← The AI Daily Brief: Artificial Intelligence News and Analysis
Loading summary
Transcript1 lines
- [00:01]
A
Today on the AI Daily Brief, the new Jobs AI will create The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. Alright friends, quick announcements before we dive in. First of all, thank you to today's sponsors, Granola, Superintelligent, Bolt and Section. To get an ad free version of the show, go to patreon.com aidaily brief or you can subscribe on Apple Podcasts. If you want to learn more about sponsoring the show or really finding out about anything else in the AI DB ecosystem, head on over to AIDAdaily Brief AI one of the big points of discussion for this week has been the idea that there is this ever so subtle shift in the AI jobs narrative. It would be going way too far to say that this was becoming anywhere near mainstream, but you're starting to see more people at least question the premise that AI and agents getting better means people will have less work. And yet, even among those who are arguing that the AI job apocalypse narrative is way overblown, the argument tends to be much more backwards looking. It's looking at the way that productivity has previously impacted industries and extrapolating that out to the future. That's well and good and all of that is important. But I think it is a significant failure of the AI industry to take the next step and actually start to explore the type of jobs that there will be in an AI enabled future. The sheer tonnage of time spent on assessing which jobs are most at risk compared to the almost zero time exploring what types of new jobs will be created, represents one of our great failures and leaves people who want to be optimistic about the future clinging to vague hand wavy notions about what those jobs might be now. From the standpoint of sheer epistemic humility, of course we have to be careful about arguing with any sort of confidence why what specific things will be created in the future. In other words, the more specific our future predictions, the less likely to be right they are. But that doesn't mean that we can't at least explore from first principles how we think the change is going to play out, if indeed it is not going to be the mass destruction of white collar work that some are promising. So what I'm going to attempt today is talk about one the fundamental problem and the hidden assumption in most of the job apocalypse narratives, 2 what opens up when you correct that assumption 3 how to deal with the AGI objection and 4 a real walk through a specific sector and some jobs that I could see very plausibly being a part of the future and representing meaningful amounts of employment. So let's talk first about that hidden assumption. AI is mostly analyzed right now as a labor supply story. In other words, AI increases the supply of labor, so labor gets cheaper and workers get displaced. But all of this rests upon the presumption that demand stays constant. This is another way of framing the lump of labor fallacy, the idea that there is a set amount of work to do. So if technology comes and does some of that work, that means that humans aren't doing that work, which means that those humans don't have any work to do. What it doesn't take into account is the expansionary nature of economies and the ability of demand to expand, to absorb more supply. Now, I don't want to spend a lot of time on the historical arguments here. As I mentioned, that's the main substance of pretty much all other arguments along these lines. Yet it is worth noting, of course, that the assumption that demand stays constant as the supply of labor increases has never held true. And so if we are just going from historical precedent, it is unlikely to be true this time either. The more important question, if we're trying to actually interact with people's lives as they're lived now, not just provide a history lesson, is to ask where demand will expand. Basically, what parts of the economy could actually grow as AI allows us to produce more? And I think what's important to note is that demand elasticity actually comes in a variety of different forms. The one that's most obvious is, of course, price elasticity. That is, I wanted the thing, but it cost too much. If AI makes a thing cheaper, new buyers can enter the market. A second category of demand elasticity is access elasticity. I wanted it to, but I couldn't get it. Now, whereas AI supported price elasticity because it lowered the cost floor, in this case, we're talking about AI reducing provider scarcity or wait times, or geographic barriers or other types of institutional bottlenecks. Basically, instead of it just being cheaper, AI makes a good or service more readily available. A third type of demand elasticity is complexity elasticity. I needed it, but the system was too confusing. Think help with taxes, insurance, immigration, benefits. The way that AI supports this type of demand elasticity is by making opaque systems navigable. You're already starting to see this as a major way that people are consuming AI interacting with legalese or contract things, or even doctors, recommendations they didn't really understand before, but now have this sherpa and guide that can help reduce that complexity. Another category of elasticity we'll call continuity elasticity. And here perhaps we're getting a little bit more abstract, but I still think it's worth noting. The idea of continuity elasticity is going from I get help occasionally to I'd benefit from help all the time. So think about areas like health or coaching, where always on monitoring and support could become