What to learn after your first month with AI (and which AI jobs are actually real)

TL;DR: For most people the highest-return path is not becoming an AI specialist — it is becoming the person in your existing field who uses AI unusually well. That is scarcer, more defensible, and considerably easier to reach from where you are. If you do want AI-specific work, the roles that reliably exist are narrower than the discourse suggests.

The fork

After a month of daily use you are at a genuine decision point:

• Path A — depth in your field. Stay in your industry, become the person who has rebuilt how the work gets done. Lower risk, faster payoff, more defensible than it sounds.
• Path B — toward AI work. Move into roles where AI is the product rather than the tool. Longer, more competitive, and requires genuine technical development for most of the good roles.

Most advice assumes you want B. Most people are better served by A, and the two are not mutually exclusive — A is a reasonable route into B.

Path A: become the person who uses it unusually well

The scarce combination is not “knows AI”. It is “deeply understands a domain and uses AI unusually well”. There are many people with each half and few with both, which is exactly what makes the combination valuable.

What to actually do, in order:

1. Map your own work. List every recurring task you do. Mark each: fully automatable, assistable, or human-only. Most people find 20-30% is assistable and they were only using AI on 5%.
2. Rebuild your three biggest time sinks as documented workflows — not ad hoc prompting, but a repeatable process with saved instructions and a quality check. This is the difference between using AI and having a system.
3. Teach it. Write up one workflow for colleagues. Teaching forces precision, and it is also how people come to know you are the person who does this.
4. Solve one problem that is not yours. Find something a different team does badly and rebuild it. This is where the reputation compounds — and where it starts showing up in performance reviews and job offers.
5. Learn the adjacent technical skill your domain rewards: SQL if you work with data, automation platforms if you work in operations, basic scripting if you work with files.

Timeline: three to six months to become visibly the person people ask. That is a genuinely different career position, reached without changing jobs.

Path B: which AI jobs actually exist

Being honest about the market matters more here than encouragement, so:

Roles that reliably exist and are hiring:

• ML / AI engineer. Builds and deploys models and the systems around them. Requires real software engineering plus ML. This is a multi-year path from zero, and there is no shortcut. Well paid because it is genuinely hard.
• Data engineer. Unglamorous, in constant demand, and the actual bottleneck at most companies trying to do anything with AI. Far more accessible than ML engineering and arguably better job security.
• AI product manager. Decides what to build and whether it works. Needs product judgement plus enough technical literacy to know what is feasible. Very reachable if you already do product or adjacent work.
• AI implementation / solutions roles. Getting AI working inside organisations — integration, change management, workflow redesign. Growing quickly, values domain knowledge, and the most natural landing spot for someone arriving from Path A.
• Evaluation and safety roles. Testing whether systems behave acceptably. Small but real, often values domain expertise and rigour over coding.

Roles that are less of a thing than the internet suggests:

• “Prompt engineer” as a standalone job. Briefly a real title, now largely absorbed into other roles — because it turns out to be a skill everyone needs rather than a job someone holds. The skill is valuable; the job title is mostly not. Covered honestly in the prompt engineering piece.
• “AI consultant” with no domain expertise. Consulting sells credibility. Credibility comes from having solved the problem in a real context. Consulting on AI in general, from a month of experience, is a difficult sell to anyone competent.
• AI content operations at scale. Producing large volumes of generic AI content is a business with a shrinking margin and rising platform hostility — see whether AI content ranks.

If you are going down Path B from zero

An honest sequence, assuming you are starting without a technical background:

1. Months 1-3: Python properly. Not “Python for AI” — Python. Data structures, functions, files, errors, libraries. Everything else builds on it.
2. Months 3-5: Data. SQL and working with real, messy datasets. This alone qualifies you for data roles and is prerequisite for everything else.
3. Months 5-8: Build things that use models via APIs. Retrieval over documents, a working tool someone else uses. This is where you learn what actually breaks in production, which is the thing employers are testing for.
4. Months 8-12: Depth in one direction — ML fundamentals, or data engineering, or productionising systems. Choose based on what the jobs you want ask for, read from actual job postings rather than from advice.
5. Throughout: build a portfolio of things that work. Three finished projects that solve real problems beat any certificate. Employers hiring for these roles can tell the difference within one conversation.

Realistically twelve to eighteen months of consistent part-time effort to become employable, longer without a technical background. Anyone quoting three months is selling a bootcamp.

How to read the job market yourself, instead of trusting anyone’s list

Including mine. Job markets move faster than articles, and the honest method takes about an hour.

1. Pull 20 real postings for roles you would want, from actual company career pages rather than aggregators, which are full of stale and duplicated listings.
2. Paste the requirements into a model and ask: “Across these 20 job descriptions, what skills appear in more than half? What appears in fewer than three? Separate genuine requirements from boilerplate that appears in every posting regardless of role. What is the most common combination of skills, rather than the most common individual skill?”
3. Look at what is absent. If a skill you were planning to spend three months on appears in two of twenty postings, that is your answer.
4. Check the seniority reality. Ask how many of the twenty are genuinely open to someone with under two years in the field. Frequently the answer is one or two, and knowing that before you start is worth more than any encouragement.
5. Repeat quarterly. This market is moving; a plan built on last year’s postings is a plan for last year.

This exercise does two things at once. It gives you an accurate, current map instead of a second-hand one — and it is itself a demonstration of the skill you are trying to build, which is using these tools to answer a real question you actually have rather than to practise on an invented one.

It also protects you from the most expensive mistake available here: spending nine months learning the thing that was in demand when the article you read was written.

What holds value regardless of path

• Judgement about when not to use it. Increasingly the differentiator, as capability commoditises.
• Verification instinct. Knowing what to check, and checking it. Most people never develop this.
• Domain depth. The thing models are worst at is knowing what is true in your specific context.
• Clear writing. Both because you have to specify what you want, and because the ability to think clearly in prose is not automatable in the way people assumed.

The realistic take on which jobs change and how fast is in the job displacement piece, and the skills currently commanding money in AI skills that actually pay.

One closing thing

The people getting the most out of this are, almost without exception, the ones who used it constantly on real work for months before they had any theory. If you are choosing between another month of daily practical use and a course, choose the practice. Start at the first 30 days if you have not already.