The New Jobs AI Will Create - CourtneyR.dev's 2nd Brain

The New Jobs AI Will Create 🔗 **Source:** [The New Jobs AI Will Create](https://www.youtube.com/watch?v=WhAAKxPlMhw) > [!note]+ Document Notes > tags: #artificial-intelligence #future-of-work #economic-transformation #job-creation #healthcare #demand-elasticity #human-premium > author: Nathan Lambert > source: YouTube > reference: https://www.youtube.com/watch?v=WhAAKxPlMhw > date created: 2024-12-19 10:30 > > # 🏷️ Title: The New Jobs AI Will Create >  ## 🧰 Action Inbox - [H1] Research demand elasticity patterns in your industry — identify which of the 6 types apply #artificial-intelligence - [H2] Map human premium categories to current role — assess which values transfer beyond AI capability #future-of-work - [H3] Explore continuous service models in your sector — look for possibility unlock opportunities #economic-transformation ## 🧭 Table of Contents - [[#📝 Summary]] - [[#📖 Main Content]] - [[#🗒️ Notes]] - [[#📚 References]] - [[#🗣️ Paraphrase]] - [[#📰 Newsletter]] - [[#✅ Action Items]] - [[#🧪 Hypotheses]] - [[#📊 Methodology]] - [[#❓ Questions]] - [[#🔮 Future Research]] - [[#💡 Implications]] - [[#🧠 Learning Styles Assessment]] - [[#📊 Accessibility Assessment]] - [[#🌍 DEIB Assessment]] - [[#✅ Logical Validity]] - [[#🛡️ DARVO Analysis]] - [[#🎯 Conclusion]] ### 🔗 References - [AI Daily Brief](https://www.youtube.com/watch?v=WhAAKxPlMhw) - [Alex Emas "What Will Be Scarce" Essay](URL) ## 📝 Summary - Nathan Lambert challenges the AI job apocalypse narrative by arguing AI will create new categories of employment through demand expansion - The hidden assumption in job displacement theories is that demand stays constant while AI increases labor supply - Six types of demand elasticity explain how AI enables market expansion: price, access, complexity, continuity, personalization, and relational - The "human premium" - seven categories of value requiring human delivery - protects many roles from AGI replacement - Healthcare case study demonstrates how continuous preventative care could create 276,000 to 1.2 million new jobs - Bottom line: AI creates both affordability unlocks (cheaper existing services) and possibility unlocks (entirely new service categories) ## 📖 Main Content Lambert argues the AI industry has focused almost exclusively on which jobs are at risk rather than exploring what new employment categories will emerge. This represents a fundamental analytical failure that leaves optimists with only vague handwavy arguments about job creation. The core flaw in job apocalypse narratives is treating AI purely as a labor supply story - more AI means cheaper labor and displaced workers. This assumes demand remains constant, embodying the "lump of labor fallacy" that there's a fixed amount of work to do. Lambert identifies six types of demand elasticity that explain how increased AI-powered supply can actually expand markets: price elasticity (cheaper access), access elasticity (reducing scarcity and barriers), complexity elasticity (making opaque systems navigable), continuity elasticity (enabling always-on support), personalization elasticity (custom solutions), and relational elasticity (human-centered value). These elasticities create two types of unlocks: affordability unlocks where existing services reach new buyers at lower prices, and possibility unlocks where entirely new service models become operationally viable for the first time. ## 🗒️ Notes #### ⏱️ [00:00](https://youtu.be/WhAAKxPlMhw?t=0) — [H1] AI jobs narrative shift beginning - Key idea: Subtle movement away from pure job apocalypse thinking - Evidence: "More people at least question the premise that AI and agents getting better means people will have less work" #### ⏱️ [02:30](https://youtu.be/WhAAKxPlMhw?t=150) — [H2] Industry failure to explore new job categories - Key idea: Massive imbalance between risk assessment and opportunity exploration - Evidence: "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" #### ⏱️ [04:15](https://youtu.be/WhAAKxPlMhw?t=255) — [H1] Hidden assumption in job displacement narratives - Key idea: AI analyzed only as labor supply increase, not demand expansion - Evidence: "All of this rests upon the presumption that demand stays constant" #### ⏱️ [06:45](https://youtu.be/WhAAKxPlMhw?t=405) — [H2] Six types of demand elasticity - Key idea: Different ways AI-increased supply can expand market demand - Evidence: "Price elasticity, access elasticity, complexity elasticity, continuity elasticity, personalization elasticity, relational elasticity" #### ⏱️ [09:20](https://youtu.be/WhAAKxPlMhw?t=560) — [H2] Complexity elasticity example - Key idea: AI helps navigate opaque systems people avoided - Evidence: "Think help with taxes, insurance, immigration, benefits... making opaque systems navigable" #### ⏱️ [12:10](https://youtu.be/WhAAKxPlMhw?t=730) — [H2] Small business professional services unlock - Key idea: Affordability unlock activates long-tail market - Evidence: "If AI reduces the cost of those jobs from 5,000 to 500 and 3,000 to 300, an enormous longtail market activates" #### ⏱️ [15:30](https://youtu.be/WhAAKxPlMhw?t=930) — [H1] Possibility unlock vs affordability unlock - Key idea: Two distinct ways AI creates new markets - Evidence: "Affordability unlock is the same menu at a lower price, the possibility unlock is a new menu entirely" #### ⏱️ [18:45](https://youtu.be/WhAAKxPlMhw?t=1125) — [H1] AGI objection and capability vs service design - Key idea: Jobs question is not just about AI capability but service delivery requirements - Evidence: "Better question I believe is does AI only delivery satisfy the demand?" #### ⏱️ [21:00](https://youtu.be/WhAAKxPlMhw?t=1260) — [H1] Seven categories of human premium - Key idea: Value that remains attached to human involvement despite AI capability - Evidence: "Relationship, embodied presence, trust, accountability, translation, behavior change, provenance and status" #### ⏱️ [24:30](https://youtu.be/WhAAKxPlMhw?t=1470) — [H2] Healthcare has every demand elasticity type - Key idea: Healthcare demonstrates all six elasticity categories - Evidence: "Healthcare has potentially every type of demand elasticity that we discussed before" #### ⏱️ [27:15](https://youtu.be/WhAAKxPlMhw?t=1635) — [H2] Continuous preventative care paradigm - Key idea: AI enables shift from episodic reactive to continuous preventative healthcare - Evidence: "AI enabled version of healthcare that is continuous, preventative and personalized" #### ⏱️ [30:00](https://youtu.be/WhAAKxPlMhw?t=1800) — [H2] Continuous care navigator role - Key idea: Human layer between patients and AI monitoring systems - Evidence: "276,000 