Skip to content
← All posts
Kello blog research · Research · November 2026

How to hire AI engineers in 2026: which role, what to pay, where to find them.

For founders and hiring managers who need AI engineers now: AI engineer vs ML engineer vs research engineer, salary by market, where AI engineers come from and what to test for.

Cite
AI engineer job postings in 2026
35,045
People titled AI engineer
134,122
ML and research engineers compared
117,801
Data as of
4 Nov 2026

Based on 35,045 AI engineer job postings from 2026, the work histories of 134,122 people titled AI engineer, and 117,801 ML and research engineers for comparison. Data as of 4 Nov 2026.

The short answer

To hire AI engineers, decide the work first. To build products on existing models, hire an AI engineer. AI engineer salary in the US: postings advertise $140k to $196k ($173k to $240k at startups), and AI engineers report a median base of $141k. To train your own models, hire an ML engineer. Test evaluation work: it's the biggest gap between what postings ask for and what candidates describe.

01 · The people AI engineers have a median 3.6 years of experience, against 5.7 for ML engineers.

The three AI titles hold different people. Across 134,122 AI engineers, 106,230 ML engineers and 11,571 research engineers, the median AI engineer has 3.6 years of experience, against 5.7 for ML engineers and 6.6 for research engineers. 4.3% of AI engineers hold a PhD, against 10.9% of ML engineers and 23.9% of research engineers. Pick the title that matches the work (chapter 04).

What is an AI engineer? An AI engineer builds software products on top of existing AI models: LLM features, retrieval (RAG), agents and the evaluation that keeps them reliable. An ML engineer trains, tunes and serves models. A research engineer turns new methods into working code. Generative AI engineer, LLM engineer and applied AI engineer are versions of the AI engineer title, so the same advice applies if you want to hire generative AI engineers.

TitleMedian years of experience
AI engineer3.6
ML engineer5.7
Research engineer6.6
Source: Kello people index, current job titles · Data as of 4 Nov 2026

Most AI engineers did something else before. Of the 134,122 people titled AI engineer today, 70.3% started their current role in 2025 or 2026. Only 22.1% of those with a previous job held the AI engineer title there. Most candidates' experience sits under other titles, which changes where you search (chapter 05).

02 · The race One in six AI engineers works at a startup.

Startups compete with the largest companies for the same people. Of today's AI engineers whose employer's stage we know (67,032), 16.5% work at early- or growth-stage startups and 29.9% at mega corporations. On the job side, 1,104 early- and growth-stage startups posted 1,689 AI engineer jobs in 2026: 8.4% of the postings from direct employers we could identify.

A quarter of postings don't come from the employer. Recruiters, staffing firms and job boards posted 24.4% of 2026 AI engineer postings whose employer we know. We leave them out of every employer and pay figure below.

The pool is split between two very different markets. India has more people titled AI engineer than the US (24.2% against 19.1%). But the median AI engineer in India has 2.1 years of experience. In the US it's 6.5.

Our view: the common mistake is one posting for three jobs. It asks one person to build agents, train models and publish research. Others copy a US experience bar onto an Indian role, or compete on a title instead of on the work. Pick the role that matches the work, pay the market rate for it, and test for the skill that's hardest to see on a profile.

03 · Winning What a good AI engineer hire looks like.

Our view: hire for the work you have, not for the label. If the job is LLM features, RAG and agents, hire an AI engineer and look well beyond people with the title. If the job is training models, hire an ML engineer. Pay the startup rate, and test for evaluation, the skill with the biggest gap between what postings ask for and what candidates describe.

Four problems usually stand in the way:

  • We don't know which AI role we need.
  • There's no one to hire.
  • We don't know what to pay.
  • We can't tell who's good.

