Which Jobs Will AI Disrupt Most?
20:40 Share

Today on the AI Daily Brief, a look at which jobs might be most impacted by artificial intelligence. The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. To join the conversation, follow the Discord link in our show notes.

Hello, friends. Quick note before we dive in. This episode ended up getting very long in the main episode. And between that and the fact that honestly, the headline news was a little bit underwhelming today, I just decided to go with this as the whole episode. Obviously, we don't do this that often, and we should be back on Monday with our normal breakdown between headlines and the main episode. But for now, hope you enjoyed this discussion.

Welcome back to the AI Daily Brief. Today we are talking about one of the most pertinent issues facing us as a society when it comes to artificial intelligence, which is where it is likely to cause economic disruption. There has been an absolute and undeniable increase in this part of the conversation, specifically as the big frontier labs seem to indicate that we are getting closer and closer to AGI.

There is a sense percolating that these big disruptive moments are just around the corner and a consequent urging from some of these leaders, including Anthropics Dario Amodei, that we need to be taking it more seriously and trying to have the conversations now on what we do in a post-AGI world.

Some of that's policy remediations. Some of that's thinking about the economics of it. And Anthropic specifically is now putting its research where its mouth is, releasing this week the first iteration of their Anthropic Economic Index. This is a research paper which sought to determine which professions were using AI, how they're using it, and how much they're using it. It drew on anonymized interactions with Claude gathered over the past two months, meaning it's extremely up-to-date.

Importantly, it's the first paper of its kind based on analyzing actual usage logs, rather than just self-reporting through surveys, which makes it a really valuable addition to this canon. On a high level, Anthropic found that over one-third of occupations are using AI across at least a quarter of their workplace tasks. However, only 4% of occupations are using AI in three-quarters of their tasks. AI usage also leans towards augmentation of workers rather than full automation of tasks, but not by that much.

57% of AI use was augmenting a human worker, while 43% was automation of individual tasks. This number I think actually should be a dramatic wake-up call. If Anthropic is already saying that 43% of this usage is AI directly performing tasks, and that's in the pre-agentic era, how much more of this behavior is going to shift in that direction?

Next up, Anthropic broke down AI usage by different categories of work. And way, way out ahead in the lead, the number one category for work was programming and development. That represented 37.2% of all use. The second highest category of work was creative and editorial, or as they call it, arts and media, with 10.3%. There were

There were also a wide range of tasks here, mostly in writing. Producing and performing in film, TV, theater, and music was actually fairly high with 1.8%, but there were also things like marketing use cases. Following that in categories was educational tasks at around 9%, office and administrative at around 8%, science, including life, physical, and social science at a little over 6%, and business and financial at 5.9%.

And again, in this study, they also get down to the task level. So you can have a sense of the use cases that are actually driving the industry right now.

Now, before we move on, it's really worth digging into the developer use case. As we were discussing in an earlier show this week, at this point, Anthropic has established themselves as the go-to model provider for coding assistants. They generate around 85% of their revenue from API usage compared to 28% for OpenAI. The point of that is not to say that this is not useful, but that it may not be representative of LLM and AI use in general, given that it's using cloud chat logs and

and programming is so far and away their biggest use. Another area that I thought was worth digging into is over here in the science category. It looks like when you dig in, it's actually a lot more of the social science fields that we're leading. The top titles were clinical psychologists, historians, and anthropologists, and the major subcategories of use seem to be around academic research.

One thing that this brings up is that in this report, Anthropic doesn't have a good way of differentiating between casual and enterprise use. In other words, it's not impossible that idle queries about historical topics or self-treatment of psychological issues could be skewing the results. Anthropic even acknowledged this in their footnotes, writing, We can't know for certain whether someone using cloud for a task was completing a task for work.

Someone asking Claude for writing or editing advice could be doing so at work, but they could also be doing so for the novel they're writing as a hobby. And you see a lot of this type of caveating with this, which I think is completely responsible and right on from Anthropic. They're trying to have a broader conversation based on a broader set of knowledge, not trying to argue that this is exactly the truest, clearest breakdown of how people are using AI right now.

Moving on, one of the features of the study that grabbed a lot of attention was a comparison between the volume of clod used per job title and the size of that workforce. In other words, this allowed Anthropic to show which sectors of the economy were using clod disproportionate to their size. Once again, and this is obvious if you are actually watching this, software engineers were the clear outlier.

