AI engineer vs ML engineer: what the job postings say

Think in layers. Latent Space's June 2023 essay on the AI engineer title argued that ML engineers and researchers handle foundation-model concerns such as pretraining-scale data and general benchmarks, while AI engineers build on top of models and own the data and evaluations specific to their product ¹⁴. The postings mostly draw the line in the same place. The table summarizes what we saw, and the exceptions follow it.

Table 1. Scroll sideways for every column.

Read the rows as tendencies, not rules. The experience figures are what each posting asks for, not what the people hired hold. Two AI-titled postings break the experience pattern: Glean's post-training AI Engineer asks for 5+ years in AI or ML engineering, and Databricks's AI FDE posting asks for extensive hands-on data science experience ^9,12.

Putting one company's AI and ML postings side by side shows the split more clearly than any definition does.

• Stripe. The AI Engineer posting sits in the Solutions Architecture organization and builds internal tools, agents and automated workflows. It asks for 4+ years shipping production systems, strong backend fundamentals and experience deploying agents or LLM tools beyond basic prompt engineering ¹. The ML Engineer postings ask for 3+ years (Growth Platform) or 6+ years (Link) of shipping models, and describe training models, building feature pipelines and, in the Growth Platform posting, recognizing data leakage ^2,11.
• Glean. Both titles carry an evaluation load. The AI Engineer for data and evals builds evalsets, LLM judges and eval infrastructure, and asks for 5+ years of software engineering ⁷. The ML Engineer for assistant quality improves Glean's assistant and agents with ML and LLM systems across retrieval, search and post-training, and asks for 2+ years in ML, applied AI or software engineering with significant ML ownership ⁸.
• Databricks. A Senior Staff Applied AI Engineer owns a retrieval stack, search subagents and the evaluation behind them, and the posting asks for 10+ years of software engineering ³. A Staff Machine Learning Engineer builds ML pipelines for preprocessing, feature engineering, model training and hyperparameter tuning, and works on LLM fine-tuning ⁴.
• Scale AI. The Applied AI Engineer builds and deploys end-to-end applications for public-sector clients, drawing on the latest developments from the biggest AI labs and on open-source models, and designs evaluation frameworks ⁵. The ML Engineer retrains and tunes production models and fine-tunes models for specific products ⁶. A third posting says its Frontier Agents Engineer role differs from traditional ML roles that focus on a single model or product ¹⁵.
• Anthropic. The Applied AI Engineer advises companies building on the Claude API, with evaluation suites and prototypes ¹⁶. The Research Engineer, Machine Learning (Reinforcement Learning) builds training infrastructure across GPU clusters and training environments, the model layer at its far end ¹⁷.

Because the role is new and its definition is still settling. The authors of a 2022 conference paper called AI engineer a relatively new role, less established than data scientist or software engineer. They analyzed 367 Dutch job ads from April 2018 to April 2021 that they had selected as AI engineering roles, treated AI and ML as synonyms, defined the work as machine learning plus software engineering aimed at production-ready ML systems, and found "ML engineer" (173 ads) far more common as a title than "AI engineer" (77) ¹⁰.

That study's data end in April 2021, well before the LLM-era postings in this article. It found three profiles among the 125 ads the authors coded by hand: 47 asked for both data science and software engineering, 38 leaned toward software engineering and 35 toward data science, and 5 gave no information ¹⁰. The mix is still visible. Stripe's Growth Platform posting wants an ML engineer who can move between modeling and software engineering ², and Latent Space itself noted that the line between the layers is permeable, with AI engineers sometimes fine-tuning and hosting models ¹⁴. Three postings show how far a title can drift from the work:

• Glean's AI Engineer for quality and post-training asks for 5+ years in AI or ML engineering and hands-on LLM post-training, and the work includes supervised fine-tuning and preference- or reward-based optimization such as DPO, PPO and GRPO ⁹. That is model-layer work under an AI title.
• Databricks's Staff Machine Learning Engineer posting refers to the people it wants as GenAI engineers ⁴.
• Scale's Public Sector ML team describes building agent frameworks, retrieval pipelines and evaluation tools alongside its model work ⁶.