jobs in conservative case, 1.2 million in aggressive case" #### ⏱️ [34:20](https://youtu.be/WhAAKxPlMhw?t=2060) — [H3] Care plan outcome specialist role - Key idea: Human interface for care plan implementation and problem-solving - Evidence: "Owns the gap between medical advice and real world execution" #### ⏱️ [37:45](https://youtu.be/WhAAKxPlMhw?t=2265) — [H3] Health data operations specialist - Key idea: Technical role ensuring reliability of AI-enabled healthcare data - Evidence: "20,000, 30,000, 40,000 jobs... owns reliability, integration, governance" #### ⏱️ [41:10](https://youtu.be/WhAAKxPlMhw?t=2470) — [H3] Six broad categories of new roles - Key idea: Patterns of job creation across industries - Evidence: "Navigators, continuous support workers, AI augmented service operators, data and operations specialists, QA safety and compliance, escalation specialists" ## 📚 References - [AI Daily Brief YouTube Channel](https://www.youtube.com/@aidailybrief) - [Alex Emas Research on Relational Economy](URL) - [Labor Economics Literature on Technological Change](URL) ## 🗣️ Paraphrase - AI job discussions focus too heavily on displacement risk and ignore job creation opportunities through market expansion - The assumption that demand stays constant while AI increases labor supply is fundamentally flawed and ignores historical patterns - Six types of demand elasticity show how AI can expand markets: making things cheaper, more accessible, less complex, continuous, personalized, and more relational - Human premium categories like trust, accountability, and relationship preserve job value even when AI can perform underlying tasks - Healthcare demonstrates how AI could create hundreds of thousands of new jobs in continuous care, implementation support, and data operations roles ## 📰 Newsletter **AI's Job Creation Opportunity: Moving Beyond Displacement Fears** The AI job debate has been stuck in apocalypse mode, but Nathan Lambert offers a refreshing economic perspective that could reshape how we think about AI's impact on work. Lambert argues we've been asking the wrong question. Instead of just "Can AI do this job?", we should ask "Does AI-only delivery satisfy market demand?" This shift reveals how AI creates new employment through demand expansion, not just job displacement. Key insight: Six types of demand elasticity show how AI enables market growth - making services cheaper, more accessible, less complex, continuous, personalized, and relationship-focused. Healthcare exemplifies this with potential for 276,000 to 1.2 million new jobs in continuous preventative care. The "human premium" - seven categories of value requiring human delivery including trust, accountability, and relationships - protects many roles from AI replacement while creating entirely new job categories. **Resources:** [Learn WordPress](https://learn.wordpress.org/), [Developer Docs](https://developer.wordpress.org/), [WP-CLI](https://wp-cli.org/) ## ✅ Action Items #### DevRel - Developer advocacy: Create content on AI-augmented developer workflows, present at conferences on human-AI collaboration in coding - Education: Develop workshops on prompt engineering for developers, AI pair programming best practices - Developer experience: Research how AI tools can improve documentation, SDK usage patterns - Community management: Survey developer attitudes toward AI tools, moderate discussions on AI impact #### OSPO - Strategic Alignment: Track AI adoption patterns in open source projects, measure impact on contribution models - Developer Support: Provide training on AI-assisted development, evaluate AI coding tools for enterprise use - Community Engagement: Fund research on AI's impact on maintainer workload, open source sustainability #### Community - Contributors: Create pathways for AI-assisted contributions, first-issue programs leveraging AI tools - Ecosystem: Coordinate cross-project standards for AI tool integration in development workflows #### WordPress - Core AI: Develop guardrails for AI-generated content, create evaluation frameworks for AI features - Training: Design lesson plans for AI-assisted WordPress development, content creation workflows - Community: Support meetups on AI tools for WordPress developers, content creators - Documentation: Use AI to improve DevHub content generation while maintaining human editorial oversight ## 🧪 Hypotheses - Industries with high complexity elasticity (legal, healthcare, finance) will see more AI-human hybrid roles emerge than those with primarily price elasticity - Roles with multiple human premium categories (trust + accountability + relationship) will be more resistant to AI displacement than those with single premiums - Possibility unlocks will create larger employment categories than affordability unlocks in sectors where continuous service models weren't previously viable - The transition period between affordability and possibility unlocks will determine whether AI adoption creates net job growth or displacement in specific sectors ### 📊 Methodology - Industry analysis comparing current service delivery models to AI-enabled continuous models across healthcare, legal, education, and financial services - Survey research on consumer willingness to pay for human premium categories in different service contexts - Economic modeling of demand elasticity coefficients for professional services at different AI-driven price points ### ❓ Questions - How do we measure and validate the seven human premium categories across different industries and cultural contexts? - What policy interventions could accelerate the transition from displacement to job creation during AI adoption waves? - How might the timeline and sequence of AI capabilities development affect which possibility unlocks become viable first? ### 🔮 Future Research - Longitudinal study tracking job category evolution in early AI-adopting industries like customer service and content creation - Cross-cultural research on human premium valuation differences across global markets - Economic impact modeling of continuous service paradigms across multiple sectors beyond healthcare ### 💡 Implications - Practice: Service businesses should identify which demand elasticity types apply to their market and design AI-human hybrid offerings accordingly - Policy: Workforce development programs need to focus on human premium skills rather than just technical AI literacy - Education: Business schools should teach demand elasticity analysis as core framework for understanding AI economic impact - Funding: Investors should evaluate AI startups based on their ability to unlock new demand categories, not just automate existing