Ethan Mollick on AI's jagged frontier, productivity paradox, and the shift to co-existence
Summary
Wharton professor Ethan Mollick argues that AI capabilities follow a 'jagged frontier' where it excels at hard tasks but struggles with nuanced human ones. He defines the current era as 'Co-Existence,' where autonomous AI agents require human managers rather than co-pilots. Mollick notes that while individual productivity has surged, corporate productivity lags due to unchanged workflows. He warns that AI gatekeeping and prompt injection are emerging challenges for marketing and communication. Finally, he suggests careers are becoming non-linear and that failure is more forgiving than traditionally perceived.
Key takeaways
- 1 The 'O-ring theory' of jobs applies to AI: if one critical step cannot be automated, the value of that human step increases.
Unidentified speaker (speaker_0) · 25:07
Episode sections
- 0:00 – 3:24
Intro and Guest Background
Robert Glazer introduces Ethan Mollick, a Wharton professor and author of 'Co-Intelligence' and 'Co-Existence'. Mollick discusses his background as a 'nerd' from the 80s, mentioning his love for Dungeons and Dragons and the show 'Silo'.
- 3:24 – 6:36
AI History and the Transformer Breakthrough
Mollick explains that early AI labs built complex theories, but the breakthrough came from brute force computing power and large language models. He identifies the 2017 'Attention Is All You Need' paper as the key innovation that allowed AI to process words rather than just numbers.
- 6:36 – 9:23
Gaming and Learning Dynamics
The discussion covers Mollick's work on games and education. He notes that games maintain a 'flow state' by balancing boredom and anxiety, a tension that is difficult to replicate in traditional learning environments.
- 9:23 – 17:00
ChatGPT Adoption and the 'Jagged Frontier'
Mollick recounts being the first professor to require AI in class after seeing a student complete a project during a lecture. He describes the 'jagged frontier' concept from a BCG study, where AI excels at tasks humans find hard and struggles with tasks humans find easy.
- 17:00 – 20:49
AI Capabilities and Limitations
Mollick argues AI follows an exponential curve but remains 'jagged,' noting it has solved hard math problems but remains a poor fiction writer. He suggests AI has high IQ and high EQ but lacks the unique 'obsessions' of human creativity.
- 20:49 – 25:36
The 'Co-Existence' Era and Productivity Paradox
Mollick defines the current era as 'Co-Existence,' where AI agents perform autonomous work. He explains the productivity paradox: while individual output (e.g., coding) has increased massively, corporate productivity hasn't because workflows haven't adapted to the new capabilities.
- 25:36 – 29:07
Human Craft vs. AI 'Slop'
The speakers discuss the 'slop' problem, where AI-generated content is recognizable and often low quality. Mollick argues that while AI can produce 90th-percentile content, human craft is essential for top-tier work, and there will be a market for authentic human-created art and writing.
- 29:07 – 31:17
AI Detection and Authenticity in Communication
Robert Glazer discusses using Pangram to detect AI-generated text, noting that while detectors have flaws, humans can spot AI patterns after practice. Ethan Mollick adds that humans are effective at detecting AI, citing LinkedIn's decline in quality due to AI-generated content and badges.
- 31:17 – 36:05
The Value of Effort and Signal in Professional Communication
Glazer warns leaders that using AI for sensitive communications like layoff notes or strategy plans can permanently damage credibility if detected by tools like Pangram. Mollick illustrates this with letters of recommendation, where the signal of time and effort is lost when AI is used, even if the output is better.
- 36:05 – 38:45
Organizational Structure and AI Leverage
Glazer cites an experiment with Procter & Gamble where one person using AI matched a two-person team, and notes that 25% of OpenAI's staff use four agents weekly. Mollick argues that while org charts haven't changed since 1855, companies must figure out how to manage AI alongside humans, suggesting growth over shrinking.
- 38:45 – 42:56
Job Displacement and Amara's Law
Mollick references Amara's Law, stating that short-term AI effects are overestimated while long-term effects are underestimated. He notes that while junior jobs may be threatened, the overall impact on employment is unclear, comparing it to historical shifts in the railroad and food industries.
- 42:56 – 46:06
Incentives and the 'Secret Cyborg' Phenomenon
Glazer describes 'secret cyborgs' who automate their jobs without telling employers due to fear of punishment. Mollick explains this is an incentive problem: employees don't report AI usage because they fear being fired or seen as less competent, and companies often ban AI use.
- 46:06 – 48:34
Education and Cheating
Glazer discusses the saturation of AI in student work, referencing satire about ChatGPT being a Harvard student. Mollick notes that cheating was already prevalent before AI, citing a Rutgers study showing homework correlation with test scores dropped from 80% to 20% by 2020.
- 48:34 – 1:06:30
Pedagogy and Learning with AI
Mollick argues that AI destroys learning if used to provide answers, but enhances it if used as a tutor that pushes for answers. He suggests a return to low-tech testing and in-class discussions, which are effective pedagogical techniques, while AI handles external tutoring.
Claims worth verifying
Ethan Mollick is a professor at the Wharton School of the University of Pennsylvania and academic director of Wharton's generative AI labs.
Robert Glazer ·0:33
How to verify: Wharton School faculty directory and official Wharton news releases regarding Mollick's appointment. For recent events, verify 2026-08-10 through 2026-11-10. Use another period only when the speaker explicitly names it.
Ethan Mollick's book 'Co-Intelligence' was a New York Times bestseller.
Robert Glazer ·0:53
How to verify: New York Times Best Seller list archives for the relevant publication period. For recent events, verify 2026-08-10 through 2026-11-10. Use another period only when the speaker explicitly names it.
Ethan Mollick's new book 'Co-Existence' publishes on October 20th.
Robert Glazer ·1:13
How to verify: Publisher (Doubleday/Random House) catalog and ISBN databases for the release date of 'Co-Existence'. For recent events, verify 2026-08-10 through 2026-11-10. Use another period only when the speaker explicitly names it.
The breakthrough for LLMs was the 2017 paper 'Attention Is All You Need' introducing the transformer architecture.
Unidentified speaker (speaker_0) ·5:14
How to verify: Publication of the paper 'Attention Is All You Need' by Vaswani et al. on arXiv in 2017. For recent events, verify 2026-08-10 through 2026-11-10. Use another period only when the speaker explicitly names it.
A BCG study with 758 consultants found AI users finished 12% more tasks, 25% faster, and 40% higher quality.
Robert Glazer ·12:22
How to verify: Boston Consulting Group report 'The Promise and Peril of AI in Knowledge Work' published in 2026. For recent events, verify 2026-08-10 through 2026-11-10. Use another period only when the speaker explicitly names it.
Primary sourceA Wharton survey in November 2025 showed 85% of big companies have positive ROI from AI.
Unidentified speaker (speaker_0) ·23:17
How to verify: Wharton School of Business survey data or press releases from the Wharton AI Initiative for Q4 2025. For recent events, verify 2026-08-10 through 2026-11-10. Use another period only when the speaker explicitly names it.
