There are roughly seven AI career paths: research, engineering, applied and solutions work, product, data, go-to-market, and policy and safety. They differ less in prestige than in daily work. To choose, match the path to the problems you enjoy, the evidence you can produce in a few months, and how much schooling you are willing to add.
Definition: An AI career path is a sequence of roles built around one kind of AI work (inventing models, building with them, deploying them for customers, deciding what to build, feeding them data, selling them, or governing them), with its own skills, entry points and ladder.
Most people start by asking which AI job pays the most or sounds the most impressive.
That question tends to produce a lot of half-finished machine learning courses.
A better first question, in our view: what do you want your Tuesday afternoon to look like? Debugging a training run and negotiating a contract are both AI careers. They just aren't the same career.
Below: the paths, a typical day in each, how each progresses, and questions to help you pick. For the job-search playbook itself, see our pillar on how to get a job in AI.
The seven AI career paths at a glance
Titles vary wildly between companies, so treat this as a map, not a taxonomy.
| Path | Core question you answer | Typical entry evidence |
|---|---|---|
| Research | Can we make models do something new? | Papers, replications, graduate work |
| Engineering | Can we build a reliable product on a model? | Shipped apps, evals, deployment |
| Applied and solutions | Can this work inside this customer? | Fast prototypes, client skills |
| Product | What should we build, and is it good? | Product judgment plus eval design |
| Data | Is the data right, and what does it say? | SQL, statistics, labeled datasets |
| Go-to-market | Who buys this, and why now? | Sales or marketing results, AI fluency |
| Policy and safety | What could go wrong, and who decides? | Writing, research, domain or legal depth |
The market signal points at several of these at once. LinkedIn's Jobs on the Rise 2026 list for the US, which ranks titles by growth in hires among its members from January 2023 to July 2025, put AI engineers first, AI consultants and strategists second, data annotators fourth and AI and machine learning researchers fifth. That's four different paths in the top five. Our guide to the fastest-growing AI jobs digs into that growth data and the new roles behind it.
The spread goes beyond tech teams too. Per the Stanford AI Index 2026, which draws on Lightcast posting data, about 2.6 percent of US job postings asked for AI skills in 2025, up from about 1.9 percent in 2024, and AI hiring spread into sectors that had historically adopted it slowly.
In the World Economic Forum's Future of Jobs Report 2025, a survey of over 1,000 employers looking ahead to 2030, two thirds said they plan to hire for specific AI skills. Forecasts get revised, so we'd read that as direction, not a promise.
So "working in AI" is less a single career than a neighborhood.
Research: the narrowest AI career path
What the work is. Research scientists and research engineers push what models can do: new training methods, architectures, interpretability, safety techniques. Scientists tend to lead the questions; engineers make experiments run at scale, and the line blurs.
A typical day, as we picture it. Reading papers, babysitting experiments, staring at loss curves, arguing about why a result didn't replicate. Progress is lumpy: a quiet month, then one result changes the roadmap.
How it progresses. Senior, then staff or research lead, with the influence of your ideas doing most of the promoting.
What it takes. The BLS lists a master's degree as the typical entry education for computer and information research scientists, a group with a median pay of $140,300 in May 2025 and projected growth of 22 percent from 2025 to 2035. In frontier research, a PhD is common.
Common is not the same as required. Anthropic's careers page says about half its technical staff have PhDs and about half had no prior ML experience. Its four-month Fellows Program states that you don't need a PhD or published papers, and OpenAI's six-month paid Residency welcomes self-taught candidates. These programs change often, so check their current pages.
Our deeper guide on how to become an AI researcher covers the PhD question in detail. If you're already a PhD student or postdoc, our guide on moving from academia into tech maps where each discipline tends to land.
Engineering: the widest door into AI
What the work is. AI engineers and ML engineers connect models to products: retrieval systems, agents, fine-tuning, evaluation harnesses, and the infrastructure that keeps it fast and affordable. Applied scientists sit between this path and research.
A typical day. Writing code, running evals to see whether yesterday's change made things better or quietly worse, chasing a latency spike. It looks like software engineering, with more doubt about whether the output is right.
How it progresses. Senior, staff, principal, or engineering management. Many arrive from ordinary software jobs.
What it takes. LinkedIn's 2026 list names LangChain, retrieval-augmented generation and PyTorch as top AI engineer skills, and the AI Index's posting data ranks Python as the most requested specialized skill in AI postings. There's no separate BLS category for AI engineers; the closest, software developers, had a median pay of $135,980 in 2025, with the wider software group projected to grow 10 percent from 2025 to 2035.
