The “AI engineer” title covers three very different roles. Learn which one your healthtech startup needs, what to pay and how to interview for it.
The Fastest-Growing Job Title Nobody Agrees On
AI engineer is the hottest title in tech. AI engineer took the top spot on LinkedIn’s annual Jobs on the Rise list, which ranks the fastest-growing US roles over the past three years.
Demand is enormous. LinkedIn added 639,000 AI-related US job postings between 2023 and 2025, including 75,000 for AI engineer roles. LinkedIn’s head of economics for the Americas, Kory Kantenga, summed it up: “Companies are just gorging on AI talent.”
Here’s the problem. The title means different things to different companies.
One company hires an “AI engineer” to train and validate machine learning models. Another wants someone to build an agent on top of a foundation model. A third wants a software engineer who uses AI coding tools well.
Same title. Three different jobs. Different skills, different backgrounds, different salaries.
When the title doesn’t match the job, everything downstream breaks. The wrong candidates apply. Interviews test the wrong skills. Offers miss the market. And the person you hire struggles in a role they were never suited for.
This article covers:
- The three jobs hiding behind the “AI engineer” title
- How to tell which one your startup needs
- Why healthtech raises the stakes
- How to write the spec and run the interview
Where the Title Came From
The term gained traction through Shawn Wang, known as swyx, who wrote the influential essay “The Rise of the AI Engineer.” He describes the role as “the software engineer building with AI.”
He also drew a useful line between AI engineers and traditional ML engineers. In his framing, the ML engineer’s job stops when they serve inference, while the AI engineer’s job begins with inference.
Swyx describes three types of AI engineer: a software engineer enhanced by AI coding tools, a software engineer building AI products, and eventually fully autonomous AI coding agents.
For hiring in healthtech today, a slightly different split is more useful. Here are the three jobs we see behind the title.
Job One: The Machine Learning Engineer
What they do: Build, train, fine-tune and validate models. They work with data pipelines, feature engineering, model evaluation and MLOps. They decide whether a model is accurate enough to trust, and keep it accurate in production.
Typical background: ML engineering, data science, research engineering or applied science. Often a graduate degree, but production experience matters more.
Signals on a CV:
- Models trained and deployed into production, with measurable results
- Experience with model monitoring, drift detection and retraining
- Work on data quality, labelling and evaluation datasets
- Experience validating models on new populations or settings
When you need this person: When your product’s value depends on a model you own. Risk prediction, diagnostic support, imaging analysis, clinical NLP trained on your own data.
Healthtech watch-out: Validation is everything. Models successful in one setting or time period may not apply to others, as Michigan researchers found when testing a widely used sepsis model. You need someone who tests on local data and understands why it matters.
Job Two: The AI Product Engineer
What they do: Build products on top of foundation models. Retrieval systems, agents, prompt pipelines, evaluation frameworks, guardrails and integrations. They rarely train models. They make existing models useful, reliable and safe inside a product.
Typical background: Software engineering, often full-stack or backend, with deep hands-on work using LLM APIs. LinkedIn found the top roles people move into AI engineering from are software engineer, data scientist and full stack engineer.
Signals on a CV:
- Shipped LLM-powered features used by real customers
- Built evaluation systems to test model outputs at scale
- Designed retrieval, agent or tool-use architectures
- Handled cost, latency and reliability trade-offs in production
When you need this person: When you’re building AI features on top of models from major labs. Ambient documentation, prior authorisation automation, patient communication agents, revenue cycle tools.
Healthtech watch-out: Governance matters as much as capability. Protected health information in prompts, audit trails for agent actions and access controls all fall on this engineer’s desk. Ask how they’ve handled sensitive data before, not only what they’ve built.
Job Three: The AI-Assisted Software Engineer
What they do: Build regular software, faster, using AI coding tools. They don’t build AI products. They use AI to ship better products.
Typical background: Strong software engineering. Comfortable with modern AI coding assistants as part of daily work.