We read 35,045 AI engineer postings and the careers of 134,122 AI engineers, 106,230 ML engineers and 11,571 research engineers. Which problems matter most depends on your situation:

If you…What the data showsStart with
Build on existing models: LLM features, RAG, agents73.4% of AI engineer postings mention LLMs; 8.9% mention training models.Chapters 04 and 07
Train or fine-tune your own models52.0% of ML engineer postings name PyTorch, TensorFlow or JAX, against 23.2% for AI engineers.Chapters 04 and 06
Need new methods or published research23.9% of research engineers hold a PhD, against 4.3% of AI engineers.Chapter 04
Hire in India63.9% of Indian AI engineer postings ask for 4 or more years. 29.3% of Indian AI engineers have that.Chapter 05
Can't pay $150kUS AI engineer postings at entry level advertise $115k to $161k.Chapters 05 and 06

04 · Which role AI engineer vs ML engineer vs research engineer: pick by the work.

If you don't know which AI role you need, start from what the person will build, because the three titles hold different people. AI engineers are the least experienced and the most likely to come from software (chapter 01). ML engineers have more model work behind them. Research engineers are the most academic and the most likely to come from a university or lab.

AI engineerML engineerResearch engineer
People with the title134,122106,23011,571
Median experience, US6.5 years7.8 years8.3 years
Previous job in software engineering32.6%22.3%22.3%
Previous job in data science or ML39.4%54.8%50.7%
Previous job at a university or lab8.4%11.5%23.3%
Work at early- or growth-stage startups (where the stage is known)16.5%13.4%19.0%

Source: Kello people index, current job titles, interns excluded · Data as of 4 Nov 2026. Research engineers here are those whose role is classed as data science and ML.

The postings for each title ask for different work. AI engineer postings are about building with models: LLMs, agents, RAG and the product around them. ML engineer postings name training frameworks and model serving. Research engineer postings ask for papers and PhDs.

Postings that mention…AI engineerML engineerResearch engineer
LLMs73.4%50.3%52.6%
Agents62.0%32.6%35.3%
RAG, retrieval or embeddings55.4%29.0%12.3%
Evaluation or benchmarks44.3%41.1%54.4%
Full stack, frontend, TypeScript or React.js23.6%9.1%9.9%
PyTorch, TensorFlow or JAX23.2%52.0%52.0%
Model serving, MLOps or inference32.8%47.4%32.6%
Training models8.9%20.9%30.0%
Papers or publications4.9%9.6%43.1%
A PhD5.4%16.6%49.4%
Unique 2026 postings35,04522,4522,747

Source: Kello jobs index, unique postings from 1 Jan to 31 Oct 2026 · Data as of 4 Nov 2026. Research engineer postings are those whose description mentions machine learning, deep learning or LLMs.

So the choice is mostly about training. If you call a model through an API and build the product around it, you need an AI engineer. If you need to train, fine-tune or serve your own models, you need an ML engineer. If the job is to invent a method nobody has shipped, you need a research engineer, and you'll pay the most for one (chapter 06).

Two variants worth knowing. If the AI engineer will work inside customers' systems, that's the forward deployed engineer, and our forward deployed engineer hiring guide covers it. If this is one of your first engineers, read our guide on how to hire a founding engineer as well.

Our view: if your product calls existing models, make your first AI hire an AI engineer from a software background, and add an ML engineer only when you have a reason to train. Most of the first year is product work: 73.4% of AI engineer postings mention LLMs, and only 8.9% mention training models. A software-trained AI engineer ships that work, and 32.6% of AI engineers came from software. If your product is your own model, as in vision, robotics or bio, hire the ML engineer first. Hire a research engineer first only when the method itself is the product, and budget from the typical US range of $170k to $272k.

05 · Sourcing Where to hire AI engineers from: 1 in 3 came from software engineering.

"There's no one to hire" usually means searching only for the title. Only 22.1% of AI engineers held the title in their previous job, so most of the pool's experience sits under older titles. Here is where today's AI engineers were in their previous job:

Previous jobAI engineersML engineers
Data science or ML role39.454.8
Software engineering role32.622.3
Already titled AI engineer22.13.4
Internship17.915.6
University or research lab8.411.5
Data analytics role5.35.6
Source: Kello people index, previous job of current AI and ML engineers · Data as of 4 Nov 2026

That gives you three pools, in order of size.

