Over the last year, I’ve changed jobs to lead an AI team. The name AI team is funny. In a few years, the name “AI team” probably won’t exist. It will simply be software or product teams. But you get my point - our team is building AI agents in a regulated space.
I started my career as an Engineer, founded a couple of startups, ended up in product - now I’m somewhere in between.
I’m interviewing a lot of AI engineers these days, and this post is trying to distil what I’ve learned about how anyone can get into AI engineering.
TLDR is the number of engineers who have built an AI agent and taken it to production (with real users) is very small. If you are one, landing a job will be easier for you. Otherwise, build an AI agent and take it to production.
When I interview people, and I ask them about the AI agents they have built, at least a few people tell me how they are using a coding agent. Not about an agent they have built. Some people talk about RAG or vector DBs. But only a few people talk about an actual agent they built.
They explain what problem it was trying to solve, why they chose an agentic route, how they implemented the memory, how was knowledge structured, what tools where it using etc. Something fun they learned.
Even less can explain how real users were using it, how they did observability, evals, what broke and what they learned. If you are in this category, there are plenty of jobs out there for you. There are a lot of interesting things you learn when real people use an agent. How expensive queries can become, how stale knowledge produces answers that sound very convincing, how conflicting memories lead to confusing answers, etc.
I don’t want to go into the details of the concept you have to learn. I believe learning has become easier these days. Just ask any LLM, and follow along.
Building AI agents is the new full stack. You can start this learning journey by asking any LLM. Pick any fun problem, and build something, take it to production and share it on the internet. In the process of building, sharing and iterating, you will learn things.
A few other things that are helpful in my experience. The lines between product and engineering are blurring. If you consider yourself a builder, you can probably get into AI engineering as a forward-deployed engineer. What you need is the ability to understand customer problems, convert that into a solution, take it back to the users and iterate.
Product-minded engineers have always been a delight to work with. But those kinds of engineers will have a lot of opportunities coming their way, and I’d argue it would probably be the default of engineering. So sharpen your product sense - again, the best way to learn it is to build things and share them on the internet.