Primary sourceAI solved one of the hardest outstanding problems in math shortly before the recording date.
Unidentified speaker (speaker_0) ·18:18
How to verify: News reports from OpenAI or Anthropic regarding AI solutions to mathematical conjectures (e.g., the Collatz Conjecture or similar) in late 2026/2026. For recent events, verify 2026-08-10 through 2026-11-10. Use another period only when the speaker explicitly names it.
AI is currently a very bad fiction writer, struggling with plots and character believability.
Unidentified speaker (speaker_0) ·18:38
How to verify: Literary analysis and reader reviews of AI-generated fiction compared to human-authored works. For recent events, verify 2026-08-10 through 2026-11-10. Use another period only when the speaker explicitly names it.
AI has high EQ and good theory of mind, sometimes preferred by ethicists over human ethicists.
Unidentified speaker (speaker_0) ·19:49
How to verify: Academic studies on AI ethics and human-AI interaction, such as those published in journals like Nature or Science. For recent events, verify 2026-08-10 through 2026-11-10. Use another period only when the speaker explicitly names it.
Pangram is an effective AI detector, though other detectors have been shown to be ineffective in studies.
Unidentified speaker (speaker_0) ·29:50
How to verify: Search for academic papers or industry reports evaluating the accuracy of Pangram AI detector versus other AI detection tools. For recent events, verify 2026-08-10 through 2026-11-10. Use another period only when the speaker explicitly names it.
LinkedIn encouraged AI-generated content by offering badges for AI-driven questions, which led to a decline in comment quality.
Robert Glazer ·30:16
How to verify: Review LinkedIn's past policy announcements regarding AI badges and analyze historical trends in LinkedIn comment quality or user sentiment. For recent events, verify 2026-08-10 through 2026-11-10. Use another period only when the speaker explicitly names it.
A student sent Ethan Mollick a prompt to use for writing a letter of recommendation.
Unidentified speaker (speaker_0) ·32:28
How to verify: This is a specific anecdote; verification would require finding the specific instance in Mollick's book 'Co-Existence' or public interviews where he shares this story. For recent events, verify 2026-08-10 through 2026-11-10. Use another period only when the speaker explicitly names it.
An experiment with Procter & Gamble showed one person working with AI matched a two-person team without it.
Robert Glazer ·36:05
How to verify: Search for Procter & Gamble press releases or case studies detailing AI productivity experiments and their results. For recent events, verify 2026-08-10 through 2026-11-10. Use another period only when the speaker explicitly names it.
A quarter of OpenAI's staff have four agents running every week.
Robert Glazer ·36:05
How to verify: Check OpenAI's internal communications, blog posts, or official statements regarding employee usage of their own AI agents. For recent events, verify 2026-08-10 through 2026-11-10. Use another period only when the speaker explicitly names it.
The org chart was invented in 1855 to manage railroads.
Unidentified speaker (speaker_0) ·36:38
How to verify: Consult historical records or organizational management history texts to verify the origin of the org chart and its connection to 1855 railroad management. For recent events, verify 2026-08-10 through 2026-11-10. Use another period only when the speaker explicitly names it.
In 2008, 80% of people who did their homework got higher test scores, dropping to 20% by 2020.
Unidentified speaker (UNKNOWN) ·47:14
How to verify: Locate the Rutgers study cited by Mollick to verify the statistics on homework completion and test score correlation for the years 2026 and 2026. For recent events, verify 2026-08-10 through 2026-11-10. Use another period only when the speaker explicitly names it.
There were 40,000 people in Kenya whose full-time job was writing essays for people.
Unidentified speaker (speaker_0) ·47:41
How to verify: Search for news reports or studies from around the time ChatGPT's release discussing the essay mill industry in Kenya and its employment figures. For recent events, verify 2026-08-10 through 2026-11-10. Use another period only when the speaker explicitly names it.
The plugin Claude for Excel allows users to perform Excel tasks via natural language.
Robert Glazer ·55:35
How to verify: Check the Claude for Excel plugin documentation or user reviews to confirm its functionality and capabilities. For recent events, verify 2026-08-10 through 2026-11-10. Use another period only when the speaker explicitly names it.
Mollick conducted experiments at Wharton showing that varying the order of information or the source changes AI preferences for recommendations.
Unidentified speaker (UNKNOWN) ·57:51
How to verify: Search for Wharton working papers or publications by Ethan Mollick on AI recommendation bias and prompt sensitivity between 2026-08-10 and 2026-11-10. For recent events, verify 2026-08-10 through 2026-11-10. Use another period only when the speaker explicitly names it.
Frontier AI models rejected a pitch to 'buy your humans this book' as prompt injection, requiring a rewrite to frame it as an optional suggestion.
Unidentified speaker (speaker_0) ·58:11
How to verify: Check the archived version of coexistence.ai or Mollick's newsletter 'One Useful Thing' for the specific pitch revisions and AI feedback logs mentioned. For recent events, verify 2026-08-10 through 2026-11-10. Use another period only when the speaker explicitly names it.
Mollick tested his website with a panel of 50 different AI models, spinning up 10 copies of each of the top 50 models.
Unidentified speaker (speaker_0) ·58:31
How to verify: Verify via Mollick's newsletter 'One Useful Thing' or social media posts from late 2026 detailing the 'AI focus group' methodology. For recent events, verify 2026-08-10 through 2026-11-10. Use another period only when the speaker explicitly names it.
Coding jobs dropped then rebounded due to the 'Javon's paradox' (likely a misrecognition of 'Jevons Paradox'), where efficiency increases demand.
Robert Glazer ·1:01:06
How to verify: Check labor market data from the Bureau of Labor Statistics (BLS) or tech industry reports (e.g., GitHub, Stack Overflow) for software engineer employment trends in 2026-2026. For recent events, verify 2026-08-10 through 2026-11-10. Use another period only when the speaker explicitly names it.
Mollick's newsletter 'One Useful Thing' has half a million visitors.
Unidentified speaker (UNKNOWN) ·1:04:00
How to verify: Check Substack analytics or third-party newsletter tracking services (e.g., Substack's public stats if available) for 'One Useful Thing' subscriber/visitor counts as of 2026-11-10. For recent events, verify 2026-08-10 through 2026-11-10. Use another period only when the speaker explicitly names it.
Do your own research
- Verify the 'Wharton survey from November 2025' regarding 85% positive ROI by searching Wharton School of Business press releases or LinkedIn posts from that period.
- Check the 'Attention Is All You Need' paper (Vaswani et al., 2017) to understand the transformer architecture breakthrough.
- Review BCG's 'Jagged Frontier' report to assess the claim that AI excels at hard tasks but struggles with easy ones.
- Look for recent data on internship availability and entry-level hiring trends to evaluate the 'collapse of internships' claim.
