This blog post was first published on Lion’s Share of data
Until now, my usual tools were Visual Studio Code and the GitHub Copilot app. In October, I attended DataMinds Connect and heard a session on agentic development by Kurt Buhler and Greg Baldini from Tabular Editor. Kurt mentioned that his favorite development tool is Claude CLI, so I decided to try a CLI myself. I went with GitHub Copilot CLI.
He also pointed out that you don’t necessarily need skills for agentic development, so I decided to install only pbir.tools.
Installation github copilot cli
Open PowerShell and run winget install Github.copilot.

Then switch to your development directory and start the CLI with the copilot command in a terminal session. You will see the screen below and need to sign in.

I’ll cover the MCP server setup in a later blog post. For my first experiments, I left the model selection on AUTO.
Installation pbri.tools
Maxim Anatsko and Kurt Buhler built a community tool for working with Power BI, and I decided to use it. I installed it from within the CLI using the commands from the documentation.

My First Steps
At the start, I had to practice a bit before I managed to change the directory. In most Windows environments cd is enough, but here it’s /cwd. My first prompt was the following, and it was very simple.

After a while, the agent had created a report.

The report worked and was fully functional, but there were a few visual things I didn’t like.
Iterations and the Final Result
For example, the slicer was tucked away in the middle of the page, which bothered me, so I asked Copilot to move it. I also made a few other optimizations.

When a request needs your confirmation, it looks like this:

Another idea was to add modern tooltips.

I quite like the final visualization after about 5-7 iterations.

pbir.tools implemented the tooltip above as follows:

I’m very surprised by the result. In total, I spent about 6 hours working with the tool, tried out various things, and used $5 in credits. I also never used a model larger than Sonnet 5.5, so the outcome is more encouraging.
My First Takeaways
I’m grateful for Kurt Buhler’s input. Without it, I would never have tried the CLI. It gave me new experiences and ideas, some of which are documented here.
- The combination of pbir.tools and the CLI works and delivers appealing, convincing results.
- Still, for me the CLI remains more of a developer tool for people who enjoy working with a command line interface.
- Until now, I tended to expect a one-shot solution, meaning the perfect result in a single prompt. The iterative process is like how I worked before. The only difference is that I can do something else in parallel.
- I also liked that I could place the report and the CLI side by side.
I will keep experimenting with CLI. For example, can I hand it a mockup, or connect further tools for validation?