Do I need to learn to code to work with AI?

For some directions yes, for several no. Automating a workflow, designing how a tool gets used, evaluating outputs and writing the material a model works from are all real work that starts without code. Find out whether you like the domain before deciding how deep to go technically.

What is a good first AI project for a beginner?

Take one task you already do repeatedly, get a model to do the first eighty per cent of it, then fix what it got wrong and write down what you had to fix. That list is the actual skill, and it takes an evening.

Is prompt writing a durable skill?

The specific tricks are not durable; models change. What is durable is being able to describe a task precisely, judge whether an output is good, and build a repeatable process around it. Test the work, not the job title.

Should I do an AI course or a project first?

A project first, almost always. A course is a good answer once you know which direction you are testing, and a poor way to discover that direction.

AI directions · 9 min read

How do I explore an AI career before choosing one?

AI work is not one job. It is a dozen quite different kinds of day. Here is how to find out which of them you would actually enjoy, using short projects instead of a programme.

Written by the OffLadder editorial team · Last reviewed 29 August 2026 · Sources listed at the foot of this page

The short answer

Treat an AI career as a set of directions to test, not a single destination to enrol in. There are at least half a dozen very different kinds of AI work, and you can do a small real version of most of them in an evening with free tools. Pick two that sound different from each other, build something tiny in each, then judge them by which parts you wanted to keep doing. Do that two or three times and you will know more about fit than any course description can tell you, and you will have made things you can show.

AI work is at least six different jobs

Automating boring processes: gluing tools together so a task runs itself. Practical, unglamorous, in demand almost everywhere.
Building with models: making a small app or tool that calls a model and does something useful.
Judging outputs: designing how you check whether a model is right, and catching where it fails.
Data work: getting messy real information into a shape a model can use.
AI plus a field you already know: health, law, teaching, logistics, music. Often the strongest position.
Helping people use it: training, workflow design, deciding what a team should and should not hand over.

Explore, try, learn, adapt

01Explore: pick two of the six above that feel genuinely different from each other. Curiosity is enough of a reason.
02Try: give each one a twenty to sixty minute version. Something with real output, however small.
03Learn: write two sentences per attempt. Which part pulled you in, which part you tolerated.
04Adapt: keep the direction that pulled, drop the other without ceremony, and pick a new second one.

You are not choosing an AI career. You are finding out which parts of it you would do on a Tuesday.

Two experiments you can do this week

First: automation. Choose one repetitive thing in your week, such as tidying notes into a summary or sorting incoming messages. Write out the steps as if explaining them to a new colleague, hand them to a model, and run it on real material. Note every correction you had to make.

Second: AI plus your own field. Take a small task from something you already understand, whether that is a school subject, a sport, a job or a hobby, and build a tiny helper for it. A revision quiz generator, a checklist, a summariser. Show it to one person who does that thing and write down what they said.

Afterwards compare them honestly. Most people find one of those two hours flew and the other dragged, and that difference is worth more than a term of research.

Questions people ask

Do I need to learn to code to work with AI?: For some directions yes, for several no. Automating a workflow, designing how a tool gets used, evaluating outputs and writing the material a model works from are all real work that starts without code. Find out whether you like the domain before deciding how deep to go technically.
What is a good first AI project for a beginner?: Take one task you already do repeatedly, get a model to do the first eighty per cent of it, then fix what it got wrong and write down what you had to fix. That list is the actual skill, and it takes an evening.
Is prompt writing a durable skill?: The specific tricks are not durable; models change. What is durable is being able to describe a task precisely, judge whether an output is good, and build a repeatable process around it. Test the work, not the job title.
Should I do an AI course or a project first?: A project first, almost always. A course is a good answer once you know which direction you are testing, and a poor way to discover that direction.

Sources

Future of Jobs Report 2025 · World Economic Forum
LinkedIn Economic Graph research · LinkedIn

How this page is written

We only state what can be checked. Where a page makes a claim about work, automation or the labour market, it links to a primary source rather than paraphrasing a secondary summary. We do not publish predictions dressed as facts, invented statistics, or lists of AI-proof jobs. Pages are reviewed and dated when the argument or the underlying evidence changes. More about OffLadder.