An AI-native engineer builds applications using AI tools as a core part of their workflow. An AI engineer works at the model layer: evaluation, fine-tuning, RAG pipelines, and inference serving. The overlap is real but limited.
The core distinction
The confusion is understandable. Both roles involve AI, and both are relatively new to job descriptions. The difference lies in where the work happens.
An AI-native engineer operates at the application layer. They use models as components, the same way an earlier generation of engineers used databases or message queues. They ship production software. Their AI fluency shows in how they build, not in what they build about.
An AI engineer operates at the model layer. Their job is to make models work correctly for a given problem: designing evals, building RAG pipelines, fine-tuning, managing vector databases, and handling the infrastructure that serves model outputs at scale.
What each role actually does day to day
AI-native engineer: - Writes and ships production features using AI-assisted coding - Reviews AI-generated code critically before it merges - Integrates third-party models via APIs without necessarily touching model internals - Uses agentic coding tools and coding agents as part of their standard workflow
AI engineer: - Designs evaluation frameworks to measure model output quality - Builds retrieval pipelines and manages embedding strategies - Fine-tunes or prompts models to meet domain-specific requirements - Decides when a model is the wrong tool entirely and advises accordingly
Where they overlap
Both need to understand context windows, prompt behaviour, and the failure modes of model outputs. An AI-native engineer who integrates a RAG system into a product needs to understand roughly how it works. An AI engineer building internal tooling will benefit from software engineering discipline.
The overlap becomes a problem when companies conflate the two roles and hire one expecting the other. A strong AI-native engineer is not automatically qualified to design fine-tuning runs. A strong AI engineer is not automatically a productive application developer.
When you need which
If your problem is shipping software faster, with better tooling, and with AI capabilities integrated into the product, you need AI-native engineers.
If your problem is model quality, evaluation, retrieval accuracy, or building the infrastructure that models run on, you need AI engineers.
Many teams need both, but at different ratios and at different stages. Early-stage products typically want AI-native engineers who can move quickly across the stack. As AI features mature and model behaviour becomes a product risk in itself, dedicated AI engineering becomes worth the investment.
A note on terminology
Neither term is standardised across the industry. Some companies use "AI engineer" to mean what others call "ML engineer" or "LLM engineer." Job posts are inconsistent. When evaluating a candidate or a placement, the title matters less than a clear account of which layer they work in and what they have shipped.
What we test for
Miyagami runs separate vetting criteria for the two roles, though both sit within the same two-session process described at how we vet. AI-native engineers are assessed across both assessments on criteria including tool orchestration, AI output verification, and judgement by risk. AI engineers are assessed on those same criteria and additionally on evaluation design and knowing when a model is the wrong tool. The distinction matters at the vetting stage because the two roles call for different habits, not just different knowledge.
Short answers
Can one person be both an AI-native engineer and an AI engineer?
Some individuals work credibly at both layers, but it is uncommon at senior level. The application and model layers each require depth. Most hiring decisions are clearer when the two roles are treated as distinct, even if there is some overlap in understanding.
Is an AI engineer the same as an ML engineer?
Roughly similar, but not identical. ML engineer is an older term associated with training pipelines and classical model development. AI engineer is more commonly used now to describe roles centred on large language models, evaluation, and retrieval. Usage varies by company.
Do AI-native engineers need to understand model internals?
Not deeply. They need enough understanding to integrate models reliably, interpret their failure modes, and make sensible architectural decisions. They do not typically need to design training runs or build evaluation frameworks from scratch.