1. Data scientists and ML people moving to applied work. 39.4% of AI engineers came from a data science or ML role, including 7.1% titled data scientist and 6.3% titled ML engineer. They know models. Test whether they've shipped product.

2. Software engineers who picked up LLMs. 32.6% came from a software engineering role, against 22.3% of ML engineers. They know production. Test whether they've built and measured an LLM feature (chapter 07).

3. People already on your team. 21.5% of all AI engineers held another role at their current employer first, and 8.0% were in software engineering there. Your strongest backend engineer may be your first AI engineer.

A search that requires the AI engineer title misses most of the market. 77.9% of AI engineers with a previous job held a different title there.

Set the experience bar by market.

In India, the bar most postings set rules out most people who already hold the title. 63.9% of Indian AI engineer postings ask for 4 or more years. Only 29.3% of India's AI engineers have that much, and 47.8% have under 2 years. In the US the two match more closely: 71.9% of postings ask for 4 or more years, and 68.4% of US AI engineers have it.

ExperienceUSUKAustraliaIndia
Under 2 years16.5%16.9%20.5%47.8%
2 to 4 years15.1%17.3%16.2%22.4%
4 to 8 years26.5%29.2%23.9%19.8%
8 or more years41.9%36.6%39.3%9.5%
AI engineers measured25,5604,2581,91332,437

Source: Kello people index, current AI engineers by current country · Data as of 4 Nov 2026. This compares postings with people who already hold the title; the wider pool of software engineers and data scientists moving into the role is more experienced.

Boolean search strings for LinkedIn.

Paste these into LinkedIn's keyword search, or into Google with site:linkedin.com/in in front. Then read what each person says they built.

("software engineer" OR "backend engineer" OR "full stack engineer") AND (LLM OR RAG OR "retrieval augmented" OR agents OR agentic OR embeddings) AND (production OR shipped OR deployed)
("data scientist" OR "machine learning engineer" OR "ML engineer") AND (LLM OR RAG OR agents OR "generative AI") AND (production OR API OR deployed)
("AI engineer" OR "applied AI engineer" OR "generative AI engineer" OR "LLM engineer") AND (evals OR evaluation OR benchmarks)

06 · Salary AI engineer salary: $140k to $196k advertised in the US, $141k reported.

If you don't know what to pay, start from the market range for the title you chose. We use two lenses. Advertised pay is what companies offer in job postings. Reported pay is what people say they earn on Glassdoor and, in India, AmbitionBox.

Advertised pay. The typical advertised AI engineer salary in the US is $140k to $196k: the median bottom and median top of the range, across 1,759 unique 2026 postings that state pay, with recruiters and job boards removed. ML engineers are advertised at $152k to $240k and research engineers at $170k to $272k.

Title (advertised)US, all employersUS, early- and growth-stage startupsUKIndia
AI engineer$140k to $196k$173k to $240k£80k to £100k₹11.8L to ₹18.5L
ML engineer$152k to $240k$178k to $240kToo few postings₹17.2L to ₹19.6L
Research engineer$170k to $272kToo few postingsToo few postingsToo few postings

Source: Kello jobs index, unique 2026 postings with an advertised annual range; recruiters, staffing firms, job boards and mismatched employer records removed · Data as of 4 Nov 2026. US AI engineer: 1,759 postings (72 from 59 startups). ML engineer: 1,882 (66 from 53 startups). Research engineer: 396 (43 from 22 startups, a fifth from one employer, so not shown). UK AI engineer: 65. India AI engineer: 1,213; ML engineer: 1,314, of which 16.6% come from one employer. AI labs are not counted as startups. These are advertised ranges, not offers.

Startups advertise more than the market as a whole. US early- and growth-stage startups advertise $173k to $240k for AI engineers, across 72 postings from 59 startups, against $140k to $196k across all direct employers. The startup sample is small. If you compete with other startups for the same people, budget from the startup range.