Just 3.4% of the U.S. economy work in some form of programming or development, but they represented 37% of clod usage. Other categories that were disproportionate between how big a sector they are and how much clod they were using include art, design, sports, entertainment, and media, representing just 1.4% of jobs, but 10.3% of usage, as well as life, physical, and social sciences, which represents 0.9% of jobs, but 6.4% of usage.

Education is a little bit higher, representing 5.8% of the economy, but still outperforming with 9.3% of clod usage. In every other sector, clod usage either basically matched what percentage of the economy they were, or wildly underperformed it.

Some of them kind of make sense, right? Transportation and material moving, 9.1% of the economy, but just 0.3% of cloud use. Just by nature of what those jobs are, this makes sense. Same with food preparation and serving related, 8.7% of the economy, but just 0.5% of cloud use. One that was super surprising, and I guess reflects the fact that a lot of sales is in person, is sales and related, which is 8.8% of the economy, but just 2.3% of cloud use.

Even assuming that there are a lot of people who are doing sales outside and off their screens, it still feels like this is low and a big opportunity for salespeople to get out ahead of their peers and colleagues.

Now, looking at job replacement or augmentation again, Anthropic wrote, as we predicted, there wasn't evidence in this data set of jobs being entirely automated. Instead, AI was diffused across the many tasks in the economy, having stronger impacts for some groups of tasks than others. The conclusion for them is that AI adoption is about automating or augmenting a subset of routine tasks rather than wholesale worker replacement. But I also just tend to think that this data still represents very nascent usage.

There was also some interesting insight to be had around salary analysis. Almost all of the job categories that Anthropic chose represented less than 1% of clod use. There were numerous roles in the $50,000 to $100,000 salary range that showed up in the data between 0.5% and 1.5% of overall usage. And all of the heavy-use roles, like software developers and copywriters, had median incomes close to this range.

The example Anthropic used for an ultra high-end profession that isn't using Claude very much was obstetricians and gynecologists. And down on the low end of the salary range, the example job that isn't getting much use out of AI was shampooers. As I mentioned a couple times, Anthropic goes to pains to provide the caveats to help you contextualize this information. The one additional that I haven't mentioned yet that is really important to note is that this was an analysis only of the free and pro tiers of chat and excluded enterprise team and API users.

I would be particularly interested to see a version of this where enterprise and team were in place, as it feels like it would change the results fairly dramatically. The other big thing that I keep coming back to is that this is still a pre-agent survey of AI usage. I tend to think that a lot is going to change as functional agents actually become available. But in any case, this is still an extremely valuable report that provides a lot more insight into where things stand.

In terms of the response to this, a lot of people seized on the fact that the distribution of use is really uneven right now. Brian Romney writes, AI is not evenly affecting all sectors, but is disproportionately displacing certain types of work. On the other side, Nick Pinkston writes, there are zero individual tasks related to manufacturing, construction, or anything IRL and physical, it would seem. Surprised, but good to see it laid out.

A Kiminismist account on Twitter makes the point that this represents a larger shift in sentiment. They write, Today, Anthropic presents a way to measure the impact.

It's clear the impact of AI on our societies is growing, and the big AI companies are trying to prepare us because it's happening much faster than we thought. Today's episode is brought to you by Vanta. Trust isn't just earned, it's demanded. Whether you're a startup founder navigating your first audit or a seasoned security professional scaling your GRC program, proving your commitment to security has never been more critical or more complex. That's where Vanta comes in.

Businesses use Vanta to establish trust by automating compliance needs across over 35 frameworks like SOC 2 and ISO 27001. Centralized security workflows complete questionnaires up to 5x faster and proactively manage vendor risk. Vanta can help you start or scale up your security program by connecting you with auditors and experts to conduct your audit and set up your security program quickly. Plus, with automation and AI throughout the platform, Vanta gives you time back so you can focus on building your company.

Join over 9,000 global companies like Atlassian, Quora, and Factory who use Vantage to manage risk and prove security in real time.

For a limited time, this audience gets $1,000 off Vanta at vanta.com slash nlw. That's v-a-n-t-a dot com slash nlw for $1,000 off. If there is one thing that's clear about AI in 2025, it's that the agents are coming. Vertical agents buy industry horizontal agent platforms.

