skip to content
<- back to blog

What AI Engineering Actually Means

8 min read
software engineering llms ai-driven engineering harness-driven engineering

In my last post I talked about software engineers splitting into two camps, depending on how they use agentic coding tools. Those who passively allow the agent to control the work and those who actively take charge, steering their agent through tasks. This post is about the natural progression into ‘AI Engineering’ that comes from that active stance.

When I first heard the job title ‘AI Engineer’ a couple years ago, I assumed it was just a rebrand of ‘ML Engineer’. Some kind of new Machine Learning specialist cooking advanced mathematics and academic papers in a big vat all day. As a Full-Stack Engineer, I figured it was out of my wheelhouse. Turns out I was dead wrong and the difference matters more than I expected.

Last year I picked up Chip Huyen’s ‘AI Engineering: Building Applications with Foundation Models’. It’s popular enough that there’s a speed-run summary of it on YouTube if you want the highlights first. The first few pages alone totally demystified the term for me. AI Engineers are just specialized SWEs that build applications on top of AI Models. That’s it. And it isn’t even a new field. Applications have integrated with AI systems for years. What’s changed is how profoundly more capable and accessible foundation models have become, and how much market demand has followed.

  • 💡 ML Engineer: Builds foundation models from scratch. They train them with huge data sets and fine-tune them to improve their behaviors. They invent novel ways to advance how they work at a fundamental level. They work at tech companies like OpenAI, Anthropic, Google and others. Commanding substantial salaries and requiring a substantial education.
  • 💡 AI Engineer: Connects applications to those foundation models. Builds the integrations, the products, the user-facing features. Incorporates methods for evaluating and optimizing the output their app recieves from the models. They monitor and secure those integrations. They work on any modern software project in nearly any domain that can benefit from AI enabled features and workflows.

So in the current era an AI Engineer is building solutions to integrate LLMs with web, desktop, and mobile apps in order to provide new exciting features. This is the same kind of split that produced ‘Full-Stack Engineer’ as a category in the 2010s. When the underlying infrastructure improved with higher internet speeds, HTML5, AJAX, and more advanced JavaScript features - Web Developers weren’t replaced. Instead, new job roles and responsibilities were added to match the new paradigm. Foundation models are doing something similar now.

Three Main Types of ‘AI Engineer’ Role

Lately curiosity has found me exploring open job postings labelled with some form of AI Engineer in the title. I figured that ‘reverse-engineering’ job requirement listings could provide some direction into how I can start studying up and training to one day fill them.

What I’m slowly finding is that since the space is so relatively new, there isn’t a definitive set of requirements for this job role. That makes sense considering that 20 years into modern web development we still see that the responsibilities defined for ‘Programmer’, ‘Software Engineer’, ‘React Developer’, ‘Full-Stack Engineer’ and ‘Backend Engineer’ are often just a mish-mash of the same main skills at their core.

Eventually I expect we’ll end up seeing traditional ‘Software Engineer’ roles simply take on ‘AI Engineer’ job requirements as AI features become a more common expectation. If that’s true, then getting familiar with what’s being expected for ‘AI Engineer’ job roles might help us prepare for what will be expected for most Software Engineers eventually.

I took a sampling of 20 recent “AI Engineer”-titled postings from late May 2026. These were jobs with above $140k salary that I’d personally target in a job search had I already been fluent with AI skills. So keep in mind, there’s some personal bias here.

The mix spanned frontier AI labs like Anthropic and Databricks; enterprise SaaS like Grafana, Pinterest, and Route; regulated industries like UnitedHealth and Mutual of Omaha; defense companies like Vantor, Defense Unicorns, and Peraton; and a handful of YC-backed startups.

From this sampling I’m seeing 3 main variations in what ‘AI Engineering’ actually means on job postings. One that focuses on the application layer, another that leans more into infrastructure, and a third that concentrates more heavily on machine learning skills.

