The summer upskilling plan for professionals who feel left behind by AI

A few months ago a project manager told me, almost apologetically, that she felt like the train had already left the station. Everyone around her was "doing AI," her LinkedIn feed was a wall of people announcing they'd become prompt engineers overnight, and she'd quietly concluded she was too late and too non-technical to catch up.

She wasn't too late. She wasn't even behind. She'd just confused using AI well with building AI, and those are completely different jobs.

Let me say the reassuring part first, because the rest of this post depends on you believing it: you almost certainly do not need to become a machine learning engineer. The world needs a small number of people who train models. It needs an enormous number of people who can use these tools confidently, catch them when they're wrong, and wire them into real work. That second group is wide open, and you can join it in a summer of evenings.

The skill that's actually scarce

Here's the thing nobody puts on a course brochure: the bottleneck isn't access to AI. Anyone can open a chat window. The bottleneck is judgment — knowing what to ask, whether to trust the answer, and where it fits in a real workflow.

I think of it as a ladder. Most people are told to jump straight to the top rung ("learn to build agents!") when the leverage is on the lower rungs.

The interesting part: rungs 2 and 4 — data literacy and judgment — are the least automatable and the most valuable, and they're the ones the "learn AI in a weekend" crowd skips entirely. A tool can write the prompt for you. It cannot decide whether the output is good enough to send to a client. That decision is your job, and it always will be.

Aim to become the person who uses AI critically, not the person who builds it. That's where the demand is, and it's a far shorter climb.

What's genuinely in demand (and what's noise)

If you strip out the hype, the durable, hireable skills around AI right now tend to cluster in three honest places:

• Data and SQL. Every AI conversation eventually hits the same wall: what's in your data, and is it clean? People who can pull, join, and sanity-check data are useful in every department. This is not glamorous and it is not going away.
• Workflow automation. Not "build an autonomous agent." More like: the report that took you three hours every Friday now takes twenty minutes because you stitched a few steps together. Multiply that across a team and it's real money.
• AI oversight and review. As organisations let AI draft, summarise, and recommend, someone has to check it. The person who can confidently say "this output is wrong, and here's why" is increasingly the person who keeps the lights on.

Notice that none of these require you to understand transformer architecture. They require clear thinking, some comfort with data, and the willingness to verify. If you want the deeper-end map of how the model side actually works — for curiosity or because your role genuinely needs it — I've laid that out separately in my 2026 machine learning roadmap. But for most professionals, that's optional reading, not the main quest.

An 8-week evening plan around a full-time job

This is built for a real life: a full day of work, dinner, and maybe three or four hours a week left over. Not forty. The whole point is that small, consistent effort beats a heroic weekend you never repeat.

The single most important rule: practise on your own job's real tasks. Generic tutorials evaporate from your memory by morning. The report you actually have to write on Thursday does not.

Weeks 1–2: Prompting that isn't a guessing game

Pick three tasks you do every week — drafting an email, summarising a document, writing a first pass of something. Do each one with an AI tool, but force yourself to iterate at least twice instead of accepting the first answer. Notice what context you had to add to get something usable. That "context you had to add" is the actual skill.

Checkpoint: you can take a vague request and turn it into a prompt that gets a usable first draft in two or three tries.

Weeks 3–4: Data literacy

Find a spreadsheet from your own work. Spend these weeks getting genuinely comfortable reading it: what each column means, what a given number is actually measuring, where a percentage could mislead. Ask an AI tool to explain a formula or summarise the data — and then check it against the raw numbers yourself. The checking is the lesson.

Checkpoint: you can look at one of your own reports and explain, out loud, what's solid and what's a shaky assumption.

Weeks 5–6: Automate one annoying thing

Choose the most tedious recurring task you own. Don't aim for clever — aim for done. Maybe it's a recurring summary, a tidy-up of messy data, a templated response. Build it, use it for real, and time the before-and-after. This is where the climb pays for itself, and it's the kind of thing I dig into on the AI & automation tutoring track.

Checkpoint: one real task in your week is now meaningfully faster, and you can describe how.

Weeks 7–8: Judgment and a small portfolio

Spend the last fortnight deliberately trying to catch AI being wrong on your domain. Ask it things you already know the answer to and watch where it drifts. Then write up — just for yourself — two or three short before/after stories: here's a task, here's how I used AI, here's what I checked, here's the result. That's your evidence. In a performance review or an interview, "I cut our Friday report from three hours to forty minutes and here's how I verified the numbers" beats any certificate.

Checkpoint: you have two or three concrete examples of AI making your actual work better, and you can explain the verification you did.

Why "use your own work" matters more than any course

There's a quiet trap in online courses: they feel like progress. You watch the videos, you nod, the dashboard fills with green ticks — and three weeks later you couldn't reproduce a single thing because none of it touched a problem you actually cared about.

Anchoring every week to a real task from your own job fixes this. The task supplies the motivation (you have to do it anyway), the realistic messiness (your data is never as clean as the tutorial's), and the memory hook (you'll remember the report you sweated over). It also means that at the end of eight weeks you haven't just "learned about AI" — you've changed how a few real things in your week get done.

If you want the version of this aimed squarely at working people balancing this against an actual career, the professionals page lays out how I tend to structure it.

When a course is fine, and when 1-on-1 actually beats it

I'll be straight with you, because pretending otherwise would be exactly the hype I'm trying to avoid: for a lot of this, a good course or even free material is perfectly fine. If your gaps are broad and you're disciplined, go for it.

One-on-one earns its keep in three specific situations:

• Targeted gaps. You don't need the whole syllabus — you're stuck on one or two things, and a course makes you sit through forty hours to reach the ten minutes you needed.
• Accountability. You already know the hard truth: most self-paced courses get abandoned around week three. A standing weekly session is a commitment device. Someone is expecting you.
• Your actual domain. A generic course teaches generic examples. Working through your spreadsheet, your workflow, your messy real data is where it clicks — and that's precisely what a tutor can do that a pre-recorded video never will.

If none of those apply, save your money and use the plan above on your own. If they do, that's the honest case for working with someone.

The recap

• You don't need to become an ML engineer. The scarce, in-demand skill is confident, critical use.
• Climb the ladder in order: prompting → data literacy → automation → judgment. The top two rungs (data and judgment) are the most valuable and the least automatable.
• In-demand and honest: data/SQL, workflow automation, AI oversight. Hype-driven and fragile: anything that promises you'll "build AI" in a weekend.
• Run the 8-week evening plan on your own job's real tasks. Small and consistent beats heroic and abandoned.
• At the end, you want two or three concrete before/after stories — not a certificate.

The project manager I mentioned? She didn't learn to train a model. She learned to use the tools well, check them carefully, and automate one painful weekly report. That was enough to stop feeling left behind — because she no longer was.

If you'd rather not figure out which rung you're actually stuck on by yourself, that's exactly the kind of thing a few focused 1-on-1 sessions can sort out quickly — no forty-hour course required.