Seniority labels on postings are loose. Mid-senior postings advertise the most, but entry-level postings advertise more than associate ones. In the US, AI engineer postings advertise:

Seniority (as posted)Typical advertised rangePostings
Entry level$115k to $161k150
Associate$107k to $134k95
Mid-senior$144k to $195k1,070

Source: Kello jobs index, unique 2026 US AI engineer postings with an advertised annual range, recruiters and job boards removed. Director-level postings are too few to show · Data as of 4 Nov 2026

Know where your number sits. A quarter of US AI engineer postings start below $111k, and a quarter top out at $250k or more.

Outside the US, recruiters post most of the pay ranges. In the UK, recruiters and job boards posted 270 of the 339 AI engineer postings that state pay, so the UK range rests on 65 postings from direct employers. In Australia they posted 51 of 57, too few direct employers to give a range.

Most postings don't state pay. 28.5% of US AI engineer postings state a range, and 16.2% worldwide. A posting that does stands out.

Reported pay: what AI engineers say they earn.

The median AI engineer in the US reports a base salary of $141k, with the middle half between $129k and $161k. That sits inside the advertised range. ML engineers report $159k and research engineers with an AI or ML title $151k.

Title (US, reported base)MedianMiddle halfSalaries
AI engineer$141k$129k to $161k1,371 at 529 companies
ML engineer$159k$136k to $186k16,809 at 2,189 companies
Research engineer (AI or ML title)$151k$126k to $184k676 at 78 companies

Source: Glassdoor self-reported salaries, regular employees, annual base pay, records updated in 2025 or 2026, weighted by number of salaries with no company above 10% of a group · Data as of 4 Nov 2026. Before the cap, one company held 52.2% of the research engineer salaries and 29.4% of the ML engineer salaries. Glassdoor holds too few UK and Australian salaries updated since 2025 to report.

On top of base, US AI engineers report a median of $11.25k in additional pay such as bonuses (950 salaries). ML engineers report $18.75k.

Why the two lenses differ. Advertised ranges show what employers posting now offer, and skew to US states that require pay in postings. Reported pay is base salary people already hold, reported over several years, across every kind of employer that uses the title. Treat advertised pay as the offer you compete with, and reported pay as what a candidate probably earns today.

India: AI engineer salary by experience.

In India, AI engineers report a median CTC of ₹13.25L (middle half ₹8.25L to ₹25.25L), across 5,956 salaries at 1,749 companies. Experience moves it most:

ExperienceAI engineerML engineerResearch engineer
0 to 3 years₹10.25L₹8.75L₹7.25L
4 to 7 years₹20.75L₹18.75L₹17.75L
8+ years₹34.25L₹34.75LToo few salaries
All levels₹13.25L₹11.25L₹11.25L

Source: AmbitionBox self-reported salaries, median CTC weighted by number of salaries; no company holds more than 6.0% of a group's salaries · Data as of 4 Nov 2026. AI engineer: 5,956 salaries at 1,749 companies. ML engineer: 22,571 at 5,350. Research engineer (data science and ML roles): 249 at 97.

The reported median sits inside the advertised Indian range of ₹11.8L to ₹18.5L. Since 70.2% of India's AI engineers have under 4 years of experience (chapter 05), the 0 to 3 year figure covers most of the pool.

07 · Interview How to interview an AI engineer: test evaluation, not prompts.

If you can't tell who's good, test for the work postings ask for and few candidates describe. We compared what 2026 AI engineer postings mention with what AI engineers write in their own role descriptions, using the same phrases for both. Every skill is asked for more often than it's described, partly because postings are longer. Evaluation has the biggest gap: 44.3% of postings mention it, and 13.1% of AI engineers describe it.

