AI Assisted debugging

[JuanSombrero23]

The idea

CLI supports pretty much all TE3 functionalities except the debugger. That makes total sense...on the other hand, as a big fan of the debugger, I have developed my own 'debugging workflow' nowadays...I build a pivot table that returns the exact same numbers as my Excel or PBI, so I can right click any value, go to debugger, and dive into the DAX to see where things go wrong. I still wanna be the driver of the debug proces, because I learn alot about DAX by doing so, on the other hand, my debugging flow is pretty structures, step 1, step 2, do this, do that,...perfectly suited for AI automation, eg a skill or whatever. So I was thinking, can you guys think of a creative approach, or hack if you want, to have my agent 'mimick my approach, do what the debugger does', with the current TE3 and te CLI capabilities. I probably could explain Claude what steps to follow, but then he would be freewheeling, basically re-building the debugger workflow from scratch, that is smth I want to avoid...let be explain myself a bit with an example:

Compare it to, let's say, editing the semantic model...there as well, I could explain the agent to open the TMDL, alter stuff, do checks, then publish...thats what I mean by 're-building'. On the other hand, the CLI took away alot of those requirements, making AI integration for semantic model editing so much easier.

That's the kind of hack I am looking for now, specifically for the debugger approach. So way where I can tell the agent to debug a spacific number, have him report his step by step findings to me (like I do when 'stepping into' and 'stepping over' in the debugger), based on my preferred way of debugging, helping me understand my own DAX, but having the technical flow facilitated by the TE technology and functionalities.

Not sure if I express myself clearly here guys, push back if smth is unclear.

Regards,
Jan

Why does it matter to you?

No response

[greggyb]

What would you want to feed to an agent harness from a hypothetical debug sub-command? The debugging interface is pretty heavily GUI driven and interactive. What would you imagine coming back as text from the CLI that the agent would then read and act upon?