Signals on a CV:
- Strong core engineering fundamentals
- Evidence of shipping at high pace
- Thoughtful use of AI tools, including knowing when not to trust them
When you need this person: Almost always. This is fast becoming the baseline for every software engineer. It shouldn’t carry an “AI engineer” title or an AI salary premium.
Healthtech watch-out: AI-generated code still needs review, especially around security and data handling. Strong engineers in this category know where AI tools get things wrong.
Why the Wrong Title Costs You
Mislabelling the role creates three expensive problems.
You attract the wrong candidates. Post “AI engineer” when you need an ML engineer, and you’ll get a flood of product engineers who’ve built chatbots. Post it when you need an AI product engineer, and research-focused candidates will apply for a role they’ll find dull.
You pay the wrong price. AI salaries carry a premium, and the premium keeps growing. Carta found median salaries for AI and machine learning engineers rose by 5.4% to 9.1% between January 2024 and June 2025, depending on startup size. Pay an AI premium for job three and you’ve overpaid. Offer a standard salary for job one and you’ll lose every serious candidate.
You test the wrong skills. An ML engineer interview tests model evaluation and data. An AI product engineer interview tests system design, evaluation frameworks and product sense. Run the wrong loop and you’ll reject strong people and hire weak ones.
There’s another trap. The title is new, so experience is short. LinkedIn data shows AI engineers have a median of 3.7 years of experience. Expecting “10 years of AI engineering” on a CV filters out almost everyone, including the people you want.
How to Work Out Which Job You Need
Answer three questions before writing the spec.
1. Do you own the model, or use someone else’s?
- Own it, train it or fine-tune it on your data: job one
- Build on foundation models through APIs: job two
- Neither, you’re building standard software: job three
2. What breaks if the AI is wrong?
- A clinical decision or patient outcome: prioritise validation and evaluation depth
- A workflow or admin task: prioritise reliability, monitoring and user trust
- Nothing AI-specific: you don’t need an AI specialist
3. What does success look like in 90 days?
- “A validated model running in one pilot clinic” points to job one
- “An agent handling prior authorisation requests in production” points to job two
- “The new patient portal shipped” points to job three
If your answers point to more than one job, you likely need more than one hire. Trying to find all three in one person is the fastest route to a search lasting six months.
How to Write the Spec and Run the Interview
Use a specific title. “Machine Learning Engineer, Clinical Models” or “AI Product Engineer, Agents” tells candidates exactly what the job is. Specific titles attract better-matched applicants.
Lead with the problem. Describe what the hire will build and why it matters. Strong AI candidates choose roles based on the problem, not the stack.
Match the interview to the job.
For an ML engineer, ask:
- “How would you validate this model before deploying it to a new hospital?”
- “Walk me through a time a model degraded in production. How did you notice?”
For an AI product engineer, ask:
- “How do you evaluate whether an LLM feature is working, beyond spot checks?”
- “How would you stop an agent from accessing patient data it doesn’t need?”
For an AI-assisted software engineer, ask:
- “Show me how you use AI tools in your workflow. Where do you not trust them?”
- “Tell me about a bug an AI tool introduced and how you caught it.”
Probe for production, not demos. Plenty of candidates have built impressive prototypes. Far fewer have run AI systems in production with real users, real costs and real failures. Ask what broke, how they found out and what they changed.
Hire for the Job, Not the Title
“AI engineer” is the fastest-growing title in tech. It’s also one of the least precise.
Behind it sit three different jobs:
- The ML engineer, who builds and validates models
- The AI product engineer, who builds products on foundation models
- The AI-assisted software engineer, who ships software faster with AI tools
Each needs different skills, a different interview and a different salary. In healthtech, each also carries different risks around validation, data and governance.
Define the job before you advertise it. Your search will move faster, your offers will land and your hire will succeed.
Hiring AI engineers for a healthtech product? Synaxia Group places senior software, product and AI engineers into healthtech, medtech and biotech startups. We help you define which AI role you need, set the right level and comp, and find engineers with production experience in regulated environments. Get in touch with Synaxia Group to scope your next AI hire.