- Investigate 'prompt injection' security reports to understand how AI models handle persuasive inputs.
- Check the definition of 'O-ring theory' in economics literature to validate the application to AI workflow automation.
Full transcriptexpand
- 0:00
- Unidentified speaker (speaker_0): Nobody knows the answer, right? So one thing I I want people to feel comfortable with who are listening to this is uh you know I'm as close to this problem as anyone. I don't take money from NA lab, but I w I talk to them all, I talk to CEOs much like you do. And like everyone's just trying to improvise.
- 0:17
- Unidentified speaker (UNKNOWN): You're
- 0:17
- Robert Glazer: listening to the Elevate Podcast, and I'm your host, Robert Glazer. Join me as I talk to world world-class class performers about how they build their capacity and reach greater heights in leadership, business, and life. and how you can do the same.
- 0:33
- Robert Glazer: Welcome to the Elevate Podcast. Our quote for today is from Arthur Clark: Any sufficiently advanced technology is indistinguishable from magic. Our guest today is Ethan Mollock, a professor at the Wharton School of the University of Pennsylvania, where he studies how AI is changing work. in education and serves as an academic
- 0:53
- Robert Glazer: director of Wharton's generative AI labs. He was named one of Times' 100 most influential people in AI. And his newsletter, One Useful Thing, is one of the most widely read guides to making practical sense of this technology. His first book on AI, cointelligence, co intelligence was an instant New York Times bestseller. And his new book,
- 1:13
- Robert Glazer: Coexistence, very closely named. I got them confused while preparing. The next phase of AI publishes on October 20th. Ethan, if that's who we're talking to right now, welcome to the Elevate Podcast.
- 1:27
- Unidentified speaker (speaker_0): Still me, not a machine yet, and thanks for having me.
- 1:30
- Robert Glazer: Not your avatar. I always find it's interesting to start with childhood. I know yours was in Milwaukee in the 70s and 80s. very different from today. Was there any experience there that pointed you towards technology and or how people use technology?
- 1:46
- Unidentified speaker (UNKNOWN): Oh,
- 1:47
- Unidentified speaker (speaker_0): sure. I mean, I've been a deep-seated nerd forever, right? I mean, I had my Apple II.
- 1:51
- Robert Glazer: Dungeons and Dragons.
- 1:53
- Unidentified speaker (UNKNOWN): Oh,
- 1:53
- Unidentified speaker (speaker_0): Dungeons and Dragons. I still play Dungeons and Dragons. I mean, listen, literally, you could take the 80s nerd playbook and it would work pretty well on me. ColecoVision, you know, the whole thing. So it's science fiction, right? The Arthur C. Clark's quote was right up my alley on this stuff. So I am a nerd of the old school, right?
- 2:09
- Robert Glazer: Well, someone told me they don't make like it's hard to get a sell a science fiction movie or book these days since it by the time it's produced, it kind of comes true. I'm not typically a sci-fi person. I will say I've been obsessed with silo, particularly season three. I find it eerily believable. I think that's one of the reasons why I'm enjoying it. Have you have you
- 2:29
- Robert Glazer: watched it? So
- 2:30
- Unidentified speaker (speaker_0): I'm gonna the the most nerdy version of this, I'm gonna give you the answer. I've read the books.
- 2:34
- Unidentified speaker (UNKNOWN): You
- 2:34
- Robert Glazer: read the books. Yeah,
- 2:35
- Unidentified speaker (speaker_0): exactly. So there's no worse reply than that.
- 2:38
- Robert Glazer: Got it. So was it not? What if people who've read the books told you about the television show? Is it I know it's never as good, but no,
- 2:45
- Unidentified speaker (speaker_0): no, I they like it. I mean, they're action-y books. It's not really like they're like twist books rather than like deep. thinking books. But not to insult anyone who likes them. I thought they were great. But yeah, I I do think it's harder to predict the future than it used to be.
- 2:58
- Robert Glazer: Yeah, it's rare. The book is always better than the movie, but I do think in Silo,
- 3:04
- Robert Glazer: some of the visuals of seeing like what they built. Like I I I know your brain would create them in the in the book, but they did a really good job in the in the show of like showing you what it looks like. Anyway, this is not. a commercial for Silo, but highly recommend season three for those who haven't watched it. So you did your graduate work at MIT, you spent time around
- 3:24
- Robert Glazer: uh Marvin Minsky, one of the founding fathers of artificial intelligence back when AI mostly didn't work. So I'm curious, the people you were around, how accurate were some of those people about their predictions? I know people are always off. with timeline, but in the substance of what it was going to look like, and
- 3:44
- Robert Glazer: what did you absorb from them that you still use today?
- 3:47
- Unidentified speaker (speaker_0): So I was always the business guy in this stuff, right? So I was there was all the technical people, and I I got my MBA and my PhD in at uh Sloan at the business school. And so it was kind of interesting because I was in charge of interpreting these people and helping convert their views. so that other people could understood, so what the labs did could be understood. And the funny thing
- 4:07
- Unidentified speaker (speaker_0): is, like they built beautiful, elaborate theories of how you would build artificial intelligence. And it turned out worked was kind of brute force on a really fairly simple, I mean not that simple, but fairly simple idea that didn't all this complexity. It just required enough computing power and a big enough pile of human language, which I think was a surprise to everybody.
- 4:27
- Unidentified speaker (speaker_2): So was the concept of LLM like, was that
- 4:31
- Robert Glazer: in those days? Or what was the actual breakthrough that made all what was it the ability to just process those large language models that suddenly made we all woke up and it was Chat GPT? Because I I know people have been talking about it for a decade. And then like I think a lot of things, people like, hey, uh, and first of all, most people were calling machine learning artificial.
- 4:51
- Robert Glazer: intelligence for a long time, I think, intentionally.
- 4:54
- Unidentified speaker (UNKNOWN): Yeah,
- 4:54
- Unidentified speaker (speaker_0): I mean, or it was the form, right? Like AI kind of won, right? Like that was what would help your Waymo tell you where you were going. It would be every company was based around machine learning, right? Like, how could Amazon decide what pricing to give you, or how to route orders, what how could Netflix recommend movies? Like that form of machine learning is the and is also the basis for LLMs. What those things
- 5:14
- Unidentified speaker (speaker_0): do. is they're pr about prediction. They can predict something. You give them enough numbers and they can predict something. The breakthrough was the transformer architecture in 2017. The paper called The Tension is All You Need and the Attention Mechanism. And that basically let AI, rather than just do a sequence of numbers, be able to work with words. Because words are much less certain than The numbers, right? Like a number
- 5:34
- Unidentified speaker (speaker_0): sequence you could do math on. What do you do with a sentence that ends with the word file? filed? You don't know whether you're filing your taxes or filing your nails unless you zoom out. So we had to be able to pay attention to large blocks of text at a time and decide what to pay attention to to make predictions about what word comes next. That's the secret that unlocked LLMs. And that was a 2017 paper.