Agents per function. If you are running a large enterprise, you will be experimenting with agents next year. And given how new this is, all of us are going to be back in pilot mode.

That's why Superintelligent is offering a new product for the beginning of this year. It's an agent readiness and opportunity audit. Over the course of a couple quick weeks, we dig in with your team to understand what type of agents make sense for you to test, what type of infrastructure support you need to be ready, and to ultimately come away with a set of actionable recommendations that get you prepared to figure out how agents can transform your business.

If you are interested in the agent readiness and opportunity audit, reach out directly to me, nlw at bsuper.ai. Put the word agent in the subject line so I know what you're talking about. And let's have you be a leader in the most dynamic part of the AI market. Hello, AI Daily Brief listeners. Taking a quick break to share some very interesting findings from KPMG's latest AI Quarterly Pulse Survey.

Did you know that 67% of business leaders expect AI to fundamentally transform their businesses within the next two years? And yet, it's not all smooth sailing. The biggest challenges that they face include things like data quality, risk management, and employee adoption. KPMG is at the forefront of helping organizations navigate these hurdles. They're not just talking about AI, they're leading the charge with practical solutions and real-world applications.

For instance, over half of the organizations surveyed are exploring AI agents to handle tasks like administrative duties and call center operations. So if you're looking to stay ahead in the AI game, keep an eye on KPMG. They're not just a part of the conversation, they're helping shape it. Learn more about how KPMG is driving AI innovation at kpmg.com slash US. Anthropic is not the only one trying to do this type of research.

Stanford PhD economist John Hartley recently added to the work in this category with a new paper. Written by Hartley and a group of contributors and published in January 2025, the paper was called The Labor Market Effects of Generative Artificial Intelligence. A couple interesting findings from that, Hartley and co. found that roughly 30% of workers had used Gen AI in their roles, in line with the adoption curve from previous studies. Gen AI use was stratified by education level, with 50% of workers with a graduate degree using Gen AI and much lower adoption for those without a college degree.

The industries with the highest adoption rates included information services, management, real estate, construction, and education. Now, obviously, one thing that's really interesting there is that construction didn't show up at all in clod use. And so it'd be really interesting to try to dig deeper into this data and understand what type of use cases Hartley & Co. were identifying in those areas. But in terms of where there was least adoption, much of it makes sense in the context of what we saw with Anthropic as well. The lowest adoption was found in agriculture, mining, oil, and gas.

Although one overly low area was government and military, which you got to think the U.S. is racing to change right now. The researchers found that Gen AI use at work increases in proportion to salary, in basically a straight line after $25,000 to $35,000 salary level. Reinforcing results from other studies about intensity of use, the research found that one-third of workers who use AI use it every day, while 15% use it between one and three days per week.

Importantly, there were basically zero examples of a worker testing AI in their workflow and then choosing to stop using their tools. The paper also reinforces Anthropic's findings that AI is more augmenting right now rather than automating. Roughly 15% of the time AI is being used to complete entire pieces of work, while 60% of usage, by far the most, is related to assisting people complete their work faster. Indeed, faster is kicking the slats out of doing it better, with doing it better just above 20%.

One interesting result, especially given that the higher income levels were using AI more, is that the less than $35,000 income bracket had the largest efficiency gains. They reported 120 minutes in time savings for AI-assisted tasks. This fell to 40 minutes for workers earning between $35,000 and $50,000, and then scaled back up to 80 minutes for the highest income bracket above $150,000.

Statistics around the time saved broken down by task were also super interesting. Hartley shared a chart, average number of minutes to complete a task with and without generative AI, that shows every single task just being a fraction of the time with AI as it was without AI.

With Gen AI, writing is at 25 minutes versus without Gen AI at 80 minutes. Negotiation, 141 minutes without Gen AI versus 29 minutes with Gen AI. Even Malik pointed out the obvious. The productivity gains appear large. Workers report when they use AI, it triples their productivity, reducing a 90-minute task to 30 minutes.

One more interesting part of the conversation around job replacement that was all over the place this week was this chart you might have seen from the U.S. government of the massive increase in software development job postings and the cratering back down subsequently.

