Write a commentThese days, AI is omnipresent - and I've been working with it for quite some time. When I discovered an article on LinkedIn, I immediately recognized the potential for one of my clients. I adapted the approaches presented, prepared myself specifically for the customer's problem and achieved an amazing result: thanks to AI, I was able to identify the exact problem within about a minute - something that would have taken much longer without AI. In this article, I report on a real-life support case and show how AI helped me to solve a customer problem in the shortest possible time.
What exactly is it about?
The customer had the problem that the print service in his multi-user system kept crashing - and nobody knew why. The event log provided hardly any information, only that splwow64.exe had thrown an exception. All I knew about the process was that a 32-bit printer driver was being used on a 64-bit operating system, which should always be avoided.
As there was no crash dump available for the incident, I first explained to the customer how to create such a dump. The customer then made the necessary changes and waited until the print service crashed again. The crash dump was absolutely essential for me so that I, or rather the AI, could get to the bottom of the problem.
In the meantime, I adapted the following blog post that I had found on LinkedIn - which wasn't so easy if you don't work in that environment all the time.
Prerequisites
- Free installation of Python 3.10 or higher.
- Free installation of Visual Studio Code (version 1.99 or higher).
- Free installation of Debugging Tools for Windows.
- A free GitHub account (if not already available).
- Free installation of Git for Windows.
After the customer contacted me and sent me the process dump of splwow64.exe, I wanted to find out how much the AI can support me. So here are the exact prompts I gave GitHub Copilot (GPT-4.1). I saved the customer's dump to C:\Dumps.
First, the AI needs to know which dump to analyze, so I just let it search for it: Duration approx. 5 seconds
The next step is for the AI to analyze the dump: Duration approx. 15 seconds
Ok, wow!
That's a very precise indication of what caused the exception! But I would like to have it a bit more precise and therefore I would like to see the status of the printer driver:
Duration approx. 15 seconds
This is now very precise and includes a summary and actions that should be taken to resolve the problem.
But now it is often the case that printer drivers are often updated and nobody is aware of this or the driver simply comes from the print server via P2P, so I would like to know the date of the printer driver. So I prompt: show date of driver koax8j*
Duration approx. 10 seconds
But now we have it exactly! The problematic printer driver is for a Konica Minolta printer and was released in April 2021.
The following recommendation for the customer
- Update of the Konica Minolta printer driver, newer than April 2021
- Use of a 64-bit printer driver
- Activation of printer isolation via GPO so that the printer service does not crash in the first place
In addition to the classic dumps, I have also had LiveKD stack traces (these are quite large log files) analyzed via the AI and I am also thrilled with them!
Summary
I have already analyzed many crash dumps with WinDBG and completed training courses with Dmitry Vostokov - this makes "correct" prompting much easier for me. Of course, a certain basic understanding is still necessary, but with AI support it speeds up the analysis enormously. So if you are well prepared for a crash dump, you can solve it in no time at all with the help of AI, or at least achieve a very good result.