cheap enough to operate at scale in a way that wasn't possible before. A fifth form of demand elasticity we'll call personalization elasticity. This is I get the generic version now, but I'd value something made for me. When AI brings the cost and ease of production down, it's going to make customization cheaper. The last category we'll call relational elasticity or value elasticity. And the idea is I want this thing that I am buying, whether it's a good or a service, to be more human, meaningful or trusted. This is what Chicago booth economist Alex Emas wrote about in his what Will Be Scarce essay, how the relational sector, where the provenance of a good or service, how it was made and who made it and who is providing it are actually an integral part of the value of the service, I. E. People deciding that they don't just want a coffee, they want an experience surrounding that coffee that involves interaction with a human and a particular quality that comes from some sort of human touch. So we've got these different categories of how AIs impact on the supply of things in terms of cost of production, ease of production, speed of production, et cetera, can interact with elasticity of demand to expand demand in new and clear ways. And one thing that's interesting to look at as we Try to understand AI's impact on different sectors, especially the positive and generative impact of AI, is to ask which of those elasticities are common in those sectors. They're present pretty much everywhere, but different sectors are going to have different elasticities more strongly represented. For example, in experiences in travel, that relational elasticity and personalization elasticity are going to be particularly high. But you're not likely to see continuity elasticity because you're not able to have experiences or travel continuously throughout your life, as awesome as that would be. Meanwhile, on the other end of the spectrum, healthcare, I think, has every single one of these types of elasticity, which is why it shows up as our case study in a little bit. Now, trying to zoom out and consolidate this argument, I think we can look at AI as expanding the demand frontier in two directions at once. It either is increasing and democratizing access to things that are available now, or unlocking things that aren't possible. Now, if you go back up to the types of elasticity, you'll note that in many of these cases, if you were currently rich, you could already get access to the thing in the way that the people want it. Now, under the conditions of the new elasticity, if you're rich, you weren't as price sensitive. You could probably pay to solve access issues or complexity issues. If you were rich enough, you could even use money to solve continuity issues. And you could probably pay for personalization and relational value as well. So the first unlock of AI is the affordability unlock. Basically the same menu, but at a lower price. And that's most of what you were probably thinking about when I was describing these six categories of demand elasticity. To take one example, let's think about small business professional services. Right now, the average small business owner might not have the budget to buy a $5,000 design project, a $3,000 marketing campaign, a $2,000 legal review, and a $1,500 analytics report. This obviously does not mean that the business doesn't have demand for design, marketing, legal and analytics. It means that at the current state, the old version of those services was unaffordable. If AI reduces the cost of those jobs from 5,000 to 500 and 3,000 to 300, an enormous long tail market activates millions of small businesses that were never agency clients before become buyers for the first time. This is again what I'm calling the affordability unlock. Where there is net new demand for services that already exist because the changing cost and availability of provision opens up new buyers for which that service wasn't viable before. The second type of unlock I'm calling the possibility unlock. To use the menu analogy, if the affordability unlock is the same menu at a lower price, the possibility unlock is a new model entirely. This is where AI makes a service model operationally possible that could not previously exist at scale. The category itself becomes possible before anyone can demand it. Previewing where we're going to get in our case study, an example of this, I think, is continuous preventative health care. The vast majority of people not named Brian Johnson don't currently have someone continuously watching their health data or tracking their care plan, or noticing drift or coordinating with their pharmacy. And to be clear, people aren't demanding this exact service. This isn't a normal category because it hasn't been operationally viable. But as AI collapses the cost of the informational layer around care, data collection, monitoring, summarization, documentation, scheduling, escalation, routing, once all of Those costs fall and the systems around them become viable. A totally different healthcare model becomes possible. One that is personalized, data driven, preventative and continuous. This is entirely net new demand, because no one had this before, because it wasn't possible until AI made it possible. When it comes to new jobs, the affordability unlock opens up whole new categories of providers that cater to different parts of the long tail of the market. The possibility unlock creates entire new markets, which entire new actors will come into to provide those goods and services that are now actually possible. But now we have to get to the question