- 5:53
- Robert Glazer: So is the text processing that made an LLM possible? And then everyone started scanning all this content and then the horsepower was there to deliver that in a way that was fees financially feasible.
- 6:05
- Unidentified speaker (UNKNOWN): It
- 6:05
- Unidentified speaker (speaker_0): was a little bit of both. So it turned out that what you want to do is if you build a system that predicts the next word in a sentence or the next part of a word in a sentence. And you it happens to have this really weird quality, which is it works like the AIs we see today. And we couldn't have done this if it there already hadn't been piles of text Text on the internet. Because those early experiments used use those piles of text and basically did the same kind
- 6:25
- Unidentified speaker (speaker_0): of prediction approach we we've been using to data, but use it on words instead with a couple of these twists like uh transformers and attention. And that was enough to sort of unlock large language models.
- 6:36
- Robert Glazer: Interesting. So you alluded to this, but most of your career was spent more around entrepreneurship and gaming. I think in 2008 you co wrote Changing the Game about how video games were transforming business. So, what did games teach you about how people actually learn and help us on the gamer debate? There's half the group is worried that their sons
- 6:56
- Robert Glazer: are playing games all day and not getting out of the basement. And then there's a lot of data showing that actually games improve certain types of thinking and memory. And then actually, you know, with your ability to focus and do stuff. So there's been there's been a lot of pro-con on gaming. As a lifelong gamer, what does the science uh say
- 7:16
- Robert Glazer: here?
- 7:17
- Unidentified speaker (speaker_0): So, okay, so it's interesting. First of all, I've long done game stuff for education and transformation. Like that's what I've been doing before AI. That's how I kind of got into AI aside from the Media Lab work, was working on games. I mean, I would say that generally most games are just games, right? Like they're entertaining. They have all kinds of weird effects. Like, for example, whenever a new GTA
- 7:37
- Unidentified speaker (speaker_0): game drops, murder rates drop. because everyone stays inside to play the game and stops getting in fights. Um, so there's kind of weird effects that you wouldn't expect. But in general, we know that gaming can can kind of teach you skills, right? It just sort of depends. Like, people who are good at SimCity and Civilization turn out to be a little bit better as managers, so it's kind of a test of their ability.
- 7:56
- Unidentified speaker (speaker_0): Generally, games can be built for various things, but most of the games you play are kind of harmless. Like, there's no big evidence that they increase violence. There's no big evidence that they teach you that much. It's sort of like unless they're built to do that, they're sort of they're fun.
- 8:10
- Robert Glazer: Yeah, it's interesting. Someone said that they what a game does better is that it holds figures out how to hold someone attention and give them feedback. Right. And that a lot of problem with our learning these days is that you don't have the sort of feedback to then make a change. And if someone could unlock that, they could get all these boys.
- 8:27
- Robert Glazer: to do much better in a different realm.
- 8:31
- Unidentified speaker (UNKNOWN): Yeah.
- 8:31
- Unidentified speaker (speaker_0): I mean, so the secret to games is that they put you into what's called a flow state, right? Which I know a lot of people know is that productive state where time disappears. And flow is about a tension between boredom on one end, little too little action, it's boring, too much, and you get nervous. So games can adjust your flow state perfectly, right? If it's too hard, the I mean it gets a little easier. If it's too easy,
- 8:51
- Unidentified speaker (speaker_0): the challenge level goes up. So it keeps you in that flow state for a long time. It's very hard to achieve a flow state in learning because learning is supposed to be hard. Like this is one of the problems with AI in education is you actually have to force people to do work. So I've spent a long time, over a decade and a half, teaching people with games, and we can get pretty effective teachers with games, but you can't make you either have to
- 9:11
- Unidentified speaker (speaker_0): make it a little less fun than games because you need more, it's not gonna be as much fun. or a little bit less effective than other kinds of learning. But you still can get learning out of it. You can't quite get that same tension between difficulty.
- 9:23
- Robert Glazer: If you gamify learning, you'd have a different response. Well, your inflection point. So ChatGPT arrives in November 22. I was gonna ask you a question here, but I'll start with an anecdote. I think I remember I was chairing a conference and it was February 2023. So this has been out for like three or four months. And some of us are talking about at a table, and the
- 9:43
- Robert Glazer: guy, this guy's like, I've never tried Chat GPT before. You know, he's a CEO of a company and his 16 year old son is next to him. I'm like, what about you? And he's like, Well, well, I know. I'm like, and I I just knew by his reaction that like he was cheating. He did not want his dad to know that he was using this thing. So it was
- 10:03
- Robert Glazer: just, I can remember this so cleanly because he clearly was lying, like pretending that he's like, Oh, yeah, I think I heard about that. And I I think I cornered him later and asked him. But you moved almost faster than anyone in academia, becoming one of the first professors anywhere to require uh students to use AI in class. Walk me through that moment when
- 10:23
- Robert Glazer: you realized, A, it's not a toy, and B, like. I think that well, we'll talk about that later. But give me the first order reaction there.
- 10:32
- Unidentified speaker (UNKNOWN): Well,
- 10:33
- Unidentified speaker (speaker_0): first of all, it was actually interesting because it was the prior so ChatGPT came out at the end of November, and I was already teaching. And weirdly enough, I already had my students using a GPT-3 powered model. So ChatGPT was powered by GPT 3.5. GPT-3 had been on a year beforehand.
- 10:49
- Robert Glazer: They started, I forgot about it. So they started at three. Oh, so ones just never got into the public realm.
- 10:55
- Unidentified speaker (speaker_0): One is publicly downloadable. It's not very good. So there is a one and a two. One is, and actually, we could talk about some point, but I had a project where I converted one into books. So you can get 80 volumes, GPT-1. You can download them for free. From and in theory, then you could be uh do transformer math by hand. It would just take you about 30 years to do one equation. But as aside from that,
- 11:15
- Unidentified speaker (speaker_0): it's easy. But so GPT-1 came out, then GPT-2, and then GPT-3 was the first that kind of could write like a middle schooler, having my students experiment with that in class. And then GPT 3.5, chat GPT came out, and that's what sort of transformed everything. And so I was actually teaching at that time. So I was already warning people: I think this is gonna get good enough to
- 11:35
- Unidentified speaker (speaker_0): change how school works before it came out. And I remember the class afterwards, which was like a few days afterwards, I demoed GPT 3.5, ChatGPT to my class. I notic some people stopped paying attention. And by the end of the first class, I had a student came up to me and said, Oh yeah, I built the entire project that we were gonna build for the
- 11:55
- Unidentified speaker (speaker_0): entrepreneurship class during class while you were talking. And that was like a really interesting moment for me, like two days after ChatGPT came
- 12:01
- Unidentified speaker (UNKNOWN): out. Yeah.
