Three New Opportunities in AI's Transition Era

The AI boom has been in full swing for more than half a year now. Experts keep insisting the robots are coming for our jobs any day, but happily, most of us are still gainfully employed. What's been more interesting these past six months is that, contrary to the doom-and-gloom predictions, a whole new category of "AI assistant" gigs has sprung up instead. Browse any freelance-work site around the world and you'll find no shortage of listings that simply bolt "autopilot" onto business as usual. Some of the job descriptions are downright bizarre — "hire me to use AI to handle some tedious task for you" — wait, isn't AI supposed to save you the labor cost in the first place?

Old Field, New Money: Vector Databases Get Their Moment

It brings to mind the old bowling machines that needed a human to reset the pins. The first robot to truly stun the public, back in the 18th century, was "The Turk" — a chess-playing automaton, a turbaned mannequin whisked about by gears, who happened to play excellent chess. Except underneath, a real chess master was hiding inside the whole time. So it was really a robot puppet powered by human intelligence. These days it feels like the roles have flipped: we're the ones playing the Turk, moving the pieces on the board, while AI is the real chess master calling the shots. In any case, the world is still some way off from a stage where robots do everything themselves — for now, we're in a transition period.

So over these past six months, I think this transition toward AI adoption has thrown up new opportunities in three distinct areas.

The first type covers jobs that were doing fine before, but where AI — specifically large language models — changed the underlying premise and turned them into critical technology. Some fields that used to have narrow applications have suddenly become mainstream investment targets.

Vector databases are a case in point. They're not exactly new technology, but their range of applications used to be pretty narrow. When we look things up, we normally use traditional databases, which run on relational logic — because looking something up is about getting an exact, correct answer.

But if what you're after is fuzzier logic — a roughly approximate answer — that's exactly where vector databases shine. Fuzzy logic, approximate answers: sounds like artificial intelligence, doesn't it? And indeed, the AI wave is exactly what's made vector databases hot again, since chatbots and text-organizing tools both rely on them. The Dutch company Weaviate just raised $50 million doing precisely this.

From an investor's standpoint, though, there won't be that many ultimate winners in vector databases. Plenty of the established database giants are going to muscle into this space too, so while the field has heated up, that doesn't necessarily make life easy for startups. It's a problem most AI companies are running into right now.

The second type is purely new fields and new jobs. Many of them are old tropes dressed up in a new field, but still genuinely illuminating.

The best-known example, mainstream among the mainstream, is generative imagery — everyone's played with it, and it's the one ordinary people feel most viscerally. The undisputed king of the category, Midjourney, is a remarkably lean startup: the ratio of how few people they employ to how much value they've created is rivaled by maybe only Instagram, a social app everybody actually uses.

Instagram had all of 13 employees when it was acquired, and Midjourney has just 11. They're lean to the extreme — their primary interface isn't even their own app, it's Discord. Investors claim the company is pulling in $100 million a year in revenue, and that's just where things stand today.

Working for the Robots: Data Labeling Demand Goes Up, Not Down

Of course, companies like this still have old-guard rivals such as Adobe — you've got it, I've got it too; every major player now offers image generation of some kind. But there's still plenty of room in this new field, and I'd say it's a long way from saturated.

The third type is the work of zipping up AI's jacket and tying its shoelaces for it. Take data labeling, for example — a crucial job in the AI world. The way it's traditionally operated is by routing datasets, through software tools, out to countries with cheap labor, or to underemployed people in rich ones. Amazon even runs its own crowdsourcing platform for this, called "Mechanical Turk" — the name tells you exactly what it is: human intelligence, working for the machines.

But now we have LLMs. Was the human labor that used to label models for a living about to get steamrolled by unsupervised learning in these big, unsporting models? As it turns out, no — full replacement isn't happening yet. If anything, as everyone rushes to adopt AI, demand for labeling has only grown. Machine learning can squeeze the human labor cost down some, but third-party labeling firms don't look like they're short on business anytime soon. Companies like Scale and Snorkel, which do exactly this kind of data labeling, seem to be doing quite well.

This field of humans helping machines is likewise a bit transitional in nature, mainly because AI itself is expected to eventually handle a lot of these impressive tasks on its own — at which point it won't need us tying its shoelaces anymore. But in the short term, as the AI field keeps expanding, this kind of business will actually grow. Then again, viewed over a 10-year or 100-year horizon, what business isn't transitional?

These are the three types of areas I've seen get immediately swept up by the AI revolution over the past six months — some are old businesses revived and reactivated by AI, some are entirely new fields, and some look like they'd be disrupted but are actually enjoying better times in the short run. Exactly how long this in-between stage will last, we'll need to keep watching before it becomes clear. But one thing I'm sure of: AI is going to change the world by more than smartphones ever did.