that stops all of these conversations mid track. I would argue that the number one reason that people claim this time is different and that AI is not a so called normal technology is that they say the people that are optimistic about the jobs created in the future aren't answering the question, won't AGI just eat those jobs too? In other words, for those small business professional services, if the first wave of AI brings costs down by a factor of 10 and makes those $5,000 jobs into $500 jobs, is that just a temporary state on the path to AGI that can do those jobs for 50 cents instead, with no human involved? To be clear, from a first principles perspective, this is a reasonable question and one that again, I think AI optimists are failing to engage with clarity and good faith. For my part, I think that there are actually quite clear answers. The question that we are asking, which I think is the wrong question, and the idea that is embedded in the assumption that AGI will just eat those new jobs too, is we think about the question of jobs only as a question of can AI perform the task? It's a capability question that assumes labor demand is grounded entirely and only in what humans can do. But that's too narrow. A better question, I believe is does AI only delivery satisfy the demand? Here we've moved from a capability question to a service design question. Many roles exist not because of capability gaps, but because of the constraints of the market's expectations in how a service is delivered. Things like trust, accountability, presence, relationships, they are part of the value. And while AGI can eat tasks and will eat many tasks, it will not automatically eat that demand for things like trust and accountability. To try to put a name on this, I'm calling it the human premium. These are seven categories of value that do not a priori transfer when you remove the human. The human premium is the portion of economic value that remains attached to human involvement even when AI can perform the underlying tasks. It is the answer to why doesn't AI eat those jobs too? Here are seven categories of human premium. The first is relationship or relational. Again, this is the one that Alex Imas explored. The idea that I want this delivered by someone who knows me. Where there is continuity, memory, accumulated trust, where the human part of the experience actually matters and is integral to the experience, where replacing a person with an AI makes the service worse enough that it actually decreases demand for that service. A second kind of related category of human premium we'll call embodied presence, where the demand constraint is I want someone there with me. This is where physical presence matters. The nurse in the room, the trainer correcting form. It's not that AI can't do some of the things that those people do. It's that the physical presence is important enough to many that there continues to be demand for that human premium of embodied presence. A third category of human premium is trust. I want to talk to a person before I act. Humans are social creatures. We value social proof. We value the experience of people who have come before us. We value understanding the context of the personal experience that we are trusting of people who came before us. As much as AI generates recommendations, there are many situations in which people will absolutely prefer the recommendation of a human, even if that human is just validating the AI that they're working with too. In many circumstances, humans will be asked to make AI generated recommendations believable, emotionally acceptable and actionable. Which gets us to of course, the fourth category of human premium, which is accountability, where the demand constraint is someone has to own this. People want a person who signs off, escalates, explains and is responsible when things go wrong. When I'm taking an experimental new peptide regime which is all the rage right now, I don't care if ChatGPT can tell me how to titrate things correctly at the right ratios, you better believe I want someone whose bonafides and credentials I can check and who, to put it crassly, I can sue if something goes wrong. The questions of legal accountability for AIs are going to make a lot of lawyers very rich for a very long time. But even outside of questions of the legal system, there is going to be a human premium on accountability for sure. A fifth area of human premium is translation. AI is interacting with the entire accumulated knowledge of the world. But some people aren't going to know how to ask for what they need. Even in a world of AGI. They will want human translators, people who can turn messy desire and constraint into usable AI mediated work. This, I think is extremely important for the example of small business professional services the AGI eats everything answer to this that won't it just compress that $500 job into a 50 cent job? Eventually that takes the job of the intermediary actor fails to understand the economic value of this translation. A small business owner might not want or have time to get good enough at using AI for design to take advantage of the 50 cent cost that AGI makes it. The market will find the price level at which a human can upsell the 50 cent AI work in a way that fits better with the process and constraints of that small business owner. It is a failure of economic understanding to assume that there is no room for margin on top of the underlying cost of goods sold. AI might dramatically compress that margin, but there will be margin there A sixth category of human premium we'll call behavior change. That's