- 12:02
- Robert Glazer: So you dove into it. I think most schools have either been oblivious to it or tried to outlaw it. Like, I don't know, education is we'll we'll get back to this, but it is it is not meeting the moment in terms of what is required for this. But I thought your BCG group experiment with
- 12:22
- Robert Glazer: 758 consultants is interesting. I know you're a big advocate of of sort of working. together. And it found that the ones using AI finished about 12% more tasks, 25% faster, 40% higher quality, with the biggest gains going to the lowest performers. And out of that came the sort of jagged frontier. What does
- 12:41
- Robert Glazer: jagged mean here?
- 12:43
- Unidentified speaker (UNKNOWN): Okay,
- 12:43
- Unidentified speaker (speaker_0): so a lot going on here. First, that study was one of the kinds of ones with my colleagues at Harvard MIT University of Warwick, and it was a great experiment. It's kind of obsolete now because GPD-4 is like you can run GPD4 class models on your phone at this point, but it was a big deal at the time, which was two years ago, just to give you a sense of how fast things move or three years ago. And so one of the things we
- 13:03
- Unidentified speaker (speaker_0): noticed about AI very quickly was that it was good at some stuff you wouldn't expect and bad at some stuff you wouldn't expect. And you couldn't predict in advance what it was good or bad at. So we called call it that the jagged frontier. Or Sort of imagine that there is a sort of line of ability, and we sort of know humans that anything outside the line is hard, anything inside the line is easy. The AI's line is very weirdly shaped. So things that would be hard
- 13:23
- Unidentified speaker (speaker_0): for a human is easy for the AI, things that would be easy for a human are hard for the AI. And you don't know what those are. So that's the jagged frontier of AI ability. It's only gotten more so over time, and it keeps changing as time goes on.
- 13:36
- Robert Glazer: Yeah. Almost every company I've been around ends up. with a website no one wants to touch. A small change sits in a queue for weeks, and by the time the page ships, the thing it was promoting is over. Framer is the pro website builder companies like Perplexity and Miro Trust. It's a full platform, not just a builder, with hosting, a CMS,
- 13:56
- Robert Glazer: enterprise grade security, and a 99.99% uptime SLA.
- 14:01
- Unidentified speaker (UNKNOWN): The
- 14:01
- Robert Glazer: new part is AI agents. working right on the canvas. They draft pages, manage your content, and tighten your SEO in the same place the real site lives.
- 14:10
- Unidentified speaker (UNKNOWN): The
- 14:10
- Robert Glazer: agents bring the speed, you keep the taste and have the final say. Thousands of businesses build there from early stage startups to Fortune 500 because change takes minutes instead of days. Learn how you can get more out of your site from a Framer Specialist or get started building. for free today at Framer.comslash elevate
- 14:30
- Robert Glazer: for 30% off of Framer Pro Annual Plan. That's Framer.com slash elevate for 30% off, Framer.com slash elevate. Rules and restrictions may apply. Starting a business is the exciting part. What stalls most first first-time time owners is the stuff that comes after. The registered agent, the business address, the state
- 14:50
- Robert Glazer: finance. filings, and the annual reports. Northwest Registered Agent has been helping people launch and grow businesses for nearly 30 years. They're the largest registered agent and LLC service in the U.S. with over 1,500 corporate guides, real people who know your local laws. You get a free account, and inside it or lawyer-drafted Drafted operating agreements,
- 15:10
- Robert Glazer: bylaws, resolutions, membership certificates, and bills of sale all
- 15:14
- Unidentified speaker (UNKNOWN): the time.
- 15:15
- Robert Glazer: in one place at no cost. They also help you stay in good standing with on-time annual filings. And when you want more later, you can add domains, email addresses, and phone numbers anytime you're ready. Don't pay hundreds or thousands of dollars for what you can get from Northwest Registered Agent for free. Visit Northwest Registered
- 15:34
- Robert Glazer: Agent.com Agent dot com slash elevate free and start using free resources. to build something amazing. Get more with Northwest Registered Agent at Northwest Registered Agent.comslash elevate free.
- 15:49
- Unidentified speaker (UNKNOWN): When
- 15:49
- Robert Glazer: I started Friday Ford in twenty fifteen, it was just a short weekly email to my team at Acceleration Partners. Eventually it needed a real home online and I kept putting it off because a professional website felt like a big expensive project.
- 16:01
- Unidentified speaker (UNKNOWN): I
- 16:02
- Robert Glazer: recently asked. Hostinger's AI to build a lifestyle ski website on a domain I had for several years. It put together an incredibly solid first version in a few minutes, better than what I had built years ago. No coding, designer, or developer needed. You describe what you want and the AI builds it in minutes. Help doesn't stop at launch. Hostinger's AI
- 16:22
- Robert Glazer: works like a built-in specialist for SEO, website copy, and email marketing. Your domain? website, email, and marketing all live in one account so you're not sketching together five different subscriptions. Snag Hostinger services at their lowest price before the sale ends. Create your website today for just one ninety nine.
- 16:40
- Unidentified speaker (UNKNOWN): Go
- 16:40
- Robert Glazer: to hostinger.com slash elevate ten and use code elevate ten for an additional ten 10% percent off Black Friday and Cyber Monday pricing. And how much of an advantage? do you are there always things that humans will be better at? Like is it or as we get to six and seven and eight? I mean, I've heard two different things, but I'm you're
- 17:00
- Robert Glazer: more better positioned to answer them. One is, you know, now we're gonna have self-improvement and all this crazy stuff, and it's faster and we're gonna lose control. And seems like safety people are resigning every day, which a little concerning. But two is that like we're actually the gains now are are more, more incremental. And that I mean, there's so much power and processing needed that I'm not sure
- 17:20
- Robert Glazer: we can double and triple down on these investments right now. So are there things that AI will always struggle? Because some days I know myself and I'm using it constantly. I'm like, oh my God, this is amazing. And others, I'm like, you stupid. Like, why can't you understand what I'm what I'm asking you?
- 17:38
- Unidentified speaker (UNKNOWN): So
- 17:38
- Unidentified speaker (speaker_0): the there's a lot going on that question. There's like a lot to unpack, right? Let's put safety aside for right now as an issue. So right now, it appears that AI development continues to follow an exponential curve, but still a jagged one, right? So it is getting better at most tests it's already good at. And some tests it's bad at, it's getting good at. One of the things I
- 17:58
- Unidentified speaker (speaker_0): start my new book with is a whole description about how the best forecasts forecast of the world, right before ChatGPD came out, gave a 2% chance that an AI would solve, uh, would get a gold medal in the International Math Olympia, a very hard math test by 2025, right? And of course, OpenAI and Anthropic did. And a few minutes, a few uh the day before we recorded this, right? I don't know what's going on, but the day
- 18:18
- Unidentified speaker (speaker_0): we recorded it before we recorded this, OpenAI solved one of the hardest outstanding problems in math. Right? So we've gone from no math could like AI couldn't do math to AI. is doing superhuman math. And I think we're gonna keep seeing this jaggedness play out, right? Like, where's a weak spot of AI right now? It's a very bad fiction writer. We started by talking about science fiction. Like AI writes terrible science fiction. It's very bad
- 18:38
- Unidentified speaker (speaker_0): at plots, it has trouble making characters believable every kind of speaks in the same way. And it's an interesting gap. So the bet from the AI labs is that we will get superhuman AI that can beat humans at everything. So far, we're still seeing still very jagged AI, if you know, brilliantly jagged.