This draws specifically from government employment data and shows a 70% collapse of software engineering jobs between 2022 and 2025. Greg Eisenberg writes, this chart is nuts. Software developer jobs down 70% from peak. People will blame the end of free money, but something way more interesting is happening. The middle class engineer is dying and it's dying because they're not needed anymore. One good dev with GitHub co-pilot ships what entire teams did five years ago. Microsoft just reported the highest revenue per employee in history.

The entry-level engineer doesn't exist anymore. Instead, we have product builders who happen to code. Armed with AI, they ship entire products in days. Meanwhile, the truly elite engineers are making more money than ever. They've shifted to working mostly on frontier tech, I mean the stuff that's really hard. AGI at OpenAI, designing rockets at SpaceX, self-driving car tech at Tesla. Product

Product builders are becoming solopreneurs and creators while frontier engineers are making hedge fund money. In 2025, software engineer doesn't mean what it meant in 2020, and that's what this chart really shows. The top is elite status, and everyone else is becoming a builder.

Economist and author Noah Smith thought the conclusion was very clear, posting, don't learn to code. Then again, after getting his clicks, he clarified that he was being sarcastic and confirmed that everyone should learn to code. And so the question was, for many people, are we witnessing the end of software engineering as a profession? Well, there are a lot of challenges with this chart. First of all, it shows job postings on Indeed, not actual jobs. The chart also shows a huge run-up in jobs in 2021. And then

and then a crash in 2022, beginning six months before the release of ChatGPT. Indeed, by 2023, jobs weren't all that different from where they were back before the run-up in 2021. We all lived through this period, and the tech industry scaling back their hiring in 2022 had nothing to do with AI coding. Indeed, as Greg pointed out, this sort of does look like a chart about post-COVID dislocation.

Programmer Andreas Liberopoulos wrote, Without pre-2020 data, this chart says more about the excesses of the 21 to 23 era than it does about the current state of the software engineering market.

Taking those criticisms into account, there's still something interesting showing up in the chart. Gergely Oroz, the author of the Pragmatic Engineer blog, commented, However, as Greg Eisenberg pointed out, the market for software has grown a lot since before COVID.

Perplexity claims a 70% increase between 2020 and 2024. Now, software revenue isn't necessarily correlated to workers in the industry, but this is a very big divergence. And thus, the collapse in job listings may not be as remarkable itself as the fact that they haven't rebounded alongside demand for software.

If anything, though, the incredible virality of this tweet, being liked 15,000 times, really demonstrated the fear that surrounds this issue. Even if software engineering isn't being replaced by AI right now, there is a visceral concern that that day is coming. Earlier this week in Paris, Sam Altman discussed how advanced agents will change software engineering. He likened the change to the launch of deep research, claiming that companies could use 50 cents of compute to complete $500 to $5,000 worth of work.

He said, companies are implementing that just to be way more efficient. I think you'll see this in a big way with the software engineering agent. Still, many think that we're getting out over our skis. DevList founder Alex Charbonneau writes, I think if someone is claiming how software engineering jobs are becoming obsolete, it's just a signal they haven't used the software engineering AI tools themselves to see how much they suck. They help with boilerplate and how do I format this date, but try building anything more complex than a template with just AI.

Entrepreneur Adam Small wrote, AI is not going to replace software engineers, developers, or whatever title you call it. AI is going to make SWEs more productive and accelerate the pace of development. If you aren't using AI as a dev tool, then you will be replaced by someone who is. That is what you have to worry about as a software engineer.

Now, I tend to think that arguments around capabilities right now are the worst arguments for understanding where AI is actually going to be disruptive in the future. It just changes so fast. And while all these critiques of what is available right now from coding agents may be true, it's not going to be the same in six months to say nothing of 12 and 24 months.

At the same time, and what might be really important about software engineers, is that there is probably no other role that more directly has the ability to translate productivity gains into building way more rather than just building the same amount with less effort and less money. When engineers are taken off the leash to be able to build as though they had a team of 10 people helping them, they're not just going to build the same stuff.

They're going to build wildly more complex stuff or customized stuff. And I think that to the extent that we see this as the model where people translate the productivity gains they get from agents into building more, better, cooler, more interesting stuff, the more that that becomes the normal expectation and the less we just see agents as ripping out every job that exists right now.

Anyways, that is going to do it for today's episode. Lots and lots of interesting things to chew on. Good job to Anthropic on the Economic Index. I'm excited to see more of where that comes from. Appreciate you guys listening or watching as always. And until next time, peace.