processes ### 🧠 Learning Styles Assessment Reading level estimate: Grade 14-15 (College/Graduate level) #### 👀 Visual learners - Present: Economic frameworks clearly explained with categorization - Improve: Demand elasticity diagrams, healthcare system flow charts, job growth visualizations #### 🔊 Aural learners - Present: Full video presentation with clear verbal explanation - Improve: Podcast-style discussion with industry experts on specific sectors #### 🗣️ Verbal learners - Present: Strong use of analogies (menu metaphors), clear definitions of technical terms - Improve: More concrete industry examples beyond healthcare #### 🧩 Kinesthetic learners - Present: Detailed case study walkthrough with specific calculations - Improve: Interactive worksheets for analyzing demand elasticity in different industries #### 🧠 Logical learners - Present: Systematic framework with numbered categories, clear cause-effect relationships - Improve: Economic modeling tools, quantitative analysis templates #### 🧑‍🤝‍🧑 Social learners - Present: YouTube format encourages discussion and community engagement - Improve: Discussion guides for industry groups, peer analysis exercises #### 🧘‍♀️ Solitary learners - Present: Comprehensive framework for independent analysis - Improve: Self-assessment tools for evaluating personal career implications #### 🌿 Naturalistic learners - Present: Healthcare ecosystem thinking, economic system interconnections - Improve: Broader connections to economic ecosystem health, sustainability implications ### 📊 Accessibility Assessment #### Screen reader semantics - Present: Clear verbal presentation in video format - Improve: Transcript with proper heading structure, alt text for any visual frameworks #### Keyboard navigation - Present: Standard YouTube video controls - Improve: Chapter markers for easy navigation to specific topics #### Contrast and readability - Present: Clear audio delivery - Improve: High-contrast transcript formatting, readable font sizes #### Media and captions - Present: YouTube auto-generated captions available - Improve: Professional captions with technical term accuracy, downloadable transcript ### 🌍 DEIB Assessment #### Cultural breadth - Present: Economic framework applicable globally - Improve: Examples from different cultural contexts, consideration of varying labor market structures #### Inclusive language - Present: Professional, accessible economic terminology - Improve: More diverse industry examples beyond healthcare, varied demographic scenarios #### Neurodiversity support - Present: Clear logical structure, systematic framework presentation - Improve: Multiple format options (audio, text, visual), consistent pacing #### Representation - Present: Focus on economic analysis rather than demographic specifics - Improve: Examples considering different socioeconomic situations, global labor market variations ### ✅ Logical Validity #### Strong reasoning - Deductive strength: Solid logical progression from demand elasticity principles to job creation predictions - Inductive strength: Good use of historical precedent regarding technology adoption and job creation - Causal reasoning: Clear causal links between AI capability improvements and demand expansion mechanisms - Analogical strength: Effective use of menu analogy for understanding affordability vs possibility unlocks #### Fallacies - Hasty generalization: Healthcare case study generalized to all industries without sufficient evidence for universal applicability, needs more cross-sector validation - False dichotomy: Presents AI impact as either displacement or creation, could better address mixed scenarios where both occur simultaneously #### Tone of voice - Overall tone: Professional and optimistic while acknowledging complexity, uses --- ## Full Transcript [00:00] Today on the AI Daily Brief, the new [00:02] jobs AI will create. The AI Daily Brief [00:05] is a daily podcast and video about the [00:07] most important news and discussions in [00:08] AI. [00:10] All right, friends. One of the big [00:12] points of discussion for this week has [00:13] been the idea that there is this ever so [00:15] subtle shift in the AI jobs narrative. [00:18] It would be going way too far to say [00:20] that this was becoming anywhere near [00:21] mainstream, but you're starting to see [00:23] more people at least question the [00:24] premise that AI and agents getting [00:26] better means people will have less work. [00:28] And yet, even among those who are [00:29] arguing that the AI job apocalypse [00:32] narrative is way overblown, the argument [00:34] tends to be much more backwards looking. [00:36] It's looking at the way that [00:38] productivity has previously impacted [00:39] industries and extrapolating that out to [00:41] the future. That's well and good and all [00:44] of that is important, but I think it is [00:46] a significant failure of the AI industry [00:48] to take the next step and actually start [00:50] to explore the type of jobs that there [00:53] will be in an AI enabled future. The [00:56] sheer tonnage of time spent on assessing [00:59] which jobs are most at risk, compared to [01:01] the almost zero time exploring what [01:04] types of new jobs will be created [01:06] represents one of our great failures and [01:08] leaves people who want to be optimistic [01:10] about the future clinging to vague [01:12] handwavy notions about what those jobs [01:14] might be. Now, from the standpoint of [01:16] sheer epistemic humility, of course, we [01:18] have to be careful about arguing with [01:20] any sort of confidence what specific [01:22] things will be created in the future. In [01:24] other words, the more specific our [01:26] future predictions, the less likely to [01:27] be right they are. But that doesn't mean [01:29] that we can't at least explore from [01:31] first principles how we think the change [01:33] is going to play out if indeed it is not [01:35] going to be the mass destruction of [01:36] white collar work that some are [01:38] promising. So what I'm going to attempt [01:40] today is talk about one the fundamental [01:42] problem and the hidden assumption in [01:44] most of the job apocalypse narratives. [01:46] Two, what opens up when you correct that [01:49] assumption. Three, how to deal with the [01:51] AGI objection. and four, a real walk [01:55] through a specific sector and some jobs [01:57] that I could see very plausibly being a [01:59] part of the future and representing [02:00] meaningful amounts of employment. So, [02:02] let's talk first about that hidden [02:04] assumption. AI is mostly analyzed right [02:06] now as a labor supply story. In other [02:09] words, AI