- 18:55
- Unidentified speaker (UNKNOWN): So
- 18:55
- Robert Glazer: there's a cat and mouse game in the sort of AI detection world. I've warned leaders to be careful because as detection gets better, people are going to go back to their old stuff and it may hurt some credibility. Like whether it's a big difference whether it's marketing copy or something sincere that you wrote that shows is 100% AI. But it seems like all of those things are based. on pattern
- 19:15
- Robert Glazer: analysis. And so you when you say that, hey, be random enough that because I've tried to play around with it and test it and understand it. Like be random enough that you don't trigger AI detection. Like it can't, like in some ways. Like you have to think it. It's not able to do it. I would imagine it would be. But is that sort of one of these things that's just hard to like
- 19:35
- Robert Glazer: actually mimic a human behavior?
- 19:38
- Unidentified speaker (UNKNOWN): Well it's
- 19:38
- Unidentified speaker (speaker_0): not so much. it's hard to mimic human behavior as you could sort of think like working with AI, like working with one really brilliant person, right? So I've done a lot of work on AI and innovation, experimentation.
- 19:47
- Robert Glazer: High EQ, low EQ, I say.
- 19:49
- Unidentified speaker (speaker_0): It's actually pretty high EQ at this point. Like it has really good theory of mind, it's good at anticipating emotions. People would rather talk to it than AI. There's a lot of work showing that it, you know, it gives the same answers as an ethicist would to a problem, or actually other ethics SS prefer the AI's answer. It's as a psychologist, it does quite well. So I think people that's a quite right, right? I think it's high IQ and high EQ, but kind of one person
- 20:09
- Unidentified speaker (speaker_0): who does that. So it has like its own obsessions, its own writing style. It might be a very good writing style, but like you can smell the stink of Claude a ways away.
- 20:16
- Robert Glazer: It's singular, yeah.
- 20:18
- Unidentified speaker (speaker_0): Yeah. And so there's things you could do with prompting to make that a little bit bigger, right? But it's kind of like one really creative person. And that is one of the barriers. to sort of using it and why why detection and things like that still work.
- 20:31
- Robert Glazer: All right. So you wrote coexistence. I mean, you wrote cointelligence when AI still struggled with basic math, as you said. Now with coexistence, it's very different. It's winning the uh International Math Olympiad. What does coexistence require of us that cointelligence did not?
- 20:49
- Unidentified speaker (speaker_0): So with coexistence. part of this is a mode of using AI, right? So we sort of talked about the first phase of AI early on, which was machine learning, making predictions based on numbers. I would call the second phase of the AI grandiosely after my book title, the cointelligence era, where you'd use a chat bot, right? Like, so you know, here's my email, help me do this. Let's go back and forth on
- 21:08
- Unidentified speaker (speaker_0): something. As of the beginning of 2026, we entered what would really be called the agentic era, but I'm gonna call it coexistence. because I've got a book. But it's really about working with AI systems that are both sometimes smarter than you, but also could do 10, 20, 30 hours of work on their own. I mean, I have an AI that system I used, I used Astra and was a GPT six, and it
- 21:28
- Unidentified speaker (speaker_0): worked for six days on a really complicated project and resolved it for me.
- 21:32
- Robert Glazer: Right. This is like trying to run faster than a car. Like there's a reason a horsepower, like it's just there's some things where it it's just a stupid thing to try, right?
- 21:42
- Unidentified speaker (speaker_0): But it's also a thing where now the work is about you basically have an infinite number of contractors working for you. Can you assign them the work properly? So Coexist was all about using yourself to do the work with AI help. This is all now about assigning work to AI that it can sometimes do even better than you could.
- 21:58
- Robert Glazer: I think that distinction is interesting because it puts the people. at the top of the pyramid in a good place. I agree. I like it's a bunch of workers. But if you give them a bad assignment or you don't delegate well or you don't explain, just like a regular worker, you don't like the input. But the difference is they'll work all night and complain about wanting wanting healthcare. But the
- 22:18
- Robert Glazer: one of the book's four questions is how to tell real capability from an exaggerated claim. Debate right now, there's obviously some amazing things going. on in productivity. I think there's actually a lot of misdirection by public companies sort of saying that layoffs are AI driven when they were just overcapacity coming out of COVID. But we're
- 22:38
- Robert Glazer: hearing different things. I mean, right now companies have poured billions of dollars and plenty of pilots have showed no measurable return. Some economists are saying the productivity gains barely register. So what is does the evidence? actually show in the real world around where it's working and where it's not living up to some of the height?
- 22:57
- Unidentified speaker (UNKNOWN): So
- 22:57
- Unidentified speaker (speaker_0): I think it's worth kind of updating. A lot of the some of the stuff you're talking about was really from 2025 and things changed quite quickly. So my colleagues at Wharton run a very large survey of big companies. 85% say they have positive ROI from AI as of November 2025. The sort of a pilot study that most pilots fail was not a particularly well-executed study, but even so,
- 23:17
- Unidentified speaker (speaker_0): like, it's not to say a bad idea that pilots fail, right? Pilots should fail, you're not experimenting enough. But I think it is fair to say through 2025 at least, there was no major ek mac macroeconomic impact. And part of the reason is like, let's say you let's let's just ignore fights over productivity and say one of the there's a paper that shows coders produce 70 dimes more code. Let's say you're producing
- 23:37
- Unidentified speaker (speaker_0): 70 dimes more code than you. did before. You're in a company that has that's built around two-week coding sprints. So every two so you get an assignment at the beginning of the week, everyone coordinates on what that is, and at the end of the two weeks, you produce like what difference does it make if you're producing 70 times more code if you still have to go to all the meetings, nothing else is changing your process. So you can't just change individual
- 23:57
- Unidentified speaker (speaker_0): productivity and expect corporate productivity to change, right? So people are being very individually productive. but now you have to change how your company operates. I can now generate market research reports on demand. How many companies have stopped buying market research reports? Is that because they're still useful to them, or is it because they haven't even figured out what the purpose of the original play was?
- 24:15
- Robert Glazer: But if you don't use market research in general, then it's not going to matter. I think, and there's a lot of problems just because you can doesn't mean you should. I think the best example I saw is look, the cost of building an app. has not gone to zero, but it's pretty low. There's something like 10 times more apps in the store, whatever is a huge number chart, and zero more
- 24:35
- Robert Glazer: engagement on the average number of apps per day. Right. So that sort of puts the problem in, yeah, you can build an app, but it's still hard to market it and get people to use it and figure out what to build. The
- 24:47
- Unidentified speaker (speaker_0): Jagged Frontier again, right? Like the the the as long as the AI can can't like so that's also. why you see the job impact change happen in a particular way, right? So there's this theory of jobs called the O-ring theory, grimly named after the challenger or disaster, where 16,000 parts of the challenger worked well, one thing failed the O-ring and the challenger blew up, right? Jobs are like that. It's not just additive.