Our read: this is the path with the most openings and the clearest proof. The Pragmatic Engineer's May 2026 look at the engineering job market found most tech companies putting AI engineering hires ahead of traditional software engineering roles. A working app with an eval set is easy for a hiring manager to judge. Our guide on how to become an AI engineer goes into the skills stack.
Applied and solutions: AI career paths that face the customer
What the work is. Solutions engineers, forward-deployed engineers and AI consultants take a general model or platform and make it work inside one customer's messy reality. Palantir, which helped popularize the forward-deployed model, describes the split between its product engineers and forward-deployed engineers as "one capability, many customers" versus "one customer, many capabilities".
A typical day. A call with a client's operations team, an afternoon wiring the product into their data, a late fix because the demo is tomorrow.
How it progresses. Lead roles, solutions architecture, or a move into product, sales leadership or founding a company. You see what customers actually struggle with, which makes it good founder training.
What it takes. Engineering skills good enough to build alone, plus patience with people who care only about the outcome. AI consultants and strategists ranked second on LinkedIn's 2026 growth list, which suggests demand for this mix.
Product: deciding what an AI product should do
What the work is. AI product managers decide what to build and how to tell whether it's good. That second part is harder with AI than with ordinary software, because the same input can produce different answers.
A typical day. Reviewing eval results, reading tickets where the model got something wrong, trading quality against cost and speed.
How it progresses. Senior PM, group PM, product leadership.
What it takes. Product judgment, comfort with metrics, and enough hands-on time with models to know what they can and can't do. In our view, being able to design a simple evaluation is the skill that most separates AI PMs from generalists. Aman Khan, a product director at Arize AI, made a similar case in a 2025 guest post for Lenny Rachitsky's newsletter, calling eval writing the defining skill for AI PMs. Our guide on how to become an AI product manager covers the path.
Data: the quiet backbone of AI careers
What the work is. Data scientists, engineers, analysts and annotation leads make sure models get good inputs and that someone measures the outputs. Annotation and evaluation work (writing test cases, grading answers) has become a path of its own, often on contract.
A typical day. Writing SQL, cleaning a dataset, reviewing labeled examples for consistency, explaining why a metric moved.
How it progresses. Analyst to data scientist, or data engineer to ML platform engineer. Annotation can lead to evaluation and data operations roles, though pay and stability vary widely.
What it takes. The BLS lists a bachelor's degree as typical entry education for data scientists, with median pay of $120,230 in May 2025 and projected growth of 35 percent from 2025 to 2035. Domain experts (nurses, lawyers, accountants) can find annotation and eval work that values what they already know.
Go-to-market: selling and explaining AI
What the work is. Account executives, sales development reps, customer success managers, marketers and partnerships leads help companies buy and adopt AI products. Selling something probabilistic, priced by usage, to a cautious buyer is a skill of its own.
A typical day. Discovery calls, a pilot review, a pricing conversation, and time in the product so you can demo it yourself.
How it progresses. SDR to account executive to sales leadership; marketers toward product marketing.
What it takes. A track record in sales or marketing plus real fluency with the product. This is, we think, the most overlooked of the AI career paths, and often the fastest one for a non-technical career changer.
Policy and safety: governing what AI does
What the work is. Policy analysts, trust and safety specialists, responsible AI leads, compliance staff and safety researchers decide what systems should and shouldn't do, and check that they behave. Anthropic, for one, lists policy among its main hiring areas.
A typical day. Reading a new regulation, red-teaming a model, writing guidance for product teams.
How it progresses. Analyst to lead to head of policy or trust and safety, often moving between companies, government and research groups.
What it takes. Clear writing, judgment, and depth in law, policy, security or a specific risk area. Technical safety research sits closer to the research path and asks for similar skills.
How to choose an AI career path: five questions
- What problem do you want in your inbox? A failing experiment (research), a broken pipeline (engineering), an unhappy customer (solutions, go-to-market), a fuzzy requirement (product), a messy dataset (data), a hard ethical call (policy).
- How do you like to be judged? By ideas, by shipped code, by revenue, by decisions, or by risks avoided.
- How much more school will you accept? Research often means graduate study. Most other paths reward projects more than degrees.
- What can you prove in 90 days? If you can build a small app in a month, engineering is open. If you can close deals, go-to-market is. Entry-level postings on the AI track of the 1752vc careers board show what a first role on each path asks for.
- What do you already know that AI companies need? A background in health care, law, finance or logistics can be the edge that makes you more valuable than a generic applicant.
