Why I'm Transitioning from Web Dev to AI Engineering

The Shift I Noticed

In 2024, something changed in client conversations. Projects that used to be pure CRUD apps started having a new question attached: "Can we add AI to this?"

At first I thought it was hype. But the requests kept coming — AI-generated summaries, chatbots, content classification, recommendation engines. The market was moving, and I needed to move with it.

What "AI Engineering" Actually Means

AI Engineering is different from Machine Learning. ML engineers build models. AI engineers integrate existing models into products.

As a web developer, I already had the infrastructure skills:

• Building APIs that serve data
• Managing state in frontend apps
• Deploying and monitoring production systems

What I needed to add:

• Working with LLM APIs (OpenAI, Gemini, Groq)
• Prompt engineering — the craft of getting reliable outputs
• Retrieval-Augmented Generation (RAG) for knowledge-grounded responses
• Understanding token limits, context windows, and costs

The Learning Path I Followed

I didn't take a course. I built things.

Step 1 — Paulo AI Chatbot Built a basic chatbot with the OpenAI API. Learned: streaming responses, conversation history, token counting, and the basics of prompt design.

Step 2 — pamangan.com AI Integration Added AI recipe generation to a real production app. Learned: provider fallback logic, response parsing, cost management, and caching AI outputs.

Step 3 — Now: RAG and Agents Currently studying vector databases and LangChain. The goal is to build an AI assistant that can answer questions about my own portfolio and projects.

The Hard Parts

Prompts are code. A poorly written prompt gives you garbage output. You have to engineer prompts with the same rigor as functions — test them, version them, iterate.

Costs are real. Every API call costs money. Design for caching. Design for fallbacks. Don't call AI when a database query will do.

The outputs are non-deterministic. Unlike a function that returns the same result for the same input, LLMs don't. You need validation layers and graceful error handling.

Where I'm Going

My goal is to become a full-stack AI engineer — someone who can build the web application and integrate the intelligence layer. Not a data scientist, not a pure frontend dev. The engineer who can do both.

The demand for that skill set is only going to grow.