AI Application Engineer

What they do: Build user-facing AI features end-to-end. A large portion of the work seems to revolve around RAG, Agents, pipelines, evaluation harnesses, and production ownership. They ship things customers actually use and interact with.

Background: Full-Stack and Backend Engineers transitioning and specializing. Experience with API design, deployment, and product thinking are common throughout.

Examples: At startups this description might include every developer on the team. At larger more established companies you might be an embedded engineer on a product team adding AI features.

Salary range: From $130k - $250k+ depending on the company and location.

AI Platform/Infrastructure Engineer

What they do: Build the layer other AI engineers work on top of. These are LLM gateways, evaluation infrastructure, prompt versioning systems, model obvservability, cost routing.

Background: Infrastructure, platform engineering, or devops. Existing strengths in distributed systems, observability, and reliability are common throughout.

Examples: At scale-ups, this is the dedicated AI platform team. At smaller companies, it’s often the senior engineer who built the internal tooling the rest of the team uses. Some Platform work provides internal business automation rather than developer tooling, but the engineering pattern is the same.

Salary range: Similar to AI Application Engineer, sometimes higher at infra-heavy companies.

ML/AI Hybrid Engineer

What they do: Traditional machine learning work with PyTorch, scikit-learn, and model training; plus generative AI integration and infrastructure work included.

Background: ML engineers, data scientists, or engineers with formal ML training. Healthcare, finance, geospatial, and ad-tech companies lean this direction in their postings.

Examples: This is a step deeper beyond the other two types of roles. You’re building and tuning internal AI models on top of integrating the LLM infrastructure tooling that interacts with them.

Salary range: Similar to the other two types of postings with large outliers in the $300k+ range where requirements are particularly ML heavy.

Between the 20 job postings I pulled, the split came out to roughly 11 Application, 6 Platform, and 4 ML Hybrid roles. Application layer roles were the most common shape by a wide margin… but again, this may be due to my selection bias.

Some application postings involved a sub-pattern of “Forward Deployed” customer-embedded roles. I’ve seen multiple articles remarking that this is one of the fastest growing positions in the space right now. Definitely worth looking more into, the pay is certainly above average. Although despite the fact that the job posting requirements I saw in my study were more “Application Engineer” flavored, that may be a red-herring. It’s reported to be the case that the FDE job role is more commonly on the Platform side, or further into ‘Solutions Architect’ category, recently.

The 6 Platform postings I saw clustered around the LLMOps stack. Gateways, evaluation infrastructure, and orchestration. There was a variant pattern for ‘internal-automation’ platform work that serves non-engineering teams as well.

The Hybrid postings stretched the widest in compensation and background, from enterprise GenAI lead roles in the low six figures to frontier-lab research engineering well past $300K. Since there were only 4 postings in my study batch, I spent some time reviewing additional postings like this. For many of them it felt like a “Catch All” job role where you’re expected to handle everything AI related - App development, platform work, and having a strong understanding of machine learning practices.

Why I’m targeting AI Application Engineer roles

Of those three main archetypes, AI Application Engineer is the closest fit to what I already do. My background is mostly full-stack and backend over the past 8 years, heaviest in TypeScript and Node, with some light Python experience. The central focus is the same - building services, building products, owning them in production. Adding AI integration to that profile is a natural extension.

Platform Engineering is interesting and I’m not closing the door on that variation of job requirements. Some of the side projects I’m working on are in-line with this category of work. It’s just not my primary positioning right now.

The Machine Learning Hybrid category would require a substantial detour into classical ML. PyTorch fundamentals, model training, the advanced math underneath. That’s a large enough gap that it wouldn’t make sense to dive into now. Instead, shipping some Application-level projects first will help give me the grounding I’d need to move further in this direction.

What’s Next

Throughout the last few weeks I’ve found a lot of great resources about ‘AI Engineering’ - roadmaps, bootcamps, guides, tutorials, and plenty more. I’ll be continuing to dive in, study up, and post about topics here on my site as I learn more.

It’s dangerous to go alone! Take these:
[<- all posts]