ataraskov.dev

About everything and nothing


Fail spectacularly with AI

What can be juicier than reflecting on a failure? Recently, I caused a production downtime.

The incident started small. A performance issue on a small project. CPU usage was high, and mongod was among the top consumers. Easy, right? Just ask your favorite LLM-based tool to guide you through the troubleshooting process. And follow its advice…

I have a DBA background (spent a bit of my time in RDBMS world), though I have little to no experience with MongoDB. So I asked AI how to investigate CPU usage. And its answer was totally logical and expected. “Index addition looks good enough”, I thought. Though the application was not ‘happy’ about that change - not because it’s badly engineered or unstable. The root cause is my overconfidence, fueled by AI’s guidance.

Lessons learned:

  • Team communication is critical.
  • AI tools are a great help within your professional domain.
  • A junior with a fancy toy tool, is still a junior.

AI-agents are a great tool, which amplifies one’s productivity. Though it also works as a magnifying glass for one’s shortcomings as well ;)

Sun Jun 8, 2025 / 180 words / Learn Failure