Hey Freelance Friends!

In April 2025, the OECD published a policy brief with a question buried inside it that most people skimmed past: Is training keeping up?

Their answer was polite. The kind of language bureaucrats use when they want to say something alarming without causing a scene. Current training systems, they said, "may not be sufficient" to meet growing demand for AI literacy.

May not be sufficient. I want you to sit with that for a second.

Because here in South Africa, we already knew this. We have been living the "may not be sufficient" version of formal education for a long time. The OECD just finally put it in a brief.

The pipeline broke. And most people haven't noticed yet.

The old model went like this: study, get certified, get hired, build a career. It worked when knowledge changed slowly. When what you learned in year one was still relevant in year five. When your degree was a reliable signal that you knew something useful.

That model had a quiet assumption underneath it, that institutions could identify valuable knowledge early enough, and that the knowledge would stay relevant long enough to justify years of training.

AI just snapped both of those assumptions in half.

According to OECD data, firm-level AI adoption across member countries more than doubled between 2023 and 2025, from 8.7% to 20.2%. Among large firms, it hit 52%. More than one-third of people across OECD countries had already used generative AI tools by 2025. Among students aged 16 and older, that number reached roughly three-quarters.

Universities approved courses on prompt engineering while the underlying models were already being replaced.

A degree designed in 2023 hit the labour market in 2026 and found parts of itself obsolete.

This is not an argument against universities.

Universities still do things nothing else can do, research, professional certification, scientific training, medicine, engineering, law. That is not what's collapsing.

What's collapsing is the idea that formal education has a monopoly on preparing people for digital work.

It doesn't. It hasn't for a while. We just didn't have a name for what was replacing it.

The internet became the backup system. Nobody announced it.

Right now, millions of people are quietly building parallel education systems for themselves. Not through expensive courses. Through YouTube tutorials that replace outdated coursework. Discord servers that outperform career centres. Reddit threads explaining new AI workflows before universities publish official guidance.

Freelancers reverse-engineering income systems through trial and error because no formal pathway exists yet.

The internet became an emergency vocational layer sitting on top of slower institutions. And the people who figured this out earliest, often people who were already excluded from or failed by formal systems, are now quietly ahead.

That includes a lot of us.

Here's the part that should concern you though.

There's a difference between having access to AI tools and being able to use them well. Researchers are calling this an AI literacy problem, and there's still no consensus on what AI literacy even means operationally.

What they do agree on is this: the interface looks simple. The skill requirements are not.

Knowing how to verify outputs. Identifying hallucinations. Structuring prompts. Combining AI outputs with real domain knowledge. Understanding where automation fails and why.

These are not technical skills. They're cognitive and procedural skills. And right now, most people building them are doing it through experimentation, not coursework.

A freelancer who has spent six months testing AI workflows for client research may have more practical operational understanding than someone who completed a formal course built around older models.

Not because informal learning is better in principle. Because the feedback loop is faster.

Entry-level workers are taking the hit first.

A Stanford-linked study analysing payroll data found employment declines among younger workers in AI-exposed industries, software development, and customer support between 2022 and 2025. Older workers in the same sectors appeared more insulated.

The evidence is still early. But the direction is clear.

Entry-level positions historically worked as training infrastructure. You learned by doing the repetitive, lower-risk work before being trusted with the complex stuff. AI is automating that entry layer first.

Which means the standard advice, "go gain experience" is quietly losing the mechanism that made it work.

This is why more and more people are bypassing formal entry pipelines entirely. Building public portfolios. Independent projects. Freelance histories. Demonstrated competence over institutional sequencing.

The labour market is shifting toward rewarding what you can show, not just what you can prove you sat through.

What this actually means for how you learn.

The research points to five things that work in unstable skill environments:

Build around real projects, not abstract knowledge. Don't "learn AI." Automate your invoice summaries. Build a research workflow. Create AI-assisted proposals. The project is the curriculum.

Treat your information sources unequally. Official documentation, academic research, experienced practitioners, reproducible workflows, these earn trust. Viral productivity influencers and unverifiable income claims do not. The AI economy currently produces enormous amounts of synthetic expertise.

Keep a public record of what you can do. Credentials still matter in some contexts. But visible proof of adaptation increasingly matters more. A portfolio, documented experiments, published analysis, freelance case studies, these function as evidence of competence in a market that's moving too fast for static credentials to keep up.

Learn in small continuous cycles. Rapid environments punish long periods of passive preparation. Frequent exposure matters more than perfect sequencing. This is closer to how you learn a language than how you sit an exam.

Build networks alongside skills. Communities share opportunities, warn about scams, explain platform changes, compare workflows. In fragmented digital economies, community knowledge is increasingly substituting for institutional guidance.

The official system still exists.

But beside it, another system has emerged. Faster, unstable, decentralised, inconsistent, often chaotic, but highly responsive to immediate economic reality.

Millions of people are already using it.

A lot of them didn't choose it strategically. They chose it because the other option wasn't available, wasn't fast enough, or wasn't built for where they are.

That's not a failure story.

That's actually the beginning of an advantage, if you're intentional about it.

~ Profreelance