How to Get a Job in AI: Paths In, Skills and Where to Look

To get a job in AI, pick one role family (engineering, research, product, solutions, sales, operations or policy), build two or three pieces of visible work that prove you can do that job with today's AI tools, and apply where AI companies post first: their own careers pages and AI-focused boards. Warm outreach to the team usually beats a cold application.

Definition: An AI job is a role whose main output is building, deploying, selling, governing or evaluating AI systems. That covers researchers training models and account executives selling them.

Many people trying to break in aim at the hardest door first: a research role at a frontier lab. No reply, so they conclude AI is closed to them.

It isn't. The building just has more doors than the one with the line outside.

This guide maps those doors, what each one asks for, how to earn real experience before anyone pays you, and a 90-day plan. If you only want the list of places to search, our roundup of the best AI job boards covers that in depth.

How to get a job in AI: the short answer

Our read is that five moves cover most successful paths in.

Choose a role family, not "AI". AI is a sector, not a job. Hiring managers hire for a function.
Close the specific skill gap. An ML engineer needs Python, model evaluation and deployment. A solutions engineer needs to build a working demo in an afternoon. A policy analyst needs to write clearly about how the systems work.
Ship proof. Two or three public projects, write-ups or contributions that look like the job tend to outweigh a certificate.
Target the right kind of employer. Frontier labs, AI startups and AI teams inside established companies hire very differently.
Apply early and warm. Track new postings weekly and pair each application with a short note to someone on the team.

The rest of this page goes deeper on each step.

What the AI job market looks like right now

Demand is real, but the numbers depend on who is counting.

Per the Stanford HAI AI Index 2026, which uses Lightcast data, AI skills appeared in 2.56 percent of all US job postings in 2025. Lightcast's own summary of that chapter puts the rise at about 55 percent over the prior year, with "agentic AI" skills growing from 0.06 percent of postings in 2024 to 0.23 percent in 2025.

Indeed Hiring Lab's public AI tracker measures something broader (postings that mention AI or machine learning terms) and shows a bigger number. Its seven-day average for the US was about 6.7 percent of postings on August 31, 2026, roughly double the 3.4 percent of a year earlier. The two sources define an AI job differently, so we'd read them as direction, not as one precise share.

The longer-run outlook is also positive. The BLS projects employment of computer and information research scientists to grow 22 percent from 2025 to 2035 (median pay $140,300 in May 2025) and data scientists to grow 35 percent over the same decade (median pay $120,230). The World Economic Forum's Future of Jobs Report 2025, a survey of over 1,000 employers, lists AI and machine learning specialists among the fastest-growing roles through 2030. Forecasts like these get revised, so treat them as a tailwind rather than a promise. Our guide to the fastest-growing AI jobs looks at the new roles behind those numbers.

Now the harder part. The August 2026 update of the "Canaries in the Coal Mine" paper by Erik Brynjolfsson, Bharat Chandar and Ruyu Chen, using ADP payroll data through June 2026, finds that employment of 22 to 25 year olds in the most AI-exposed occupations stands 19 percent below where it would be had it kept pace with less-exposed peers. The declines concentrate where AI substitutes for work, not where it complements it.

Our takeaway: AI is creating jobs and squeezing generic entry-level work at the same time. That's why proof of skill matters more for a first AI job than it did a few years ago.

The AI role families and what each one needs

Here is one way to sort the field. The titles vary by company, so read job descriptions rather than trusting labels.

Role family	What the job is	What tends to get you hired
ML or AI engineer	Builds products and pipelines on top of models	Shipped apps, evals, deployment skills
Research scientist or engineer	Trains and studies models	Papers, replications, strong math and code
Solutions or forward-deployed engineer	Builds with customers inside their systems	Fast prototyping, client-facing skills
AI product, sales, ops, policy	Decides, sells, runs and governs AI products	Domain depth plus hands-on AI fluency

Technical roles

ML or AI engineer. The largest technical door. You connect models to real products: retrieval, agents, fine-tuning, evaluation and the infrastructure around them. Python sits at the top of Lightcast's 2025 list of specialized skills in US AI postings, and skills tied to running systems at scale, such as workflow management, automation and scalability, also make its top 10. Many strong candidates come from ordinary software engineering. A 2025 Pragmatic Engineer interview by Gergely Orosz traced one such move: an engineer who taught herself to build apps on large language models through side projects and hackathons, then landed an AI engineering role. Its advice for others was much the same: build LLM products on the side and learn the core concepts.

Applied scientist. Common at larger tech companies. Part research, part product: you design experiments, pick models and measure whether they work. A master's or PhD is common here, though not universal.

Research scientist or research engineer. The narrowest door. Labs hire a small number of people to push model capability, safety and interpretability. Publications, replications of known results and serious open-source work are the usual currency.

Data and ML infrastructure. Data engineers, ML platform engineers and inference and GPU specialists keep training and serving costs under control. If you already work on distributed systems, this can be the shortest path in.