where I know what to do, but I need help doing it. Anyone who has tried to use a ChatGPT workout plan only to abandon it or at least integrate it with a human trainer who can help them stay accountable to themselves will understand what I'm talking about here. Simply put, there are certain categories of behavior change for which we will listen to a human and we will not listen to an AI. The last category of human premium we'll call provenance and status. This is what we were talking about in the relational sector that a human made it is part of why it matters. This could be arts crafts, custom goods, live performance, bespoke services. The human signature is part of those products. Today's episode is brought to you by Granola. Granola is the AI notepad for people in back to back meetings. You've probably heard people raving about Granola. It's just one of those products that people love to talk about. I myself have been using Granola for well over a year now and honestly, it's one of the tools that changed the way I work. Granola takes meeting notes for you without any intrusive bots joining your calls. During or after the call, you can chat with your notes, ask Granola to pull out action items, help you negotiate, write a follow up email, or even coach you using recipes which are pre made prompts. Once you try it on a first meeting, it's hard to go without. Head to Granola, AI AIDAily and use code AIDAily. New users get 100% off for the first three months. Again, that's Granola, AI AIDAily, OpenAI and Anthropic are both launching enterprise AI consulting efforts because everyone is realizing that the challenges and the capabilities of AI. The challenge is getting individuals in the organization actually ready to use it. The truth though, is that all the forward deployed engineers in the world aren't going to help you if you don't actually have a coherent strategy based on an understanding of your actual AI readiness. Super Intelligent Maturity Maps give you a chance to see where you stand relative to the industry on deployment, depth, systems integration, data access, outcomes, people and governance. And from there, our customized AI planning assessments can help you figure out what you need to do to improve your readiness and how to sequence it. Go take your own Maturity Maps quiz@BESuper AI and send us a note if you want to go deeper. Today's episode is sponsored by Bolt New. Bolt New is agentic engineering on multiplayer mode. Designers, product managers, and engineers build in the same environment, and the design system agent keeps every screen on brand no more Frankenstein UIs stitched from a dozen prompts. Whether you're shipping internal tools, moving from prototype to production, or replacing a legacy admin panel, Bolt New takes your team from concept to deployed app. One Personal Recommendation Hit Plan mode before you build I had a project I had half described in three different prompts, and Plan mode made me actually think through it with Bolt New before a single line got written. It saved me from rebuilding the same screen probably about four times. Build better apps faster Start with the link in the description. Here's a harsh truth. Your company is probably spending thousands or millions of dollars on AI tools that are being massively underutilized. Half of companies have AI tools, but only 12% use them for business value. Most employees are still using AI. To summarize Meeting notes if you're the one responsible for AI adoption at your company, you need Section. Section is a platform that helps you manage AI transformation across your entire organization. It coaches employees on real use cases, tracks who's using AI for business impact, and shows you exactly where AI is and isn't creating value. The result. You go from rolling out tools to driving measurable AI value. Your employees move from meeting summaries to solving actual business problems and you can prove the roi. Stop guessing. If your AI investment is working, check out section@section AI.com that's S E C T I O N A I.com. So with all of these concepts in mind, let's look at an industry case study in healthcare. I want to be very clear. I have no expertise in healthcare whatsoever. It's not a sector I've worked in. I'm coming at this from the perspective of a a technologist and b a consumer of health care. But I think it's quite clear that healthcare has potentially every type of demand elasticity that we discussed before. People could sue more of it if it was cheaper. They would probably consume more of it if they had easier access, like better clinics in their hometown. They would certainly consume more of it if the complexity was reduced, or they had better ways to navigate that complexity, especially when it comes to billing, insurance, etc. I think that likely many people would want a relational element and a personalized element of their health care rather than just being a number in a system. And I think the biggest demand elasticity, or at least the one that's most interesting when it comes to the idea of the difference between the affordability unlock and a possibility unlock, is around continuity. My argument is that we consume an incredibly small part, even a vanishingly small part, of the amount of healthcare we would consume if the labor math underneath it made a different kind of healthcare possible. Right now, we all live in a paradigm of reactive care that is episodic by default. Every once in a while we make an appointment, either because it's our annual checkup, or because there's a thing we're now at the age where we have to do once every three to five years, or