Hi ,

If you've spent any time on LinkedIn lately, you've probably run into two camps: the ones who think AI will make developers obsolete, and the ones who think these tools are just glorified autocomplete. Neither camp spends much time on the actual mechanism.

I wrote a piece that tries to fix that.

"Under the Hood: What Your AI Coding Assistant Actually Is" starts with a real bug I ran into while working with Quarkus, Jakarta Data, and Hibernate. When I asked a coding agent to help diagnose it, the output was fluent, confident, and almost entirely useless. Not because the tool is bad, but because I had stepped outside the regime where these models are reliable. Understanding why that happens changes how you use them.

The post covers:

- How transformers actually work (the parts that matter for everyday use, no PhD required)

- Why training on human-written text explains both the strengths and the failure modes

- Why confidence in the output tells you nothing about correctness

- Where these tools genuinely earn their keep, and where they will lead you wrong

This is written for developers who want a working mental model, not a marketing pitch and not a research paper. If you have colleagues who are either over-relying on AI assistance or dismissing it entirely, it might be worth passing along.

Read it here: https://empatheticdeveloper.wordpress.com/2026/03/23/under-the-hood-what-your-ai-coding-assistant-actually-is/

As always, I'm curious what your experience has been. Hit reply if anything resonates or if you disagree.

Markus