Hybrid and non-technical roles

AI product manager. Decides what to build and how to measure model quality in the product. Evaluation design is a real differentiator here.

Solutions engineer or forward-deployed engineer. Sits with customers and builds working deployments inside their systems. a16z's Joe Schmidt called it the hottest job in startups in 2025, because enterprises buying AI need someone to connect the product to their own data and workflows. The title has spread fast: both OpenAI and Anthropic have advertised forward-deployed engineering roles on their public job boards.

Sales and go-to-market. Probably the most underrated door. When we reviewed Anthropic's public Greenhouse board, account executive openings outnumbered research engineer and research scientist openings combined. AI companies need people who can sell a technical product to a skeptical buyer.

Operations, policy and safety. Strategy and operations, trust and safety, policy research and compliance teams are growing alongside the products. These roles reward clear writing and judgment.

Evals and data annotation. Labs and data vendors hire people to write test cases, grade model outputs and produce training data, often on contract. Domain experts (lawyers, doctors, coders, linguists) can earn real money here. Pay and stability vary widely, so read the terms carefully before treating it as a career step.

"You need a PhD to work in AI"

This is the most common reason people never apply. There's a fair version of it: frontier research roles do lean heavily on graduate training, and the BLS lists a master's degree as the typical entry education for computer and information research scientists.

But.

Research is one door out of many. Anthropic's careers page notes that about half its technical staff have PhDs, and that about half had no prior ML experience before joining. BLS lists a bachelor's degree as the typical entry point for data scientists. Most of the roles in the table above care more about what you have built than where you studied.

In our view, a PhD is a strong signal for one job family, and a weak excuse for skipping the others.

How to get experience working in AI before anyone hires you

The classic trap: no job without experience, no experience without a job. In AI the trap is softer than in most fields, because the tools are public and the work is visible.

Some ways to build experience that hiring teams can check:

Ship a small app on a model API. Pick a narrow, real problem (summarizing permit filings for a contractor, triaging support tickets for a friend's shop), build it, and write up what broke. A working link beats a list of courses.
Build an evaluation set. Write 100 to 200 test cases for a task you know well and measure how two or three models perform. Evals are something many AI teams need and few applicants show.
Contribute to open source. Fix documentation, add a test or reproduce a bug in a library you use. Small merged pull requests are easy to verify.
Do competitive or benchmark work. Kaggle competitions and public leaderboards give you a scored, comparable result.
Replicate a paper. For research-leaning roles, reproducing a published result (and noting where it did not reproduce) is about as credible as unpaid work gets.

Anthropic's careers page, for one, invites candidates to put independent research, a thoughtful blog post or open-source work at the top of the resume. We'd take that advice for any AI employer. Our guide to getting an AI job with no experience turns these into a step-by-step plan.

How to transition into an AI career from another field

Career changers have an edge that new grads lack: domain knowledge. AI companies selling into health care, law, finance or logistics need people who know how those buyers think.

A practical way to make the switch:

Stay in your domain, add AI. A nurse who builds a documentation-review prototype is a stronger candidate for a health AI startup than a generic bootcamp graduate.
Aim for the adjacent role first. Marketers move into AI growth roles, account managers into AI sales, analysts into data and evals roles, software engineers into AI engineering. One step sideways is easier than two.
Use AI visibly in your current job. An internal tool you built, with a measured result, makes a strong interview story.

Our sibling guide on how to get a job in tech covers the broader switch for people coming from non-tech backgrounds.

How to get hired at an AI startup vs an AI lab vs an AI team

"AI company" covers three very different employers.

Frontier labs. The handful of companies training the largest models. The bar is high and the visible research roles draw enormous applicant volume. Their go-to-market, policy and operations teams are often easier entry points than research.

AI startups. Smaller companies building products on top of models. Hiring is faster, titles are looser and you will usually do several jobs at once. Founders often care most about whether you can ship this month. Our playbook on how to get a job at a startup covers the process.

AI teams inside other companies. Banks, retailers, insurers and software firms building AI features. Stability is higher and the domain matters a lot. This is often the most realistic first AI job for career changers.

Before you join a startup, it is worth asking whether the company will still matter in two years. A thin wrapper on someone else's model can be overtaken by the next model release, which is the case we make in our guide to AI startup moats. Ask about runway too: our explainer on default alive or default dead gives you the math to check a startup's answer. If an offer includes equity, our guide to the employee equity offer letter explains the terms worth checking before you sign.

Where to look for AI jobs

AI companies tend to post first on their own careers pages, which usually run on applicant tracking systems such as Greenhouse, Ashby or Lever, and only later (or never) on large general boards. That makes a weekly habit of checking new postings more useful than daily scrolling.

The 1752vc careers board has a dedicated AI jobs track listing open roles at AI companies alongside venture-backed startups and VC firms, with a US focus. It is refreshed every week, shows each employer's own posting date and keeps only recent postings, with filters for level, location and time posted. The full list of boards and search tricks is in our AI job boards roundup.