because something stops working the way that it's supposed to. From there we have some tests, get a diagnosis which turns into a prescription, and then we're left alone again to fend for ourselves. At some point, a new crisis arises and the whole cycle starts again. There is, I believe, an AI enabled version of healthcare that is continuous, preventative and personalized. It involves AI for data collection, continuous monitoring, and even triage and ranking. But in doing so, it creates a whole new set of opportunities for human roles, net new human roles that emerge in this new continuous, preventative, personalized paradigm. There's an AI background layer that watches the data and a human action layer that helps convert that to meaning. So for the sake of this episode and for coming along this journey with me, let's assume that this idea that continuous preventative care becomes more viable and more normalized. If that happens, it will create demand for new types of jobs. And these are not AI jobs in the narrow sense, as we think about prompt engineers or agent operators. Now, these are healthcare service jobs that the AI layer makes possible, and each of them maps to some set of specific human premium categories that protect them from being eaten by stronger AI. The first role I'll call a continuous care navigator. This is the human layer between a patient and an AI enabled monitoring system. They oversee a caseload of patients in continuous monitoring and handle the moments that matter, such as the pattern having changed, the call that has to happen, the family that needs reassurance, or the escalation that has to land. So while an AI is doing things like ingesting data from devices, labs, pharmacies, patient reports, establishing baselines and detecting deviations, ranking cases and changes by urgency, updating documentation, routing escalation, the human, the continuous care navigator, is doing things like reviewing flagged cases, calling patients to clarify and ask the questions AI didn't know to ask in those personal interactions, noticing things like fear, shame, avoidance or unexpressed family dynamics, coordinating with clinicians to escalate things, and closing the loop after intervention. The human premiums involved here include trust, accountability, translation, behavior change, and relationship. So how many people could the market support? When it comes to health care, the total addressable market is everyone with a body. But for the sake of our arguments, let's zoom out a decade and take a conservative, middle and aggressive case among American adults. We'll call conservative 40 million enrolled patients, middle 80 million enrolled patients and aggressive 120 million enrolled patients. The conservative case then is about 14% of US adults enrolled, with the aggressive being about 43% of US adults enrolled now. Also on the conservative end, let's assume that one continuous care navigator can handle 150 patients. If that's the case, you're talking about about 276,000 jobs. On the other end of the spectrum, where we've got 120 million enrolled patients and we're assuming that they can only handle 100 people each, you're talking about 1.2 million jobs. A couple examples of jobs that are in that 267,000 person range include personal financial advisors, loan officers and coaches, and scouts. For the 1.2 million. That's about the number of high school teachers there are. And if you took advertising, marketing, promotions, PR and sales manager, they combined come in at around that number, meaning that we're talking about a ton of people and a ton of jobs. The next job in our hypothetical example, we'll call a care plan outcome specialist. This is the person who owns the gap between medical advice and real world execution. So if the continuous care navigator is the person who's helping assess and escalate issues as they come up, the care plan outcome specialist is the person on the other end of the cycle who helps with the implementation layer of healthcare. If AI is tracking each patient's care plan and milestones and monitoring medication adherence, tracking labs, screenings, et cetera, and integrating all of that into their continuous care plan. The care plan outcome specialist is the person who can provide a human interface for talking with patients about why the plan isn't working or help solve practical problems like cost, transportation or fear or family. Now, I'll leave it to you to decide whether there would be a smaller number of those folks as relative to the continuous care navigators. But you're still talking about potentially hundreds of thousands of net new jobs that don't exist today, that only exist because of the way that AI changes the overall healthcare system. Finally, representing another category of role that's likely to be fairly ubiquitous is the new type of technical roles that will emerge to support the AI enabled economy. This one I'm calling a health data operations specialist. The premise is that continuous healthcare is only as good as the data flowing through it. Wearables, labs, pharmacy systems, patient reported data, insurance records. The health data operations specialist owns reliability, integration, governance and clinical usability of that layer. Now, AI is going to be doing a lot of the work around data. It's going to pull and normalize data across sources. It's going to flag anomalies and integration failures. It's going to, it's going to generate pipeline, diagnostics, audit logs. But the human that sits on top of them with the premium of accountability, trust and translation is going to do things like