A worked example: one person, three candidate paths
An illustrative case. Priya is a financial analyst with solid SQL, a little Python and no ML background. She scores each path from 1 to 5 on enjoying the work, time to credible proof, and how much her finance background helps.
| Path | Enjoys the work | Time to proof | Background helps | Total |
|---|---|---|---|---|
| Engineering | 3 | 2 | 2 | 7 |
| Data | 4 | 4 | 4 | 12 |
| Applied and solutions | 4 | 3 | 5 | 12 |
Two paths tie at 12. One tie-breaker: start where proof comes soonest (data), then step toward solutions work at a fintech AI company. The scores are hers, not a formula.
"Just pick the highest-paying path"
There's a fair case for this. Pay differences are real, and if money is your main constraint, it's reasonable to weigh it heavily.
But.
The highest-paying paths tend to have the steepest entry bars, and a path you dislike is hard to stick with long enough to reach the pay. In our view, picking the path you can enter and stay on usually beats aiming at the top of a ladder you can't reach yet. The 1752vc salary guide can help you compare once you've narrowed the list.
Where to find roles on each path
AI companies often post first on their own careers pages; our roundup of AI job boards covers where to look. The 1752vc careers board has an AI jobs track listing open roles at AI companies, refreshed every week with each employer's own posting date. Read 20 postings for each of your two candidate paths; the repeated skills tell you what to build. Filtering to the past 7 days shows which path is hiring now.
Common mistakes when choosing an AI career path
- Picking by title. "AI engineer" means different things at different companies. Read the responsibilities.
- Starting with the hardest door. Research gets the attention, but it is one path of seven.
- Hiding your old career. Domain knowledge is often your differentiator.
- Collecting courses instead of proof. One finished project tends to say more than three certificates.
Where we land
Our one suggestion: pick the path whose daily problems you'd happily solve on a slow weekend, then check you can show proof for it within a few months. Title, pay and prestige tend to follow from staying on a path long enough.
That's our answer, not the only one. Your finances, location and visa situation may change the math.
The bottom line
AI has more doors than its reputation suggests.
Choose the one you'd walk through twice.
Key takeaways
- The main AI career paths are research, engineering, applied and solutions, product, data, go-to-market, and policy and safety.
- Growth data from LinkedIn, the Stanford AI Index and the BLS points to demand across several paths, not only research.
- A PhD is common in research but many other AI paths hire mainly on projects and domain knowledge.
- Choosing by daily work, proof you can show soon, and your existing background tends to work better than choosing by title or pay.
Frequently asked questions
Go-to-market and policy paths are usually the most open to non-technical people. Sales, customer success, marketing, trust and safety, and compliance roles at AI companies value domain knowledge and communication, plus hands-on fluency with the product. Data annotation and evaluation work can also be a starting point for subject-matter experts, though its pay and stability vary.
Most of them. Engineering, solutions, product, data, go-to-market and most policy roles hire mainly on skills and proof of work. The BLS lists a bachelor's degree as typical entry education for data scientists. Research is the exception where graduate degrees are common, though some labs state that a PhD is not required for their research programs.
Yes, and many people do. Common moves include software engineering to AI engineering, AI engineering to solutions or product, data analysis to data science, and solutions work to sales leadership or founding a company. Skills overlap between neighboring paths, so a switch often means adding one new skill rather than starting over.
By LinkedIn's Jobs on the Rise 2026 list for the US, AI engineer was the fastest-growing job title, with AI consultants and strategists second. That list measures growth in hires among LinkedIn members from 2023 to mid-2025. Other datasets measure demand differently, so treat any single ranking as a signal rather than a forecast.
Sources
- US Bureau of Labor Statistics: Computer and Information Research Scientists, Occupational Outlook Handbook
- US Bureau of Labor Statistics: Data Scientists, Occupational Outlook Handbook
- US Bureau of Labor Statistics: Software Developers, Quality Assurance Analysts, and Testers, Occupational Outlook Handbook
- LinkedIn: Jobs on the Rise 2026, The 25 fastest-growing roles in the U.S.
- Stanford HAI: AI Index Report 2026, Chapter 4: Economy
- The Pragmatic Engineer: State of the software engineering job market in 2026
- Lenny's Newsletter: Beyond vibe checks, A PM's complete guide to evals (Aman Khan)
- World Economic Forum: The Future of Jobs Report 2025, Digest
- Anthropic: Careers
- Anthropic Alignment Science: Anthropic Fellows Program for AI safety research
- OpenAI: Residency
- Palantir Blog: Dev versus Delta, Demystifying engineering roles at Palantir
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.