increases the supply of labor, [02:12] so labor gets cheaper and workers get [02:14] displaced. But all of this rests upon [02:16] the presumption that demand stays [02:19] constant. This is another way of framing [02:21] the lump of labor fallacy. the idea that [02:23] there is a set amount of work to do. So [02:25] if technology comes and does some of [02:27] that work, that means that humans aren't [02:29] doing that work, which means that those [02:30] humans don't have any work to do. What [02:32] it doesn't take into account is the [02:34] expansionary nature of economies and the [02:36] ability of demand to expand to absorb [02:38] more supply. Now, I don't want to spend [02:41] a lot of time on the historical [02:42] arguments here. As I mentioned, that's [02:44] the main substance of pretty much all [02:46] other arguments along these lines. Yet, [02:48] it is worth noting, of course, that the [02:49] assumption that demand stays constant as [02:51] the supply of labor increases has never [02:53] held true. And so, if we are just going [02:55] from historical precedent, it is [02:57] unlikely to be true this time either. [02:59] The more important question, if we're [03:01] trying to actually interact with [03:02] people's lives as they're lived now, not [03:04] just provide a history lesson, is to ask [03:07] where demand will expand. Basically, [03:10] what parts of the economy could actually [03:13] grow as AI allows us to produce more? [03:16] And I think what's important to note is [03:17] that demand elasticity actually comes in [03:20] a variety of different forms. The one [03:22] that's most obvious is of course price [03:23] elasticity. I.e. I wanted the thing but [03:26] it costs too much. If AI makes a thing [03:29] cheaper, new buyers can enter the [03:30] market. A second category of demand [03:33] elasticity is access elasticity. I [03:36] wanted it but I couldn't get it. Now, [03:39] whereas AI supported price elasticity [03:41] because it lowered the cost floor, in [03:43] this case, we're talking about AI [03:45] reducing provider scarcity or weight [03:47] times or geographic barriers or other [03:50] types of institutional bottlenecks. [03:52] Basically, instead of it just being [03:54] cheaper, AI makes a good or service more [03:56] readily available. A third type of [03:58] demand elasticity is complexity [04:00] elasticity. I needed it, but the system [04:02] was too confusing. Think help with [04:04] taxes, insurance, immigration, benefits. [04:07] The way that AI supports this type of [04:08] demand elasticity is by making opaque [04:11] systems navigable. You're already [04:13] starting to see this as a major way that [04:15] people are consuming AI, interacting [04:17] with legal ease or contract things or [04:19] even doctor's recommendations they [04:20] didn't really understand before, but now [04:22] have the sherpa and guide that can help [04:23] reduce that complexity. Another category [04:26] of elasticity we'll call continuity [04:28] elasticity. And here perhaps we're [04:30] getting a little bit more abstract, but [04:31] I still think it's worth noting. The [04:33] idea of continuity elasticity is going [04:35] from I get help occasionally to I'd [04:37] benefit from help all the time. So think [04:40] about areas like health or coaching [04:42] where always on monitoring and support [04:44] could become cheap enough to operate at [04:46] scale in a way that wasn't possible [04:48] before. A fifth form of demand [04:50] elasticity we'll call personalization [04:52] elasticity. This is I get the generic [04:55] version now but I'd value something made [04:57] for me. When AI brings the cost and ease [05:00] of production down, it's going to make [05:02] customization cheaper. The last category [05:05] we'll call relational elasticity or [05:07] value elasticity. And the idea is I want [05:10] this thing that I am buying, whether [05:11] it's a good or a service to be more [05:13] human, meaningful or trusted. This is [05:15] what Chicago Booth economist Alex Emas [05:18] wrote about in his what will be scarce [05:19] essay. how the relational sector where [05:22] the provenence of a good or service how [05:24] it was made and who made it and who is [05:26] providing it are actually an integral [05:28] part of the value of the service i.e. [05:30] people deciding that they don't just [05:32] want a coffee, they want an experience [05:33] surrounding that coffee that involves [05:35] interaction with a human and a [05:36] particular quality that comes from some [05:38] sort of human touch. So, we've got these [05:40] different categories of how AI's impact [05:43] on the supply of things in terms of cost [05:45] of production, ease of production, speed [05:47] of production, etc. can interact with [05:49] elasticity of demand to expand demand in [05:52] new and clear ways. And one thing that's [05:54] interesting to look at as we try to [05:56] understand AI's impact on different [05:57] sectors, especially the positive and [05:59] generative impact of AI, is to ask which [06:02] of those elasticities are common in [06:04] those sectors. They're present pretty [06:06] much everywhere, but different sectors [06:08] are going to have different elasticities [06:09] more strongly represented. For example, [06:12] in experiences and travel, that [06:14] relational elasticity and [06:15] personalization elasticity are going to [06:17] be particularly high, but you're not [06:18] likely to see continuity elasticity [06:20] because you're not able to have [06:21] experiences or travel continuously [06:23] throughout your life, as awesome as that [06:24] would be. Meanwhile, on the other end of [06:26] the spectrum, healthcare, I think, has [06:28] every single one of these types of [06:30] elasticity, which is why it shows up as [06:32] our case study in a little bit. Now [06:34] trying to zoom out and consolidate this [06:37] argument, I think we can look at AI as [06:39] expanding the demand frontier in two [06:41] directions at once. It either is [06:44] increasing and democratizing access to [06:46] things that are available now or [06:48] unlocking things that aren't possible [06:49] now. If you go back up to the types of [06:52] elasticity, you'll note that in many of [06:53] these cases, if you were currently rich, [06:56] you could already get access to the [06:58] thing in the way that the people want it [06:59] now under the conditions of the new [07:01] elasticity. If you're rich, your word is [07:03] price sensitive. You could probably pay [07:04] to solve access issues and complexity [07:06] issues. If you're rich enough, you