A 90-day plan to get your first job in AI

An illustrative plan for someone with a full-time job and about 8 to 10 hours a week. Adjust the numbers to your situation.

Days 1 to 30: choose and close the gap - Read 30 to 40 real job descriptions in one role family and list the skills that repeat. Filtering the AI track by Entry level and Past 30 days is one quick way to collect them. - Pick the two most common skills you lack and spend most of the month on them. - Start project one, scoped to finish in three weeks.

Days 31 to 60: ship proof - Publish project one with a short write-up and numbers. - Build project two, ideally an eval set or a small contribution to an open-source tool the target companies use. - Make a target list of 30 companies: 10 startups, 10 AI teams at established companies, 10 stretch employers.

Days 61 to 90: apply and reach out - Apply to 5 to 8 well-matched roles a week, each within days of posting. - For each, send a short note to an engineer, manager or recruiter on the team. - Track every application: company, role, posting date, contact, date sent, follow-up date, outcome.

Run the numbers on that plan: 6 applications a week for 4 weeks is 24 applications. If 1 in 8 turns into a first conversation (an illustrative rate, not a benchmark), that is 3 interviews to learn from, and your projects give each one something concrete to discuss.

A short note you can adapt:

Subject: Your [role] opening, and a small thing I built

Hi [Name], I saw the [role] posting on your team. I recently built [one-line project] that [result with a number], write-up here: [link]. It overlaps with [specific thing from the job description]. I've applied through the site and would value 15 minutes if you think there's a fit. Thanks, [Name]

Common mistakes when trying to break into AI

Applying to "AI" in general. A generic application to 200 roles across five role families usually reads as no fit for any of them.
Collecting certificates instead of shipping. Courses help you learn. Hiring teams mostly want to see what you made with what you learned.
Only targeting the famous labs. The applicant pool there is huge, and many good first AI jobs sit at startups and inside non-tech companies.
Hiding your domain. Career changers who drop their old field from the story throw away their best edge.
Applying late. A posting that is a month old may already have a shortlist.

Where we land

AI hiring rewards proof over pedigree more than most fields we know, mainly because the work is so visible. A merged pull request, a working demo or a thoughtful eval write-up can be checked in minutes.

So our suggestion is narrow and practical: one role family, two or three projects that look like the job, a weekly check of fresh postings, and a personal note with every application. It is our answer, not the only one. Your background, location and timeline may point somewhere else.

The bottom line

The research door gets the headlines. The sales, solutions, ops and engineering doors get most of the hires.

Pick a door.

Then show up with something you built.

Key takeaways

Choose one AI role family first; engineering, research, solutions, product, sales, operations and policy each hire on different evidence.
AI demand is rising by every measure we checked, but sources disagree on the size, and early-career workers in AI-exposed jobs face a tighter market.
Visible proof (shipped apps, eval sets, open-source contributions, replicated papers) tends to matter more than certificates.
A PhD matters mainly for research roles; many AI jobs hire on skills and domain knowledge.
Frontier labs, AI startups and AI teams at established companies hire differently, so it helps to target each one on purpose.
A 90-day plan of choosing, shipping and then applying early and warm gives most candidates a realistic path in.

Frequently asked questions

Can you get a job in AI without a degree?

Yes, in many roles, though it is harder for research. Anthropic's careers page, for one, says it cares about what you can do, not where you learned it, and roles in AI engineering, solutions, sales, operations and evals often hire on demonstrated work. A strong portfolio of shipped projects, open-source contributions or evaluation work can stand in for a degree, while research science still leans toward graduate training.

How long does it take to get into AI?

It depends on your starting point. A software engineer moving into AI engineering can often become competitive in a few months of focused building. Someone without coding experience aiming for a technical role should plan for a year or more. Non-technical people moving into AI sales, operations or product roles can sometimes switch within one to three months.

Do AI companies hire non-technical people?

Yes. AI companies hire account executives, solutions and customer success staff, marketers, recruiters, operations, policy and legal teams. On some large AI company job boards, sales listings outnumber research listings. Hands-on fluency with the product and domain knowledge of the customers you would serve make a non-technical application much stronger.

Is it too late to start a career in AI?

We don't think so. Job postings mentioning AI skills rose in Lightcast data for 2025 and kept rising in Indeed data into 2026, and federal projections show fast growth for data and research scientist roles. The market is more competitive for generic entry-level work, so starting now with visible projects matters more than starting early.

What programming language should I learn for an AI job?

Python is the standard choice. It was the top specialized skill in US AI job postings in 2025, per Lightcast data in the Stanford AI Index, and in our view most model libraries and APIs support Python first. SQL is a useful second skill for data roles, and TypeScript helps for engineers building AI product interfaces.

Sources

Disclaimer: This guide is for general education only and is not legal, tax or investment advice. Laws, market data and program terms change, so it may not reflect the latest developments or fit your situation. Treat it as a starting point, not a source of truth, and talk to a qualified lawyer, accountant or financial adviser before you make decisions.