you prefer.

"The tone is too informal. Make it board level."

"I like that section but the summary at the top needs more detail."

You can go back and forth as many times as you need. Each round only takes a minute or so. If you can describe what you want in English, that is all you need.

Pro Tip

Getting Better Results

A useful technique is to ask the AI to pause before it builds anything. For example:

"Before you do anything, create a plan and present that to me. Also give me your understanding of what I have asked for, just so I know we are aligned."

This forces the AI to show you its thinking before it starts building. You can correct it early, which means higher quality output and fewer rounds of revision. The AI may also ask you clarifying questions, which is a good sign. Answer them and it will produce something much closer to what you need.

Repeat Runs

Making It Easier Next Time

You do not need to upload the files every single time. Here are two ways to make repeat runs faster:

◆

Rename the thread. After your first run, give the conversation a name (like "Ops Pulse Report"). Next time, just come back to that thread and drop in your new data. The agent files are already loaded.

◆

Use Projects. In ChatGPT, you can create a Project and upload the agent files once. Every conversation you start inside that project will have access to those files automatically. This is the cleanest way to run an agent repeatedly.

Maintenance

Updating Your Agent

Everything in your agent is written in plain English. There is no code. If you want to change how it works, you just edit the relevant file and upload it again.

◆

Want the output to look different? Edit the golden example file, or describe the changes you want and ask the AI to update it for you.

◆

Want it to know something new about you or your role? Edit the context file.

◆

Want to change the process it follows? Edit the process file.

You can even do this inside the conversation. Just tell the AI what you want changed and ask it to update the golden example with that version instead. Then save the file it gives you back and use it next time.

How To Use Your AI Agent