could [07:08] even use money to solve continuity [07:10] issues, and you could probably pay for [07:11] personalization and relational value as [07:13] well. So, the first unlock of AI is the [07:16] affordability unlock. Basically, the [07:18] same menu, but at a lower price. And [07:20] that's most of what you were probably [07:21] thinking about when I was describing [07:22] these six categories of demand [07:23] elasticity. To take one example, let's [07:26] think about small business professional [07:27] services. Right now, the average small [07:30] business owner might not have the budget [07:32] to buy a $5,000 design project, a $3,000 [07:35] marketing campaign, a $2,000 legal [07:37] review, and a $1,500 analytics report. [07:40] This obviously does not mean that the [07:41] business doesn't have demand for design, [07:43] marketing, legal, and analytics. It [07:44] means that at the current state, the old [07:47] version of those services was [07:48] unaffordable. If AI reduces the cost of [07:51] those jobs from 5,000 to 500 and 3,000 [07:54] to 300, an enormous longtail market [07:57] activates. Millions of small businesses [07:59] that were never agency clients before [08:01] become buyers for the first time. This [08:03] is again what I'm calling the [08:04] affordability unlock where there is net [08:07] new demand for services that already [08:08] exist because the changing cost and [08:10] availability of provision opens up new [08:12] buyers for which that service wasn't [08:13] viable before. The second type of unlock [08:16] I'm calling the possibility unlock. To [08:18] use the menu analogy, if the [08:20] affordability unlock is the same menu at [08:22] a lower price, the possibility unlock is [08:24] a new model entirely. This is where AI [08:26] makes a service model operationally [08:28] possible that could not previously exist [08:31] at scale. The category itself becomes [08:34] possible before anyone can demand it. [08:37] Previewing where we're going to get in [08:38] our case study, an example of this, I [08:40] think, is continuous preventative [08:42] healthare. The vast majority of people, [08:44] not named Brian Johnson, don't currently [08:46] have someone continuously watching their [08:48] health data or tracking their care plan [08:50] or noticing drift or coordinating with [08:52] their pharmacy. And to be clear, people [08:54] aren't demanding this exact service. [08:56] This isn't a normal category because it [08:57] hasn't been operationally viable. But as [09:00] AI collapses the cost of the [09:02] informationational layer around care, [09:04] data collection, monitoring, [09:06] summarization, documentation, [09:08] scheduling, escalation, routing, once [09:10] all of those costs fall and the systems [09:12] around them become viable, a totally [09:14] different healthcare model becomes [09:16] possible. One that is personalized, [09:18] datadriven, preventative, and [09:19] continuous. This is entirely net new [09:23] demand because no one had this before [09:25] because it wasn't possible until AI made [09:28] it possible. When it comes to new jobs, [09:31] the affordability unlock opens up whole [09:33] new categories of providers that cater [09:35] to different parts of the long tale of [09:37] the market. The possibility unlock [09:39] creates entire new markets which entire [09:42] new actors will come into to provide [09:43] those goods and services that are now [09:45] actually possible. But now we have to [09:48] get to the question that stops all of [09:50] these conversations midtrack. I would [09:53] argue that the number one reason that [09:55] people claim this time is different and [09:57] that AI is not a so-called normal [10:00] technology is that they say the people [10:02] that are optimistic about the jobs [10:04] created in the future aren't answering [10:06] the question, won't AGI just eat those [10:09] jobs, too? In other words, for those [10:12] small business professional services, if [10:14] the first wave of AI brings cost down by [10:16] a factor of 10 and makes those $5,000 [10:18] jobs into $500 jobs, is that just a [10:21] temporary state on the path to AGI that [10:24] can do those jobs for 50 cents instead [10:25] with no human involved? To be clear, [10:29] from a first principles perspective, [10:31] this is a reasonable question and one [10:33] that again I think AI optimists are [10:35] failing to engage with clarity and good [10:37] faith. For my part, I think that there [10:40] are actually quite clear answers. The [10:43] question that we are asking, which I [10:44] think is the wrong question and the idea [10:46] that is embedded in the assumption that [10:48] AGI will just eat those new jobs too, is [10:51] we think about the question of jobs only [10:53] as a question of can AI perform the [10:55] task. It's a capability question that [10:58] assumes labor demand is grounded [10:59] entirely and only in what humans can do. [11:03] But that's too narrow. A better question [11:05] I believe is does AI only delivery [11:08] satisfy the demand? Here we've moved [11:11] from a capability question to a service [11:13] design question. Many roles exist not [11:17] because of capability gaps but because [11:19] of the constraints of the market's [11:21] expectations in how a service is [11:23] delivered. Things like trust, [11:25] accountability, presence, relationships. [11:27] They are part of the value. And while [11:29] AGI can eat tasks and will eat many [11:32] tasks, it will not automatically eat [11:34] that demand for things like trust and [11:36] accountability. To try to put a name on [11:38] this, I'm calling it the human premium. [11:41] These are seven categories of value that [11:43] do not a priority transfer when you [11:45] remove the human. The human premium is [11:48] the portion of economic value that [11:50] remains attached to human involvement [11:52] even when AI can perform the underlying [11:54] tasks. It is the answer to why doesn't [11:57] AI eat those jobs too. Here are seven [12:01] categories of human premium. The first [12:04] is relationship or relational. Again, [12:06] this is the one that Alex Emos explored. [12:08] The idea that I want this delivered by [12:10] someone who knows me, where there is [12:12] continuity, memory, accumulated trust, [12:15] where the human part of the experience [12:16] actually matters and is integral to the [12:18] experience, where replacing a person [12:21] with an AI makes the service worse [12:23] enough that it actually decreases demand [12:24] for that service. A second kind of [12:27] related category of human premium we'll [12:29] call embodied presence, where the demand [12:32] constraint is I want someone there with [12:34] me. This is where physical presence [12:37] matters. The nurse in the room, the [12:39] trainer correcting form. It's not that [12:42] AI can't do some of the things that [12:44] those people do. It's that the physical [12:47] presence is important enough to many [12:49] that there continues to be demand for [12:51] that human premium of embodied presence. [12:54] A third category of human premium is [12:57] trust. I want to talk to a person before [12:59] I act. Humans are social creatures. We [13:02] value social proof. We value the [13:04] experience of people who have come [13:05] before us. We value understanding the [13:07] context of the personal experience that [13:09] we are trusting of people who came [13:10] before us. As much as AI generates [13:12] recommendations, there are many [13:14] situations in which people will [13:16] absolutely prefer the recommendation of [13:18] a human, even if that human is just [13:20] validating the AI that they're working [13:22] with too. In many circumstances, humans [13:24] will be asked to make AI generated [13:26] recommendations believable, emotionally [13:29] acceptable, and actionable. Which gets [13:31] us to of course the fourth category of [13:32] human premium which is accountability [13:34] where the demand constraint is someone [13:36] has to own this. People want a person [13:39] who signs off, escalates, explains and [13:41] is responsible when things go wrong. [13:44] When I'm taking an experimental new [13:45] peptide regime, which is all the rage [13:47] right now, I don't care if Chat GBT can [13:49] tell me how to titrate things correctly [13:50] at the right ratios. You better believe [13:52] I want someone whose bonafides and [13:54] credentials I can check and who to put [13:56] it crassly I can sue if something goes [13:58] wrong. The questions of legal [14:00] accountability for AIS are going to make [14:02] a lot of lawyers very rich for a very [14:04] long time. But even outside of questions [14:05] of the legal system, there is going to [14:07] be a human premium on accountability for [14:10] sure. A fifth area of human premium is [14:13] translation. AI is interacting with the [14:15] entire accumulated knowledge of the [14:17] world. But some people aren't going to [14:18] know how to ask for what they need. Even [14:20] in a world of AGI, they will want human [14:23] translators. people who can turn messy [14:25] desire and constraint into usable AI [14:27] mediated work. This I think is extremely [14:31] important for the example of small [14:32] business professional services. The AGI [14:35] eats everything answer to this that [14:37] won't it just compress that $500 job [14:39] into a 50cent job eventually that takes [14:41] the job of the intermediary actor fails [14:43] to understand the economic value of this [14:45] translation. A small business owner [14:47] might not want or have time to get good [14:50] enough at using AI for design to take [14:53] advantage of the 50cent cost that AGI [14:55] makes it. The market will find the price [14:58] level at which a human can upsell the [15:00] 50cent AI work in a way that fits better [15:03] with the process and constraints of that [15:05] small business owner. It is a failure of [15:07] economic understanding to assume that [15:09] there is no room for margin on top of [15:11] the underlying cost of goods sold. AI [15:13] might dramatically compress that margin, [15:15] but there will be margin there. A sixth [15:18] category of human premium we'll call [15:19] behavior change. That's where I know [15:22] what to do, but I need help doing it. [15:24] Anyone who has tried to use a chat GBT [15:27] workout plan only to abandon it or at [15:30] least integrate it with a human trainer [15:32] who can help them stay accountable to [15:33] themselves will understand what I'm [15:35] talking about here. Simply put, there [15:38] are certain categories of behavior [15:39] change for which we will listen to a [15:40] human and we will not listen to an AI. [15:43] The last category of human premium we'll [15:45] call provenence and status. This is what [15:47] we were talking about in the relational [15:48] sector that a human made it is part of [15:50] why it matters. This could be arts, [15:52] crafts, custom goods, live performance, [15:54] bespoke services. The human signature is [15:57] part of those products. So with all of [15:59] these concepts in mind, let's look at an [16:01] industry case study in healthcare. I [16:03] want to be very clear. I have no [16:05] expertise in healthcare whatsoever. It's [16:07] not a sector I've worked in. I'm coming [16:09] at this from the perspective of a a [16:11] technologist and b a consumer of [16:13] healthare. But I think it's quite clear [16:15] that healthcare has potentially every [16:17] type of demand elasticity that we [16:19] discussed before. People consume more of [16:21] it if it was cheaper. They would [16:22] probably consume more of it if they had [16:24] easier access like better clinics in [16:26] their hometown. They would certainly [16:27] consume more of it if the complexity was [16:29] reduced or they had better ways to [16:30] navigate that complexity, especially [16:32] when it comes to billing, insurance, [16:33] etc. I think that likely many people [16:35] would want a relational element and a [16:37] personalized element of their healthcare [16:38] rather than just being a number in a [16:39] system. And I think the biggest demand [16:41] elasticity or at least the one that's [16:43] most interesting when it comes to the [16:44] idea of the difference between the [16:45] affordability unlock and a possibility [16:47] unlock is around continuity. My argument [16:51] is that we consume an incredibly small [16:53] part, even a vanishingly small part of [16:56] the amount of healthcare we would [16:57] consume if the labor math underneath it [16:59] made a different kind of healthcare [17:00] possible. Right now, we all live in a [17:03] paradigm of reactive care that is [17:05] episodic by default. Every once in a [17:07] while, we make an appointment. Either [17:08] because it's our annual checkup or [17:10] because there's a thing we're now at the [17:12] age where we have to do once every 3 to [17:13] 5 years or because something stops [17:15] working the way that it's supposed to. [17:17] From there, we have some tests, get a [17:19] diagnosis, which turns into a [17:21] prescription, and then we're left alone [17:22] again to fend for ourselves. At some [17:24] point, a new crisis