manage permissions, consent and audit governance. They're going to resolve device and integration issues, they're going to translate between clinical and IT requirements, and they're going to act as institutional accountability for the data layer. Now, it's likely that you'll have this type of role, be able to support a far larger number of people, maybe one specialist per thousand or two thousand patients. But even in those scenarios, you're talking about net new 20,000, 30,000, 40,000 jobs. And I want to be clear again that there is no guarantee that this is the way that healthcare plays out. There's obviously a million factors other than just the affordability of the care. But I also don't think that this is implausible. And I do think it is the type of change that you're going to see AI enable where AI unlocks vast amounts of new data, new types of data coordination, new types of monitoring, access, provisioning, and the system will over time adapt to accommodate those new opportunities. And whether it's these jobs or other types of jobs, there will be new jobs that support the new shape of those industries. Some of them will be transitional and eventually done by the AGI as well. But others will have different categories of that human premium and will become a durable integral new role in the new AI economy. As AI changes industry paradigms, it won't just create individual spot jobs, but ecosystems of new roles that come to meet the new challenges of those new paradigms. Now I picked healthcare, but it doesn't take all that much imagination to see similar patterns in other sectors. We talked about small business professional services where human plus AI service operators could deliver smaller, cheaper, more frequent professional services to businesses that weren't agency clients before. It's not hard to imagine that people would consume a lot more legal services if they were both affordable and we might even see more possibility unlocks around things like preventative legal maintenance that provides ongoing support instead of crisis only engagement. Education is another area where we could see always on personalized learning plus human pathway guidance AI provides content humans help learners persist, choose paths, improve competence, and so on and so forth. In mental health we could see a broader support layer between nothing and licensed therapy, peer support, group facilitation, continuous check ins with clear escalation. Personal finance could see once again continuous financial life support, taxes, benefits, insurance, debt, retirement, family finances with AI doing analysis and human zoning judgment and follow through elder childcare family support where AI can solve huge amounts of coordination and data issues but is not going to solve putting trustworthy people in the room with your children or parents. Across all of these you start to see new broad categories of roles navigators that help people enter and move through systems that are too complex to face alone. Continuous support workers who provide ongoing human support around AI monitored systems AI augmented service operators who use AI to deliver cheaper versions of professional services to new market tiers data and operations specialists who make AI enabled service models reliable in real institutional systems QA safety and compliance roles who ensure AI mediated services are safe, auditable, legal, fair and reliable. And the sixth and final family of roles are escalation specialists who handle the hardest cases that AI routes upward. My argument in a nutshell then is that the AI jobs question, which is really an AI economy question, can't be discussed only in terms of labor supply. It also has to be discussed in terms of demand. The second part of my argument is that demand stretches. It can stretch because of price, access, complexity, continuity, personalization and relational value. The unlocks that stretch demand sometimes fall into the affordability category where existing services reach new buyers, while others fall into the possibility category where new service models become viable for the first time. And as those new opportunities come online. The reason that AGI doesn't just do it all is the human premium. These at least seven categories of value that require human delivery, relationship, presence, trust, accountability, translation, behavior change and provenance. And of course when you combine human premium plus those unlocks in new economic systems you get new types of roles navigators, continuous support workers, AI augmented service operators, data and ops specialists, QA safety and compliance and escalation specialists. Simply put, we are heading to a world where there is more AI, which yes, does create more supply of labor. But as that increased supply of labor changes the cost and availability structure of other industries, it also means more demand. And more demand, even in a scenario of AGI, means more human work ahead too. My argument when it comes to AI and jobs has always been that in the short term there are going to be painful disruptions that require meaningful interventions and that we should not be Pollyannish or unconsiderate about. But that on the other side I think things are massively better and even bigger than they are today. I hope this episode goes a little way to showing what I see on the other side and why I'm so genuinely bullish about what comes in the future. I'm sure this is a theme we will continue to explore, but for now that's going to do it for today's AI Daily Brief. Appreciate you listening or watching. As always and until next time, peace.