The gap nobody's talking about between "I can prompt GPT" and "I can ship AI systems"
I've spent the last few years watching engineers try to break into AI/ML, and there's a pattern I keep seeing.
Most people think the barrier is math. It's not. Most working software engineers can handle the math fine if they sit with it for a weekend.
The real barrier is that almost nobody teaches the engineering of AI systems — the part between "I fine-tuned a model in a notebook" and "I shipped a reliable AI feature that doesn't fall over in production."
Things like:
Why your RAG pipeline works in the demo and falls apart with real user queries
How to actually evaluate an LLM feature instead of eyeballing outputs and hoping
What "agentic" actually means once you strip the hype — and when it's the wrong tool for the job
How to structure prompts and context windows like you'd structure any other system input, with versioning and testing
The unglamorous stuff: cost control, latency budgets, fallback logic, when to fine-tune vs. when to just write better retrieval
If you're a software engineer who already knows how to build software but keeps bouncing off AI/ML because every course either goes too academic (linear algebra proofs) or too shallow (prompt engineering listicles) — that gap in the middle is exactly where I focus.
Curious how others here are approaching this. Are you self-teaching, doing side projects, or leaning on your team's ML engineers to learn by osmosis?