arises, and the [17:26] whole cycle starts again. There is, I [17:28] believe, an AI enabled version of [17:30] healthcare that is continuous, [17:31] preventative and personalized. It [17:33] involves AI for data collection, [17:35] continuous monitoring and even triage [17:37] and ranking. But in doing so, it creates [17:40] a whole new set of opportunities for [17:42] human roles, net new human roles that [17:45] emerge in this new continuous [17:47] preventative personalized paradigm. [17:49] There's an AI background layer that [17:51] watches the data and a human action [17:53] layer that helps convert that to [17:54] meaning. So for the sake of this episode [17:57] and for coming along this journey with [17:58] me, let's assume that this idea that [18:00] continuous preventative care becomes [18:02] more viable and more normalized. If that [18:05] happens, it will create demand for new [18:07] types of jobs. And these are not AI jobs [18:10] in the narrow sense we think about [18:12] prompt engineers or agent operators. Now [18:14] these are healthcare service jobs that [18:16] the AI layer makes possible. And each of [18:18] them maps to some set of specific human [18:21] premium categories that protect them [18:23] from being eaten by stronger AI. The [18:26] first role I'll call a continuous care [18:28] navigator. This is the human layer [18:30] between a patient and an AI enabled [18:31] monitoring system. They oversee a case [18:34] load of patients in continuous [18:35] monitoring and handle the moments that [18:37] matter such as the pattern having [18:39] changed, the call that has to happen, [18:40] the family that needs reassurance or the [18:42] escalation that has to land. So while an [18:44] AI is doing things like ingesting data [18:47] from devices, labs, patient reports, [18:50] establishing baselines and detecting [18:52] deviations, ranking cases and changes by [18:54] urgency, updating documentation, routing [18:57] escalation, the human, the continuous [18:59] care navigator is doing things like [19:01] reviewing flagged cases, calling [19:03] patients to clarify and ask the [19:05] questions AI didn't know to ask in those [19:07] personal interactions, noticing things [19:09] like fear, shame, avoidance, or [19:11] unexpressed family dynamics. [19:13] coordinating with clinicians to escalate [19:15] things and closing the loop after [19:17] intervention. The human premiums [19:19] involved here include trust, [19:21] accountability, translation, behavior [19:24] change, and relationship. So, how many [19:26] people could the market support? When it [19:28] comes to healthcare, the total [19:30] addressable market is everyone with a [19:31] body. But for the sake of our arguments, [19:34] let's zoom out a decade and take a [19:36] conservative, middle, and aggressive [19:37] case among American adults. We'll call [19:39] conservative 40 million enrolled [19:40] patients, middle 80 million enrolled [19:42] patients, and aggressive 120 million [19:44] enrolled patients. The conservative case [19:46] then is about 14% of US adults enrolled [19:49] with the aggressive being about 43% of [19:51] US adults enrolled. Now, also on the [19:53] conservative end, let's assume that one [19:55] continuous care navigator can handle 150 [19:57] patients. If that's the case, you're [19:59] talking about about 276,000 [20:02] jobs. On the other end of the spectrum [20:04] where we've got 120 million enrolled [20:06] patients and we're assuming that they [20:08] can only handle a 100 people each, [20:10] you're talking about 1.2 million jobs. A [20:13] couple examples of jobs that are in that [20:15] 267,000 person range include personal [20:18] financial adviserss, loan officers, and [20:21] coaches and scouts. For the 1.2 million, [20:24] that's about the number of high school [20:25] teachers there are. And if you took [20:27] advertising, marketing, promotions, PR, [20:29] and sales manager, they combined come in [20:31] at around that number. meaning that [20:32] we're talking about a ton of people and [20:35] a ton of jobs. The next job in our [20:37] hypothetical example, we'll call a care [20:39] plan outcome specialist. This is the [20:41] person who owns the gap between medical [20:43] advice and real world execution. So, if [20:45] the continuous care navigator is the [20:47] person who's helping assess and escalate [20:49] issues as they come up, the care plan [20:51] outcome specialist is the person on the [20:53] other end of the cycle who helps with [20:55] the implementation layer of healthcare. [20:57] If AI is tracking each patient's care [20:59] plan and milestones and monitoring [21:01] medication adherence, tracking lab [21:03] screenings, etc., and integrating all of [21:05] that into their continuous care plan, [21:06] the care plan outcome specialist is the [21:09] person who can provide a human interface [21:11] for talking with patients about why the [21:12] plan isn't working or help solve [21:14] practical problems like cost [21:15] transportation or fear or family. Now, [21:18] I'll leave it to you to decide whether [21:20] there would be a smaller number of those [21:21] folks as relative to the continuous care [21:23] navigators, but you're still talking [21:24] about potentially hundreds of thousands [21:26] of net new jobs that don't exist today [21:28] that only exist because of the way that [21:29] AI changes the overall healthare system. [21:32] Finally, representing another category [21:34] of role that's likely to be fairly [21:35] ubiquitous is the new type of technical [21:38] roles that will emerge to support the AI [21:40] enabled economy. This one I'm calling a [21:42] health data operations specialist. The [21:45] premise is that continuous healthcare is [21:46] only as good as the data flowing through [21:47] it. Wearables, labs, pharmacy systems, [21:50] patient reported data, insurance [21:51] records. The health data operations [21:53] specialist owns reliability, [21:55] integration, governance, and clinical [21:56] usability of that layer. Now, AI is [21:59] going to be doing a lot of the work [22:00] around data. It's going to pull and [22:01] normalize data across sources. It's [22:03] going to flag anomalies and integration [22:04] failures. It's going to generate [22:06] pipeline diagnostics, audit logs. But [22:08] the human that sits on top of them with [22:10] the premium of accountability, trust, [22:11] and translation is going to do things [22:13] like manage permissions, consent, and [22:15] audit governance. They're going to [22:16] resolve device and integration issues. [22:18] They're going to translate