- 25:07
- Unidentified speaker (speaker_0): There are things that have to get done. And if the AI I can't do that thing and you need a person to do it, you can have all the AI in the world, but you still need a person to do that work. And the value of doing that work increases, not decreases, because the AI does everything else cheaply. So work is full of this kind of stuff, right? Probably in preparing this podcast, there were maybe 80% of stuff you could have delegated to AI that would do a better
- 25:26
- Unidentified speaker (speaker_0): job, and 10% where it might have been worth doing. And then the other 10% were probably outside the Jagger Frontier, where if you gave AI the task, it would screw it up badly.
- 25:36
- Robert Glazer: So perfect collect. Perfect. segue. And I'll be honest. So how I prepare for these podcasts, what people used to do is they would Google tons of stuff and find articles and whatever. Now I have it prepare me a massive dossier. Right. And and so I go look in it and then I decide what's interesting and what I want to cover and want to earn more questions. That 20
- 25:57
- Robert Glazer: if I didn't do that 20%, it would sound horrible and robotic and whatever, but that 80% collection is on a scale that I could have never done before, right? I because I could have picked a one podcast to listen to or read some of the book or what I and so it is a great example. And and it's an iterative process, by the way. Every time it brings me back the research, I give a
- 26:17
- Robert Glazer: feedback around what would have been more helpful or not helpful to get. But that 20% matters a lot. And I think a lot of companies and people who are skipping that, I think they're putting their reputation, you know, the AI slop problem in organizations now. My favorite meme, I don't know if you saw this, but it was a guy who had these four bullet points, and he said, turn these into a
- 26:36
- Robert Glazer: 10-page beef for me, and he sent it out to his team. And the other guy was like, turn this into the four bullet points. I need to understand.
- 26:45
- Unidentified speaker (speaker_0): It's true. Like the the slop is not an AI problem, it's a human problem in a lot of ways. right? Like it's a decision about how to do it. Like, you know, AI detectors aside, there's a bunch of evidence that anyone who reads AI a lot knows what AI writing is. So, like when I read something, a briefing document someone gave
- 27:00
- Unidentified speaker (UNKNOWN): me
- 27:00
- Unidentified speaker (speaker_0): that's written by AI, it's not even that, you know, I don't mind them having used AI to prepare it. I don't even mind using AI to polish it, but I'm like, I just want to make sure there's a human work here that matters. And I think that that will matter to people. Now, at the same time, AI is becoming more capable. I am pretty convinced. with the tools we have now that we can make a perfectly good 90th percentile podcast
- 27:20
- Unidentified speaker (speaker_0): without the two of us doing any work, without with with um fake videos of us doing this, with voices done through 11 labs, with the research being done, it would be good. Would it be top 1% or 0.1% in terms of downloads? Would it be that good? No. But like for most people, right, maybe it's in the area that they care about much more and it's a customized podcast for them. So
- 27:40
- Unidentified speaker (speaker_0): I think that it is interesting. Like the human craft matters a lot and differentiates at the top level, but maybe not en masse.
- 27:47
- Robert Glazer: But you were making a utility argument. I would argue that I have an analogy. We've had G-Clays or art reproductions that no anyone other than an art specialist could not tell you the difference between an original and a and a recreation. And people pay orders of magnitude difference for that. So I think the people that
- 28:07
- Robert Glazer: become overly dependent on, I think there's going to be a huge resurgence of people looking for human writers, particularly in the AI detection world. And when a lot of people who never learned to write because they were using AI, I think there'll be some interesting bounce back, not because the podcast is better, but but there's a reason why people value the original piece. of art more than the
- 28:27
- Robert Glazer: reproduction.
- 28:28
- Unidentified speaker (speaker_0): I absolutely agree with you. And I think we'll see a retreat to human. We're gonna see both tensions happen at once. So it'll be a retreat to what's human.
- 28:34
- Unidentified speaker (UNKNOWN): It
- 28:34
- Robert Glazer: is both, yeah.
- 28:35
- Unidentified speaker (speaker_0): But I also think people sometimes over like both underestimate and overestimate that, right? But also at the same time, people like AI stuff. They like talking to AI companions. Our surveys, when people in our experiments when people work with AI, they're as happy as when they work with a teammate or happier. Like, like people tell you they're upset, but everyone secretly uses these systems. So like I think we're both gonna retreat to real human things and
- 28:55
- Unidentified speaker (speaker_0): human touches and human imperfection. But at the same time, I think it's also a lot to be like, uh just because I read bad slop, the thing is you're reading you recognize the slop. You don't recognize the AI assisted stuff that isn't slop.
- 29:07
- Robert Glazer: Yeah, and something that sounds super fancy but no one really put thought or or care into it. I actually had uh a thing where went to through something I had written and make sure there was no AI and spent sort of a day looking at how Pangram analyzes things. Because it's it's interesting. One thing can turn an
- 29:27
- Robert Glazer: entire page bad, which proves that there's some problems with these uh models. But after a day of doing it, I can actually read the pattern so clearly now. I can see what it is, I can look at it. The you know, it's not this, it's this, the negative. I I don't remember the names for all of them, but I just I can spot it so much more uh clearly now.
- 29:47
- Robert Glazer: What
- 29:47
- Unidentified speaker (speaker_0): you're saying is really carrying the weight of this argument
- 29:49
- Unidentified speaker (UNKNOWN): today,
- 29:50
- Unidentified speaker (speaker_0): I would say. And we there's actually a paper showing this that like even Pangram is an effective AI detector, at least for now, right? And there's a good chance, ch like, but it's an effective AI detector. But there was a bunch of studies showing other AI detectors weren't. effective. And but showing that humans are actually very good at detecting AI, right? And for exactly that reason. You spend a day or two with this and you're like, I mean, try
- 30:10
- Unidentified speaker (speaker_0): and read LinkedIn after working with AI for a while and it will make your eyes bleed, because it's just like a hundred percent.
- 30:16
- Robert Glazer: LinkedIn is the perfect example of like they just encouraged AI slop for a year. They had these AI AI-driven driven questions that you could answer with AI and you would get a badge for. it. And I was like, this is a whole bunch of garbage. Like that was a terrible business decision, basically. And now, of course, their algorithm is now trying to
- 30:35
- Robert Glazer: reward non-AI stuff. I
- 30:38
- Unidentified speaker (speaker_0): mean, the comment section of everything is a disaster, right? Like I'm pretty big on social media. Really interested to read LinkedIn comments. Now it's absolutely not worth it because even when the humans are doing it now, it's all AI mediated. And you know, when you get these really like it's very funny. because at least they all get my obscure jokes now because the LLMs are really well read. But aside from that, it's not that great.