MentionedPostingsAI engineers' role descriptions
LLMs73.436.4
Production or deployment80.440.5
Agents62.030.3
RAG, retrieval or embeddings55.429.1
Evaluation or benchmarks44.313.1
Model serving, MLOps or inference32.811.5
Fine-tuning18.57.2
Training models8.93.1
Source: Kello jobs index (unique 2026 postings) and people index (current AI engineers' role descriptions) · Data as of 4 Nov 2026

Not everyone writes a role description, so these shares show what people choose to describe, not everything they did. A phrase missing from a profile doesn't mean the skill is missing. With a stricter evaluation list (evals, model evaluation, evaluation frameworks or pipelines, benchmarks) the gap is wider: 26.4% of postings against 6.5% of role descriptions.

Turn each gap into an interview question.

  • Evaluation (3.4x). Ask how they measured whether an LLM feature worked: the test set, the metric and a change they made because of it. It's the skill hardest to see on a profile, so give it time in the interview.
  • Production (80.4% of postings). Ask about an LLM feature real users relied on, and what broke after launch.
  • RAG and agents (1.9x and 2.0x). Ask them to draw a retrieval or agent system they built, and to explain one design choice and its cost.
  • Software basics. 23.6% of AI engineer postings mention full stack, frontend, TypeScript or React.js work. If the role ships product, include a normal coding round.

Leave training and papers off an AI engineer scorecard. Only 8.9% of AI engineer postings mention training models and 4.9% mention papers. If you need either, you're hiring an ML or research engineer (chapter 04).

AI engineer job description template.

Write the job description around the work AI engineers do, and name evaluation: 44.3% of AI engineer postings mention it, but only 13.1% of AI engineers describe it. Naming it tells candidates it matters and gives you something to test. The words in square brackets are for you to fill in.

SectionWhat to write
About the roleYou'll be our [first / next] AI engineer. You'll build [product or feature] on large language models and ship it to [users or customers].
What you'll doBuild LLM features and agents that run in production. Design retrieval (RAG) over [your data]. Write the evaluations that tell us whether a change made the product better. Decide with [founders or product] what to build next.
What we're looking for[Years] of building and shipping software. LLM features you've shipped to real users. Evaluations you've designed yourself.
Nice to haveModel training or fine-tuning, if your product needs it.
Pay[Salary range] plus [equity]. US startups advertise $173k to $240k for AI engineers.

Why this shape: AI engineer postings mention production or deployment (80.4%), LLMs (73.4%), agents (62.0%) and RAG (55.4%). In the US, 71.9% of postings that state a minimum ask for 4 or more years.

Putting it together: one founder's plan.

Daniel runs an eight-person seed-stage startup in the US, building an AI agent for finance teams. Daniel is a fictional founder; every figure comes from the chapters above.

  1. The role. The product calls existing models through an API, so he hires an AI engineer, not an ML or research engineer (chapter 04).
  2. The search. Software engineers who have shipped LLM features, data scientists moving to applied work, and his own backend engineer. He doesn't require the AI engineer title (chapter 05).
  3. The experience bar. In the US, 4 or more years matches the market (chapter 05).
  4. The salary. He budgets $173k to $240k, the startup range, and states it in the posting (chapter 06).
  5. The interview. One LLM feature in production, one agent design, and how they evaluated it, plus a normal coding round (chapter 07).

If Daniel later needs to fine-tune his own models, that's a second hire: an ML engineer at $178k to $240k.

How we know, and how Kello can help.

This guide is based on 35,045 unique AI engineer job postings from 2026, the work histories of 134,122 people titled AI engineer, and 106,230 ML engineers and 11,571 research engineers for comparison, with data as of 4 Nov 2026.

Titles are matched as phrases with a plain text analyzer. Kello's default people search treats "AI" and "machine learning" as the same word, which would count ML engineers as AI engineers, so we turned that off. "AI/ML engineer" counts as an ML engineer here. Interns are excluded. Research engineers are those whose role is classed as data science and ML (people) or whose posting mentions machine learning, deep learning or LLMs (postings). People figures cover people's current role and current country. Previous job means the job before the current one; a job can hold several roles.

Job postings are 2026 job postings (January to October 2026), each counted once however often it was reposted.

