From a Software Engineer to an AI Engineer

Currently, I work as a full-stack software engineer, with a backend focus. My preferred tech stack is .NET Core and TypeScript, and I also do a lot of work in the cloud. The amount of code I've written in the last 8 months has dropped a lot with using GitHub Copilot. It's caused me to have peak interest in the field of AI and where the developers in my industry are headed. AI Engineering is a confusing profession and so many developers don't understand what they actually do. With proper research it's not that confusing. In this post I'll talk about the approach I'm currently taking to transition into AI engineering as a software engineer.

What AI Engineering Actually Is

AI Engineering is building applications on top of foundation models rather than training them from scratch. Essentially this includes:

Building applications using base models with APIs
- e.g. OpenAI, Anthropic
Designing prompts and structuring inputs/outputs
Building RAG pipelines to incorporate external data
Writing evaluation frameworks to measure performance
Not focused on training models from scratch

Although the name doesn't sound like it, it's way closer to backend engineering than traditional machine learning. The focus is on system design, data flow, and reliability.

Why Software Engineers Are In A Good Position

Many software engineers overthink about how they would need to transition to AI. A lot of them take on courses, and even go back to univesrsity for masters and doctorates. Although there's nothing wrong with this, it can be unecessary. In reality, software engineers alreayd have most of the skills required for AI Engineering

Experience working with APIs and backend systems
Familiarity with databases and data flow
Understanding of system design, scalability, and reliability
Experience with cloud platforms

The main gap is learning how LLMs behave, where they fail, and how to design systems around those limitations.

Key Concepts to Learn

You don't need to go deep into model training or advanced mathematics. The prime concepts include understanding tokens and context windows, which determines how much information a model can process at once, and temperature, which controls how deterministic or creative the output is.

It's also important to understand why models hallucinateand how prompt structure can significantly influence output quality. On the data side, concepts like embeddings and semantic search are key to allow for retrieval used within RAG pipelines.

Finally, one of the most overlooked but critical areas is evaluation. AI outputs are fuzzy and things can look right but be wrong. Most people rely on intuition, but in practice, building AI systems requires deliberate ways to measure and improve output quality.

Practical Transition Plan

One thing many AI Engineers agree on is to not spend too much time doing theory upfront, but rather to jump right into building. The guide below serves as a good way to transition

Week 1: Add a small AI feature to an existing project

This can be really small. The purpose of it is to understand the foundations and to be uncomfortable

Examples:
- AI search over blog posts (RAG)
- AI-generated summaries
- Inbox organiser
- Comment moderation
Use an LLM API directly or a framework
Focus on shipping something, not perfection

Weeks 2–4: Fixing problems and gaps

The small feature you'll build will bound to have little problems and gaps when you hit them. You might encounter issues and possible improvements including

RAG Pipeline returning irrelevant results
Knowing if outputs are actually good
An agent going in circles
Getting incorrect or hallucinated outputs
High latency and cost of API usage
Improving prompt structure and consistency

Month 2+: Building something more ambitious

There's a large number of possibiliies to build at this stage. Ideally it should be something multi-step. It should be something you could demo about and write a lot about. At this point, you'd become comfortable with conceptualising how to add AI to existing workflows. Some examples could be adding evaluation methods, improving the quality of existing RAG, or introducing a basic agent

What I'm Currently Building

I'm currently looking to get into AI Engineering. One of the first things that I'm doing is adding some AI features into existing projects. One of them is this blog. The other is my job tracker app. I've been experimenting with RAG based search, and structured outputs from LLMs.

Final Thoughts

The key takeaway from this guide is that the most effective approach is learning by building. Spending a lot of time in the theory phase isn't as beneficial as building practical skills.

Another important takeaway is positioning. You don't need to wait for permission to start working with AI, you can begin by writing about what you're building and contributing to your workplace by adding AI features to existing systems. Lots of companies are actively looking for software engineers who can own and deliver AI features and workflows.

By positioning yourself as someone with AI expertise, you might not even need to change jobs at all. With a background in production software engineering and a few strong AI projects, your company may begin to rely on you and invest more heavily in AI initiatives.