between [22:19] clinical and IT requirements, and [22:21] they're going to act as institutional [22:22] accountability for the data layer. Now, [22:25] it's likely that you'll have this type [22:26] of role be able to support a far larger [22:28] number of people, maybe one specialist [22:30] per thousand or 2,000 patients. But even [22:32] in those scenarios, you're talking about [22:34] net new 20,000, 30,000, 40,000 jobs. [22:38] And I want to be clear again that there [22:40] is no guarantee that this is the way [22:42] that healthcare plays out. There's [22:43] obviously a million factors other than [22:45] just the affordability of the care. But [22:47] I also don't think that this is [22:48] implausible and I do think it is the [22:50] type of change that you're going to see [22:51] AI enable where AI unlocks vast amounts [22:55] of new data, new types of data [22:57] coordination, new types of monitoring, [22:59] access provisioning, and the system will [23:02] over time adapt to accommodate those new [23:05] opportunities. And whether it's these [23:07] jobs or other types of jobs, there will [23:09] be new jobs that support the new shape [23:11] of those industries. Some of them will [23:13] be transitional and eventually done by [23:14] the AGI as well, but others will have [23:16] different categories of that human [23:18] premium and will become a durable, [23:20] integral new role in the new AI economy. [23:22] As AI changes industry paradigms, it [23:25] won't just create individual spot jobs, [23:27] but ecosystems of new roles that come to [23:30] meet the new challenges of those new [23:32] paradigms. Now, I picked healthcare, but [23:35] it doesn't take all that much [23:36] imagination to see similar patterns in [23:38] other sectors. We talked about small [23:40] business professional services, where [23:42] human plus AI service operators could [23:44] deliver smaller, cheaper, more frequent [23:45] professional services to businesses that [23:47] weren't agency clients before. It's not [23:49] hard to imagine that people would [23:50] consume a lot more legal services if [23:52] they were both affordable. And we might [23:54] even see more possibility unlocks around [23:55] things like preventative legal [23:56] maintenance that provides ongoing [23:58] support instead of crisis only [23:59] engagement. Education is another area [24:02] where we could see always on [24:03] personalized learning plus human pathway [24:05] guidance. AI provides content. Humans [24:07] help learners persist, choose paths, [24:09] improve competence, and so on and so [24:11] forth. In mental health, we could see a [24:13] broader support layer between nothing [24:14] and licensed therapy, peer support, [24:16] group facilitation, continuous check-ins [24:17] with clear escalation. Personal finance [24:20] could see once again continuous [24:21] financial life support, taxes, benefits, [24:23] insurance, debt, retirement, family [24:25] finances with AI doing analysis and [24:27] human zoning judgment and followrough. [24:29] elder, child care, family support, where [24:31] AI can solve huge amounts of [24:33] coordination and data issues, but is not [24:35] going to solve putting trustworthy [24:36] people in the room with your children or [24:38] parents. Across all of these, you start [24:40] to see new broad categories of roles. [24:43] Navigators that help people enter and [24:45] move through systems that are too [24:47] complex to face alone. Continuous [24:49] support workers who provide ongoing [24:51] human support around AI monitored [24:53] systems. AI augmented service operators [24:56] who use AI to deliver cheaper versions [24:58] of professional services to new market [24:59] tiers. Data and operations specialists [25:02] who make AI enabled service models [25:04] reliable in real institutional systems. [25:06] QA safety and compliance roles who [25:08] ensure AI mediated services are safe, [25:10] auditable, legal, fair, and reliable. [25:12] And a sixth and final family of roles [25:14] are escalation specialists who handle [25:16] the hardest cases that AI routes upward. [25:19] My argument in a nutshell then is that [25:21] the AI jobs question which is really an [25:23] AI economy question can't be discussed [25:26] only in terms of labor supply. It also [25:28] has to be discussed in terms of demand. [25:30] The second part of my argument is that [25:32] demand stretches. It can stretch because [25:34] of price, access, complexity, [25:35] continuity, personalization and [25:37] relational value. The unlocks that [25:40] stretch demand sometimes fall into the [25:42] affordability category where existing [25:44] services reach new buyers while others [25:46] fall into the possibility category where [25:48] new service models become viable for the [25:49] first time. And as those new [25:51] opportunities come online, the reason [25:53] that AGI doesn't just do it all is the [25:55] human premium. These at least seven [25:57] categories of value that require human [25:59] delivery. Relationship, presence, trust, [26:01] accountability, translation, behavior [26:03] change, and providence. And of course, [26:05] when you combine human premium plus [26:07] those unlocks and new economic systems, [26:09] you get new types of roles. Navigators, [26:12] continuous support workers, AI augmented [26:14] service operators, data and ops [26:16] specialists, QA safety and compliance, [26:18] and escalation specialists. Simply put, [26:21] we are heading to a world where there is [26:23] more AI, which yes does create more [26:25] supply of labor, but as that increased [26:27] supply of labor changes the cost and [26:29] availability structure of other [26:30] industries, it also means more demand. [26:33] and more demand even in a scenario of [26:35] AGI means more human work ahead too. My [26:39] argument when it comes to AI and jobs [26:41] has always been that in the short term [26:43] there are going to be painful [26:45] disruptions that require meaningful [26:47] interventions and that we should not be [26:49] polyianish or unconsidered about but [26:51] that on the other side I think things [26:53] are massively better and even bigger [26:55] than they are today. I hope this episode [26:57] goes a little way to showing what I see [26:59] on the other side and why I'm so [27:01] genuinely bullish about what comes in [27:02] the future. I'm sure this is a theme we [27:04] will continue to explore, but for now, [27:06] that's going to do it for today's AI [27:08] daily brief. Appreciate you listening or [27:10] watching as always and until next time, [27:12] peace.