Recruiters, staffing firms, job boards and recruiting marketplaces are removed from every employer and pay figure: by company type and industry in Kello's company data, a list of known intermediaries, and a check of the ten largest employers in each group. We also drop employer records that clearly don't match the posting, such as a one-person company with dozens of postings. Startups are employers tagged early-stage or growth-stage in Kello's company data, matched by website domain; AI labs that train their own frontier models are a separate group. Many postings have no employer domain, so some recruiter postings may remain. Advertised pay figures are annual ranges in local currency from postings, not offers. Reported pay comes from Glassdoor self-reported salaries (US; annual base pay, regular employees, records updated in 2025 or 2026, no company above 10% of a group) and AmbitionBox self-reported salaries (India; CTC), weighted by the number of salaries behind each record and read from pay bands, so each median is accurate to within half a band. Groups under 20 salaries are not shown. Phrase shares count mentions, not requirements, and use the same phrase list for postings and role descriptions. A posting is a plan to hire, not a hire. The advice in each chapter is our reading of these figures.

Kello uses the same data to find AI engineers for companies like yours. Send us the role, and your shortlist usually arrives in about two hours. Most roles have people ready to interview within 7 days.

Hiring an AI engineer? Send Kello the role. Your first role is free.

Hiring under the member of technical staff title? Read our member of technical staff guide.

Updates

  • First published.

Questions.

What is an AI engineer?

An engineer who builds products on top of existing AI models: LLM features, retrieval (RAG), agents and their evaluation. 73.4% of 2026 AI engineer postings mention LLMs, and only 8.9% mention training models. The typical AI engineer has 3.6 years of experience.

How much do AI engineers make?

In the US, AI engineer postings advertise a typical salary of $140k to $196k, and early- and growth-stage startups $173k to $240k. AI engineers report a median base salary of $141k on Glassdoor, plus a median $11.25k in bonuses and other pay. In the UK, direct employers advertise £80k to £100k, though most UK postings with pay come from recruiters.

What is the AI engineer salary in India?

AI engineers in India report a median CTC of ₹13.25L on AmbitionBox: ₹10.25L with 0 to 3 years of experience, ₹20.75L with 4 to 7 years and ₹34.25L with 8 or more. Indian AI engineer postings advertise ₹11.8L to ₹18.5L.

What does an AI engineer do?

In their own role descriptions, AI engineers most often mention production or deployment (40.5%), LLMs (36.4%), agents (30.3%) and RAG (29.1%). Few mention evaluation (13.1%) or training models (3.1%).

What's the difference between an AI engineer and an ML engineer?

AI engineers build with existing models; ML engineers train and serve them. 52.0% of ML engineer postings name PyTorch, TensorFlow or JAX, against 23.2% of AI engineer postings. ML engineers have more experience (median 5.7 years against 3.6) and are advertised higher in the US ($152k to $240k against $140k to $196k).

How do you hire AI engineers for a startup?

Pick the title that matches the work, search beyond people with the title (1 in 3 AI engineers came from software engineering), budget $173k to $240k in the US, and test how candidates evaluated what they built. One in six AI engineers whose employer's stage we know works at an early- or growth-stage startup (16.5%).

What should an AI engineer job description include?

The LLM features they'll build and ship to production, the retrieval and agent work, and evaluation. 44.3% of AI engineer postings mention evaluation, but only 13.1% of AI engineers describe it, so naming it tells candidates it matters. State the pay: US startups advertise $173k to $240k.

Which companies hire AI engineers?

Companies of every size. Of today's AI engineers whose employer's stage we know, mega corporations employ 29.9% and early- and growth-stage startups 16.5%. In 2026, 1,104 early- and growth-stage startups posted AI engineer jobs. A quarter of postings with a known employer (24.4%) came from recruiters, staffing firms and job boards.

Cite this report

Kello blog. “How to hire AI engineers in 2026: which role, what to pay, where to find them.” 4 November 2026. https://kello.ai/blog/hire-ai-engineers/

Get the next Kello report by email.

Hiring data you can act on, sent when we publish.

Free. Unsubscribe any time.

Almost there. Check your inbox to confirm.

In this report

    Back to top ↑