None of this makes the distinction fiction. It means the layer you work on is a more useful unit than the label on the job.

• Read the first three responsibilities. Building, training, evaluating and deploying models points to ML, as in Stripe's Link posting. Evaluating and integrating AI capabilities and building agents and workflows points to AI engineering, as in Stripe's AI Engineer posting ^11,1.
• Check the experience line. Postings say whether they want years of software engineering or years of ML work, as the table above shows. A few AI-titled postings ask for AI or ML years instead, so read the line and not just the title ^1,11,9.
• Look for training verbs. Retraining, hyperparameter tuning, feature engineering and reward-based optimization mean model-layer work whatever the title says ^6,4,9.
• Look at who the work is for. Customer-facing applied AI roles add technical discovery with customers and occasional travel to the build work, as the Anthropic and Scale postings describe ^16,5.

If you are preparing for the AI engineer side, see our guide to AI engineer interview questions.

There is no clean official answer. The Bureau of Labor Statistics' A-Z index of occupations, which we checked on October 23, 2026, has no entry for either title ¹⁸. Nearby official categories are software developers, with a median annual wage of $135,980 in May 2025, data scientists, at $120,230, and computer and information research scientists, at $140,300 ^19,20,21. Those medians cover all employers and experience levels, and no category is AI or ML engineering, so they frame the question without answering it.

Employer-posted ranges are the title-level pay data this article uses. The table lists the US ranges in the postings above as they read on October 23, 2026, with the location each posting names. Glean labels its figures a standard base salary range and says certain roles may also be eligible for variable compensation and equity. Databricks calls its figure the expected salary range (on-target earnings for commissionable roles) and says the total package may add an annual performance bonus and equity ^7,4.

Table 2. Scroll sideways for every column.

The ranges overlap, and the title does not predict them. Glean posts the same base range for two AI Engineer roles and one ML Engineer role, and the one lower range belongs to an ML role with a lower experience floor of 2+ years ^7,9,13,8. At Databricks the highest range goes to the most senior posting, an AI-titled role asking for 10+ years ³. Each range is one employer's target for one posting, and we have not averaged them.

The widest gap we saw is a warning about comparing across employers. Anthropic's postings list an annual salary of $500,000 to $850,000 for its Research Engineer, Machine Learning (Reinforcement Learning) and $200,000 to $320,000 for its Applied AI Engineer, Enterprise, both for San Francisco or New York ^17,16. One is frontier-lab research engineering and the other is customer-facing, and the postings do not say what drives the gap, so it cannot be read as a premium for ML over AI engineering.

• You are a software engineer with little ML experience. Most AI engineer postings in our sample lead with engineering experience. Stripe's AI Engineer asks for backend fundamentals and agent-building experience and lists no ML or statistics minimum, and Glean's evals role asks for 5+ years of software engineering ^1,7. Our guide to moving from software engineer to AI engineer lays out a build-first plan.
• You have production ML experience or a strong statistics background. ML engineer postings reward it: Stripe asks for statistics, model evaluation and experimentation, and for people who can tell an offline improvement from business impact ². Some AI-titled postings ask for the same background, such as Glean's post-training AI Engineer role (5+ years in AI or ML engineering) and Databricks's AI FDE posting (hands-on data science experience) ^9,12.
• You want to train models, not build on them. Look at research engineer postings. Anthropic's asks for experience with ML frameworks such as PyTorch, TensorFlow or JAX and industry experience in ML research, and says candidates need not have academic research experience or publications ¹⁷. BLS lists a master's degree as the typical entry-level education for computer and information research scientists, an official research-focused category, though not the same job ²¹.
• You are not sure. Pick three postings you would want and mark which requirements you already meet. The pattern, engineering years or ML years, is your answer. The generalist profile is real: the Dutch study found it in 47 of 125 hand-coded ads ¹⁰.