The Current Status of AI Job Displacement

Are we getting replaced?

November 30, 2022.

OpenAI threw ChatGPT out into the wild, and millions of people were floored by just how much this thing could actually do.

For the hardcore AI crowd, it was the kind of breakthrough they thought they’d only read about in science fiction books. For creative workers, it suddenly looked like an enemy at the gates, ready to suck the joy out of making things. And for everyone else? Well, now you could ask a robot to write Hemingway-style love letters or crank out a Donald Duck rap, and it would actually deliver.

But deep down, everyone had the same thought: is this thing going to take my job? Are my skills suddenly worthless? ChatGPT seemed ready to do just about anything you and me could do (and do things we needed to spend years learning).

If that thought crossed your mind, it doesn’t make you a luddite or a doomsayer. It just makes you human. The fear is natural, especially when we’re staring at a tool that can juggle so many tasks we once thought were safe in human hands.

Back in 1908, if you drove a carriage and cars became a thing, you could switch to driving a car and ride out the rest of your career. Today we don’t know if we’re the driver or the horse in this analogy. Picking up a new skill isn’t a long-term safety net anymore. This thought is enough to give anyone a knot in the stomach.

The famous “Leader of the Luddites” Engraving

Like with most things (and after losing more sleep than I’d like to admit), I eventually decided to stop panicking and just wait to see how AI actually plays out in society, and what it really means for jobs. Will it take over everything I do? Will it leave my friends and family suddenly out of work?

… and, the awkward question I need to ask myself: am I personally speeding this up as the founder of an AI company?

To get some answers, I started digging into how the job market is shifting under AI’s weight. That’s what this post is about: where we are right now and what we actually know, circa August 2025.

To write this post, I’ll lean on economic studies. Sure, they come with the usual trap of mixing up correlation and causation, but they’re still the best evidence we’ve got for measuring AI’s impact at the macro level.

Office Work + Entry Level have the biggest impact so far

Let’s start with a fresh paper by Erik Brynjolfsson, Bharat Chandar, and Ruyu Chen, released just two days ago (August 26). It backs up what many of us may have suspected. Using ADP (Automatic Data Processing) data from the past 4 years, the study analyzed job payments and trends.

The first big takeaway from the paper: employment for young workers has dropped in jobs most exposed to AI. What are these roles exactly? The white-collar ones: software engineering, customer support, marketing, etc.

Source: Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence

Source: Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence

The trend is pretty clear: early-career professionals are struggling more than ever to land a job, with entry-level roles taking the biggest hit.

On the flip side, jobs that deal heavily with the physical world barely show any impact at all. And roles where empathy is the core skill (like health aides) are not just safe, they’re booming. Partly because empathy is hard to automate, and partly because populations are shrinking and aging, which only fuels the demand.

It’ll be interesting to watch how this plays out:

• Will this squeeze on early-career jobs eventually ripple upward into mid- and late-career roles?
• Once robots get better at actually interacting with the physical world, will today’s “safe” jobs face the same fate as office work, or will human touch keep them protected?

Another interesting take from the article: the shifts so far aren’t about paychecks getting smaller. It’s the roles themselves that are changing, not the salaries.

… but unexperienced workers may also benefit from Generative AI.

In another study (again with Brynjolfsson on the author list) looked at how Generative AI plays out in customer support.

The findings are pretty interesting:

• Newer, less experienced workers suddenly looked a lot sharper, with productivity jumps of roughly 15%.
• Veterans and high-skill reps, on the other hand, barely moved the needle. At best, they left the study with marginal gains.

There were some other interesting effects too. Customers were less likely to escalate their issues when AI was in the loop, which also meant lower turnover among junior staff. On top of that, metrics like “resolutions per hour” and “average handle time” got a boost — almost certainly thanks to the rookies who suddenly had AI covering their blind spots.

This ties back to the study we’ve seen before: today’s junior employees can get a lot more done than their pre-AI counterparts.

Which raises the question — if one junior now produces as much as two used to, do companies really need as many entry-level hires going forward? In other words, the bar for “minimum productivity” maybe just got pushed way up.

The research reads like a tug-of-war

In 2023, the International Labour Organization (ILO) came out with a study basically saying: relax, work isn’t going anywhere, it’ll just be transformed and there’s no immediate job loss.

Their analysis leaned on ISCO-08 classifications and ILO survey data. ADP works with payroll transactions (money actually moving), while ILO works with surveys (people saying what they do) and that can cause some bias.

The ILO study made some interesting statements:

• Certain jobs are more exposed than others. Clerical and administrative tasks are more affected, with broader office work following behind. That lines up with what other research conclusions.
• ILO was less dramatic about job losses. Timing could be part of it, though. Remember, this study was done just a year after ChatGPT dropped.

Another important angle they flagged: the divide. Low-income countries probably will feel less impact, mostly because infrastructure and costs will not enable adoption. But that also means they risk missing out on the productivity boost AI could deliver.

The 2025 update takes a more cautious stance, acknowledging some job exposure, but still concludes the overall impact is limited compared to the earlier study.

The Inflection Point

Another paper takes the debate a step further, and here the story splits in two directions. Economists call it the “displacement effect” vs. the “productivity effect”.

This study analyzed freelancer work, and the results again pointed out to some impact. After ChatGPT arrived, jobs for translators dropped by about 9%, and their earnings sank almost 30% (using data from freelance platforms).

In contrast, web developers saw the opposite. Job volume jumped 6.4%, and earnings shot up by roughly 66%. Here, AI turned into a sidekick, helping them code faster and ship more projects.

But there’s a catch: researchers introduce an interesting idea they call the “inflection point.” Before AI gets too good, it boosts your output, but once it crosses a certain threshold in your field, the machine takes over, and humans get sidelined. Will every type of work may have its own inflection point?

So… what do we actually do with all this? For me, the best protection is learning how to live with ambiguity and navigate it. If your role is basically “take orders, execute, repeat”, you’re probably sitting in the high-exposure zone.

Looking at these results, I don’t think the right reaction is panic. Yes, there’s tension around job displacement, and yes, entry-level roles are getting thinner on the ground. But if you position yourself as an AI-native worker (someone who can actually leverage these tools) you’ll be better prepared for the shock waves ahead.

In my own field (data science), the real edge isn’t just “using AI”. It’s AI plus the fundamentals. Code, statistics, domain knowledge, that’s definitely the bedrock. Only by knowing the fundamentals can you unlock the productivity boost AI promises and leverage these AI tools.

.. Of course, you could also move into the physical world. If you’re fixing pipes, moving boxes, or caring for people, the robots aren’t lining up to replace you tomorrow. At least not until they get out of the factories.