What should you do about your career and AI?

You do not need a completely new identity because AI is changing work. Begin with a simpler question: what is one realistic step that would improve your options? The answer depends on whether you need income now, already have a job, want to change direction or are responsible for a team.

If money is tight, keep current applications, paid training routes and public employment support first. If you are working, ask what tools are permitted and whether you can have paid learning time. If you have room to explore, choose one small practice project. Caring responsibilities, health and access needs should shape the pace.

The companion journey asks about those circumstances, your AI experience and the work you want to explore. It gives you one next step—not a score or a career picked for you. Change an answer whenever your situation changes.

AI is software that performs tasks associated with intelligence. Generative AI can draft text, images or other output from your instructions. Those instructions are often called a prompt. You do not need to code to try a basic exercise.

The first habit to learn is checking. NIST identifies confidently presented false answers as a generative-AI risk. A polished answer can still invent facts, and asking the same tool whether it is right is not an independent check.[3]

Try the short exercise in the journey: read an information sheet, compare a draft answer and spot what is unsupported. It needs no account. If you want a course afterward, start with the opening chapter of Elements of AI, not a whole stack of resources.

Use public or fictional material. Do not upload private, customer or employer records to an unapproved tool. A tool allowed to draft should not automatically be allowed to send messages, spend money or change records.

Choose something you can understand and check: a visitor information sheet, a comparison of two roles or a simple checklist. Write down who it helps, which sources it uses and what a correct result looks like. If you cannot judge the answer, narrow the task or ask someone qualified to help.

Make a first attempt yourself. Then, if useful, compare an AI-assisted attempt. Keep the original information, your corrections and the time spent checking. The aim is to learn the work—not just produce something that looks finished.

A course is something you studied. A sample shows a bounded piece of work. An offer depends on an employer and the market. Keep those separate so that an unanswered application does not automatically send you back to buy another certificate.

• Make one small sample. Explain what you did, how AI helped and one mistake you corrected. Label fictional work as practice.
• Ask for feedback. Show it to someone familiar with the work and ask what is wrong, missing or useful. Make a revision.
• Check a paid route. Compare current listings or ask an employer about requirements, pay, supervision and training.
• Choose what happens next. Apply, deepen one skill, explore a nearby role or pause. Set a limit on unpaid work and course spending.

Research does not give us a reliable list of safe careers. The ILO warns that estimates of tasks AI could affect are not direct predictions of job losses or individual training needs.[4]

Results also vary by setting. One customer-support study found 15% more issues resolved per hour with assistance, with larger gains for less-experienced workers. A Danish study found little average change in earnings or recorded hours during the first two years after ChatGPT. They answer different questions in different places.[6][5]

Newer does not always mean more certain. METR’s February 2026 developer follow-up pointed toward speedups, but selection and measurement problems made the estimates unreliable. That is a reason to check the work you actually do, not rely on a headline.[7]

AGI, or artificial general intelligence, has competing definitions. One research framework separates breadth of ability, level of performance and autonomy. Being strong on one test does not settle all three.[2]

In our optional scenarios, AGI means broadly skilled-human-level cognitive ability; superintelligence means substantially exceeding human cognitive ability across domains. These are hypothetical assumptions, not an arrival date. Cheap, reliable deployment is a separate question.

The International AI Safety Report describes uncertainty about future capability and labor-market effects. If systems surpass people across nearly all tasks, employment and wages could fall; it is uncertain whether enough new human work would emerge.[1]

No personal learning plan can guarantee income in that scenario. Employer transition support, worker voice, public services and how the gains are shared would matter too. Your worth is not your ability to outperform a machine. For today, make a useful, affordable decision and keep connections to people and institutions that can help.

After trying a step, return to the journey and say what happened. Stuck? It suggests a smaller task or someone to ask. Made a sample? It points to feedback. Found no opportunity? It asks you to check demand before doing more training.

Review your direction when your work changes, a trial fails, local openings change or the plan no longer fits your life. You are not trying to finish adapting forever. You are making the next decision better informed and easier to change.