Power BI with AI agents 101
Want to be ready for where Power BI is going?
You've probably heard that AI agents are changing how Power BI reports get built. Join me for a morning to take your first steps in agentic development. You get an agent to explain a prepared report you've never seen and write notes on its calculations. Then you have it build one new calculation, and you check its answers yourself.
By the end, you can judge where Power BI is going from your own hands-on use, not from what you've read about it.
If you've built a report in Power BI Desktop, even a simple one, you're ready. No coding or AI experience needed. If you already use coding agents on Power BI work every day, this covers ground you know.
What you'll do
One prepared report runs through the whole morning.
Get your bearings.
You see what an agent is, how a Power BI report's files fit together and what to check before giving a tool access to work files. Then you point your agent (GitHub Copilot, Claude or ChatGPT/Codex) at the teaching report.
Understand a report you didn't build.
You ask the agent to explain the report: what each page shows, the tables behind it, how they connect, how the data is prepared and what each calculation does. Nothing changes yet.
Document its calculations.
You give the agent a short set of house rules, then have it add comments and descriptions to the existing calculations, so whoever opens the report next can see what each one does. You bring the changes back into Power BI Desktop and check them there.
Build one new calculation and check it.
You describe a new calculation and have the agent build it. Then you check its total, a filtered result and an awkward case. If an answer is wrong, you show the agent the evidence so it can put it right, and you can always go back to a clean copy.
We finish with where this goes next: tools that let an agent check its own work, and agents doing more of the building while you decide what's right.
Read the diagram as text
You ask for a change in the interface, which passes your request to the harness; both run on your computer. The harness sends your request and what it has read to the AI model on a server elsewhere, which replies with the next step, and this repeats. The harness runs the tools the AI model asks for, within the permissions you set, and reads and edits the text files of your Power BI project, which define its pages, visuals, tables, relationships and calculations. Power BI Desktop opens the same files and shows you the changed report, where you check it.
How you'll check its answers
An agent can sound sure of itself and still be wrong. You learn to check its work against numbers you can confirm, correct it with evidence and go back to a clean copy when you need to. Deciding whether an answer is right stays with you.
Who's teaching
I'm Alan Harman-Box. I've spent about five years working with organisations as they adopt Power BI. I've used coding agents on Power BI since before they could write DAX. Now they're better than me on a good day, so it's time to show other people how it's done, including how to check their work.
How I trainEssay: We're all middle management nowWhat you'll need
- A Windows computer with a version of Power BI Desktop released in the 3 months before the course.
- One of three agent tools, on at least its lowest paid plan: Claude Desktop, ChatGPT or GitHub Copilot. Using something else? Tell me, and we can probably make it work.
- Permission to install and use that tool. On a work laptop, check with your IT team early. If your work laptop is locked down, you can use your own Windows computer.
- A folder on your computer for the teaching files, outside OneDrive and SharePoint, plus reliable internet and audio. A second screen helps.
- A short setup check before the day. You'll get the details with your course information once your booking is confirmed.
You work on a prepared report and sample data, not your organisation's own reports.
Afterwards, you keep a video recording of the morning, along with the teaching project and clean copies, so you can work through it again. How recordings work.
Book a seat
, 9:30am to 12:30pm UK time, live online. 4 to 8 people. £195 a seat for this first run, the founding rate. Later runs are £225 a seat. VAT is added at checkout.
Before you book: you'll need a Windows computer, a paid plan on Claude Desktop, ChatGPT or GitHub Copilot, and permission to install it. See what you'll need.
Booking terms
- Cancel up to 14 days before the course, by , for a full refund.
- Can't make it after that? Send a colleague in your place: tell me at least 24 hours before.
- If your setup can't be got working before the day, you can move to a later run, send a colleague or have a full refund.
- If fewer than 4 people have booked by , the course moves once to a date in February. If you can't make the new date, you get a full refund.
- If I have to cancel for any other reason, you get a full refund.
If your organisation needs an invoice or a purchase order, email alan@burningsuit.co.uk and I'll send an invoice instead. Invoices are due within 30 days.