How to Transition Into an AI Career: Complete Guide

Say you've spent years in sales ops, or support, or finance. AI jobs keep showing up in your feed, and you keep wondering which one you could actually get.

Your old job gives you something most candidates don't have: you know the problems in your field. It doesn't give you the technical skills the new role asks for. Moving into AI work means being honest about both.

This guide helps you pick one of six roles based on the work you do now, learn things in the right order, build one project that proves it, and read what employers actually post as pay. Pay figures come from postings opened in September 2026 and listed on the site's career pages.

Pick a target role before you pick a course

"Learning AI" isn't a goal. Each role below asks for different proof, so choose one before you spend a dollar or a weekend. Start from what you already do well.

If you don't write code yet and want an engineering role, that is your first gap. The free foundations guide orders the free courses. If you'd rather not become an engineer, the sales and GTM routes reward commercial experience first.

Every role has a career page with a reading order, posted pay and practice interview questions. The careers hub lists them.

What employers ask for

AI engineer topped LinkedIn's list of fastest-growing US job titles in 2025, and it topped the list again in 2026. Job postings that ask for AI skills pay about 28 percent more than equivalent postings that don't, according to Lightcast's July 2025 analysis.

Demand doesn't lower the bar. Postings ask for evidence, not interest. Baseten's AI Engineer posting wants "demonstrated experience building AI or LLM-powered products, agents, or agentic workflows that real users depend on." A course certificate doesn't answer that. A finished project in a field you already know does.

What to learn, in order

This order is for the engineering roles. Each stage ends in something you can show. How long it takes depends on where you start and how many hours you have, so plan by stage, not by month.

Stage 1: Programming fundamentals, if you need them. If you don't already write code, learn Python through a project-driven course such as Harvard's free CS50P. Get comfortable with git, virtual environments, REST APIs and the command line. Goal: a small command-line tool on GitHub that solves a real problem from your current job.

Stage 2: Calling a model and checking the answer. A model will sometimes send back something broken or wrong, so your code has to check every answer before using it. The site's AI APIs lesson sends a request, checks the response and tests what happens when it fails. Prompt contracts with evaluation fixtures turns "the answer looks right" into checks that pass or fail. Goal: a script that calls a model and rejects answers in the wrong shape.

Stage 3: Retrieval, tools and evaluation. Build a small RAG pipeline, which is code that finds the right documents and hands them to the model. Make it check who's allowed to see a document before it searches. Then let a model ask to take actions, like looking up an order, that your code approves or blocks. Finally, score the results honestly, counting every failed run in the total. Goal: an eval set, meaning a saved list of test cases with known right answers, that you can rerun after every change.

Stage 4: One project in your domain. Pick a problem from the field you know and build it end to end, with the evidence described below. Goal: a public repository with a README, an eval report and a short demo.

Stages 2 and 3 map onto the site's 17-lesson agent course. The labs run locally on synthetic data, so none needs a paid API key to reproduce its recorded result.

For sales roles, the order is different: learn the product category, the buyer and how an AI deal runs. What is different about selling AI and the 90-day plan in breaking into AI sales cover it.

Use your old job, don't hide it

Don't present yourself as a fresh entry-level engineer. A former salesperson who builds an agent that cleans up a CRM already knows the problem that other candidates would have to learn on the job. That knowledge doesn't replace the technical skills the role asks for, so name both: what you know about the work, and what you're still learning.

Your years in sales, finance, healthcare, marketing or operations are where you'll find problems you understand well enough to solve. You're not starting over in a new field. You're building AI for the problems you already know.

Build one project a reviewer can check

One finished project with evidence tells a reviewer more than several demos. Choose the type that matches your target role:

1. An agent project: an agent that takes several steps, uses tools, remembers what it did between runs, and has a real interface, ideally against a real API like Slack, Notion or Stripe. Fits agent engineer and FDE roles.
2. A retrieval project: answering questions over a real set of documents, showing how you split them up, how you tested the search, and that each answer points to its source. Fits AI engineer roles.
3. A revenue-systems project: a signal that becomes a reviewed action in a CRM, with the data model written down. Fits GTM engineer roles, and the GTM engineer guide specifies one.
4. A vertical project: something specific to your domain, such as a sales-call summary agent, a financial model audit tool or a medical chart abstraction pipeline. This is the one that uses your experience, so make it your first choice if it fits one of the types above.

Whichever you pick, the project needs the same evidence: a written description of the task that says when the system should refuse, a saved eval set, separate scores for each kind of failure, a starting score and one measured improvement, and a list of what still fails. The AI engineer path lists each piece and what it shows a reviewer.

Put it in a public repository with the README first and the demo second. A second project can wait until the first one is finished.

A worked example

Here's a made-up case to show how the pieces fit. Dana has spent five years as a claims adjuster at an insurance company. She knows which claims get stuck and why. She can write basic SQL, but not Python.

She wants an engineering role, so code is her first gap. She starts with stage 1 and builds a small command-line tool that sorts a folder of claim notes by type. It's not impressive, but it's real and it's hers.

After stages 2 and 3, she picks a vertical project: a tool that reads a claim file and lists which required documents are missing. She writes down when it should refuse to answer, like when a scan is unreadable. She builds an eval set from old claims where she already knows the right answer, scores it, fixes the most common failure, and scores it again. The README says what it still gets wrong.

When she applies, her paragraph doesn't say "career changer." It says she spent years watching claims stall on missing paperwork, then built a tool that catches it, with the scores to show how often it's right.

Read pay from postings, not surveys

Salary surveys blend titles that do different work. Employer postings are narrower and checkable. The ranges below come from postings the site's career pages list, opened on September 26, 2026 (September 17 for the forward deployed engineer rows). They are not labeled the same way: some are base salary, some are on-target earnings (OTE), and some don't say. Postings close, so open the link before relying on a number.

A posting's experience bar tells you more about the level you'll be hired at than your previous title does. Baseten's AI Engineer posting, for example, asks for five or more years building and shipping software. If you are switching fields, expect to be read against that bar in the new field, and let the project and your domain story carry the rest. How OTE and sales pay plans work explains the sales rows.

Find the openings

Three habits make a search shorter.

1. Read responsibilities before titles. "Applied AI Engineer" is a product role at one company and a customer-facing deployment role at another. The directory of companies hiring forward deployed and GTM engineers and the directory of companies hiring AI sales roles list checked postings with their scope.
2. Ask for introductions with the project attached. A referral request that links a finished project and a one-line reason you fit the team is easy to forward.
3. Write about what you build. A short public post per project shows how you think about failure, which is what the eval report is for. It also gives a referrer something to send.

Three ways a switch stalls

The first is finishing course after course without ever shipping a project. The second is hiding your old experience to look like a junior engineer, which throws away your biggest advantage. The third is aiming at a role your background doesn't fit yet, such as research roles without a research record, or a title you can't find in current postings.

Next steps

• Pick your row in the role table and open that role's career page from the careers hub.
• If you need programming fundamentals, start with the free foundations guide.
• Work through the agent course stages that match your target, then build the project.

Related Guides

• Agent Engineer: A Practical Career and Portfolio Guide
• AI Engineer: A Practical Career and Portfolio Guide
• GTM Engineer: A Practical Career and Portfolio Guide