We're still early in AI's emergence. Capabilities are growing fast, and it's fair to wonder how the job market will react.
Some jobs will go away. Many more will change dramatically. I don't buy the mass-unemployment crisis story though. Markets need both buyers and sellers. If large chunks of the workforce lose purchasing power overnight, demand collapses with them. Sellers need a stable market as much as buyers do. Extreme disruption that hollows out the customer base is bad for everyone — including the companies shipping the tools.
The more interesting shift is quieter, and I think it's under-discussed.
Knowledge was already cheap. Skill wasn't.
The internet already turned most knowledge into a commodity. Documentation, tutorials, Stack Overflow, courses, open source. If you could search, you could find the answer.
What stayed scarce was the skill to apply it. Judgment under constraints. Knowing which answer mattered for this system. Taste for trade-offs. The ability to ship something that survives contact with production.
AI changes that. Not perfectly, not everywhere — but enough that a lot of applied skill is getting cheaper and more accessible. Prototypes appear faster. Boilerplate disappears. Junior-shaped work gets compressed.
So if skill becomes easier to rent from a model, what stays scarce?
Know-how might get expensive again
I'd expect companies to sit tighter on their proprietary knowledge. Processes that actually work. Domain heuristics. Data that isn't on the public web. The messy operational truth you only learn by running the business.
When public knowledge plus applied skill is abundant, private know-how becomes the moat.
That doesn't mean secret PDFs in a vault. It means:
- institutional memory that isn't written down cleanly
- decision frameworks tuned to a specific market
- customer and operational data used as context, not just analytics
- the judgment of people who've seen this failure mode before
In my view, that's also where a lot of new work shows up. Not "prompt engineer" as a job title fad. Roles around capturing, protecting, structuring, and applying proprietary knowledge so AI systems can use it without leaking it. Knowledge ops for real businesses. Context engineering for products that can't afford to be generic.
What this means if you're building a product
If I were advising a UK startup founder or CTO right now, I wouldn't spend the strategy offsite on "will AI take our jobs."
I'd ask sharper questions:
- What skill work are we paying for that AI already compresses? Automate or assist there first. Don't romanticise busywork.
- What know-how is actually ours? If a competitor with the same model and public docs can match you in a week, you don't have a moat. You have a head start.
- Can our product use private context without becoming a liability? RAG, permissions, audit trails, and evals matter more than flashy demos when the valuable stuff is proprietary.
- Are we hiring for application taste or for facts people can Google? Facts got cheap years ago. Judgment under constraints is still expensive. AI makes that distinction clearer, not softer.
The job market will adapt. The scarce resource will move.
Jobs shifting is real. Dramatic change inside roles is real. A sudden unemployment cliff that kills demand for the same products AI is meant to sell into is a weaker story.
The deeper move is the scarcity flip: public knowledge stayed abundant, applied skill got cheaper, and proprietary know-how gets more valuable.
Build like that's the world you're entering. Because for most software products, it already is.