I’ve been working as a backend developer with PHP and MySQL for a while now. Recently got moved to an AI focused team at my company about 8 weeks ago.
Right now I’m building stuff with LangChain and doing some basic RAG implementations with vector databases like Chroma. Mostly trying to make our chatbots give better answers for our specific business needs. I use Python with Flask and FastAPI for the API work.
But honestly it feels like I’m just doing regular web development except now I’m making API calls to OpenAI and Anthropic instead of a normal database. Is this what AI engineering really is?
I want to get into a real AI role at another company in a few months. For people who work full time in AI, what do you actually do each day? Do you work on the same kind of integration stuff or is it completely different?
What skills should I focus on learning? Any good learning resources you’d recommend? I’m really excited about this field but want to make sure I’m headed in the right direction.
your experience sounds pretty typical for ai roles right now. I’m at a startup doing similar work - endless prompt tweaking and fighting with llms that love to hallucinate. the tech stack isn’t the hard part. It’s dealing with unpredictable outputs and explaining to stakeholders why the ai went off the rails again. focus on debugging llm behavior and basic mlops like monitoring model drift.
Yeah, what you’re dealing with is totally normal for AI roles right now. I’ve been doing this for 3 years and honestly, most of my work is exactly what you described - building APIs that hit foundation models and setting up RAG systems. It really depends on where you work. Big tech companies might have you training models or doing research, but most businesses just want you to take existing models and make them work for their specific problems. That means tons of prompt engineering, cleaning up data, and building the same web services you already know. I’d focus on getting solid with ML pipelines. You’ll need to know how to evaluate model performance, manage training data, and handle model versioning - even when you’re mostly hitting APIs. Vector databases and embedding strategies are becoming must-have skills too. Don’t sell yourself short though. Getting RAG working well and pulling good results from LLMs is genuinely valuable stuff that pretty much every company needs right now.
Sounds like you’re already doing real AI work. Most of us aren’t training models from scratch or doing deep learning research daily.
I spend 70% of my time on exactly what you described - integrating APIs, fine-tuning prompts, and keeping systems stable when users throw weird queries at them. The other 30% is data preprocessing and figuring out why our embeddings suddenly return garbage.
The big difference from regular backend work? Everything’s probabilistic now. Your API calls work perfectly 95% of the time, then completely fail on edge cases you never saw coming. You’ll build way more monitoring and fallback logic.
For skills, get comfortable with evaluation metrics. You need to measure if your changes actually improve things. Also learn embeddings beyond just stuffing them in Chroma. Understanding when and how to chunk documents can make or break your RAG performance.
This course helped me early on - covers the practical side:
Don’t worry about not doing “real” AI work. What you’re building is what most companies actually need right now.