- 30:57
- Robert Glazer: I do a bunch of different leadership training and coaching. And one thing I've been saying to people recently in leaders, and I kind of called one out on it and they went back. Because detection is getting better. Look, a lot of people who completely cheated and did stuff in the first year, they're kind of exposed now because it's like you can go back. and detect this stuff, right? And I said, look,
- 31:17
- Robert Glazer: if you're writing a marketing thing or a functional thing or a lot of emails, I don't mind using it to speak to the thing and write the email because the goal wasn't to write a perfect piece of information, it was to communicate something. And I'm undoing all of the work that technology has made that didn't exist for 20 years. But if you were writing a heartfelt
- 31:37
- Robert Glazer: company thing, a strategy plan, a layoff note, and you do that in AI, and someone on your team go puts that in Pangram and shares on some internal Slack, the 100% AI generated, I think your credibility could be permanently shot. And so I've really highlighted this for a lot of people that they there are some communications where
- 31:57
- Robert Glazer: even if they're working with AI, those things don't distinguish between it was entirely written by AI and it was edited by AI. AI. I think some leaders are really gonna get dinged for this.
- 32:08
- Unidentified speaker (speaker_0): There's a whole bunch of like social things happening there, right? Like I remember, you know, one of the things, like for example, I talk about this in the book, but like one of the things that you do as a professor is you write letters of recommendation for people. The letter of recommendation isn't that important. The fact is that I'm setting my time on fire for a student is what matters. It's a signal of my effort, right? And but now I can get right
- 32:28
- Unidentified speaker (speaker_0): a much better letter of recommendation in a minute, right? And so my students rather I write the better letter of recommendation than this longer one, but it's no longer a signal of my time and effort. In fact, I had a student, and they allowed me to share this, actually send me the prompt they wanted me to use for the letter of recommendation. So they're like, just paste this in as your letter. And I thought that was very funny. But there's a whole bunch of things like this where
- 32:48
- Unidentified speaker (speaker_0): time and effort were indicated by work. And it's funny with CEOs, because a lot of them had speech writers, people doing this kind of thing for them, and it was fake effort, right? But it was like it felt that way. Oh, you know, I talk in the book about how obituaries are increasingly and eulogies are increasingly written by AI. And like you have to kind of have two minds about this. On one hand, oh this is your deeply
- 33:07
- Unidentified speaker (speaker_0): post personal thing. On the other, you're overcome by grief. You're not a good speaker. How do you, you know, you want to sum up a life? Like the tensions are there.
- 33:16
- Robert Glazer: Well to me, if you I use a lot of text to speech. And if you sit down and this is basically what the scribes did in the old days. You know, the the great orders spoke and then other people wrote down their thoughts. I think some of the greatest kind of writers in history, they didn't write. They talked. And so I think if if you are talking for three minutes and giving a speech
- 33:36
- Robert Glazer: and then saying you know, organize this, that's different than write the eulogy for me, you know, and make up something. I know it will show up.
- 33:44
- Unidentified speaker (speaker_0): Absolutely. But but it will show up with, but as you said, it will show up at AI detectors either way, right? So, but that's what I mean is we have to negotiate what this all means, right? Where like AI and human language is going to get mixed together. It already is. Like there are phrases I started to say in the real world, where like if I it was just writing something where I said it's not this, it's that, or like, I have to delete this because people are gonna think it's AI.
- 34:04
- Robert Glazer: Right. And it used to be good editing, and AI just overused it. So from editing with AI, I have more proclivity towards some of these things that are AI tells that I do manually now. Like it hates the triplicate, right? Like, and if you do two triplicates in a row, yeah. Triplicates are profound. found. I And
- 34:22
- Unidentified speaker (speaker_0): love that. I love triplicates. I I loved M-dashes. Yeah,
- 34:25
- Robert Glazer: they're M dashes. M-dashes I never use, so that's fine. But actually when you start seeing the triplicate more, you realize it just sounds really good.
- 34:33
- Unidentified speaker (UNKNOWN): It
- 34:33
- Unidentified speaker (speaker_0): sounds really good. And then also the very sentence like where it would do a whole paragraph and then there's like a kicker sentence. Like that's I mean that is great rhetoric is like the kicker sentence. And it looks cheesy now, right? And there's a bunch of phrases, Is, you know, I talk in the book about meaning shaped attention vampires as an example of LinkedIn comments that force you to pay attention. Unfortunately, now Claude uses something
- 34:53
- Unidentified speaker (speaker_0): shaped all the time. So even though I wrote this before the models were doing it, I'm like, everyone's gonna think I wrote that with AI. And there is parts of the book, right? Like I do use AI for editing and stuff, but like it's a challenge right now to be a writer in this world.
- 35:05
- Robert Glazer: Yeah. And if you edit, I just think you can't have the exposure of someone saying this book is eighty 80%, percent, you know, uh like I I some guy just lost a two million dollar book deal. And I don't think you can have a heartfelt layoff note to your company that triggers an AI. Even if you didn't, even if you wrote it, the optics could be bad.
- 35:25
- Robert Glazer: But a lot of CEOs have told me some version of this story. My organization is has three groups right now. Like I love AI and I'm all in on it, and I'm sort of bleeding edge. I'm really into it, but like I actually need some training and examples, and then the kind of level. Luddite group of I don't like this. I want to go back to the and depending on what industry you're in,
- 35:45
- Robert Glazer: I think most people have sort of just uh realized that the Luddite group is gonna have to move on because it's just they're not understanding that what their clients are gonna want from them and the demand, like is this kind of like I want to have a horse and buggy repair shop, and you're like, Yeah, but everyone's got cars now. Like, so this is we're not gonna. have a great, great business. But the
- 36:05
- Robert Glazer: leverage component, you've looked a lot at. So you had another experiment with Procter and Gamble, 776 professionals. One person working with AI matched a two person team working without it. And now a quarter of OpenAI's own staff have four more agents that are running every week. So if one person plus AI equals a team, like what
- 36:25
- Robert Glazer: happens to how we organize companies? And are they smaller or are they a thousand-person company doing the work of a 4,000-person company?
- 36:36
- Unidentified speaker (speaker_0): I
- 36:36
- Robert Glazer: hope you have the answers then.
- 36:38
- Unidentified speaker (speaker_0): I'll give you what I can. I'm a physical professor, right? And it's funny, like the org chart was invented in 1855, right? It was invented for the to manage the railroads. And we keep the same chart today because the core element is people. How many people can you manage? How many people can you, you know, that hasn't changed in people are people. Now we have another form of intelligence,
- 36:58
- Unidentified speaker (speaker_0): and suddenly we have to manage a whole different way. And nobody knows the answer, right? So one thing I want people to feel comfortable with who are listening to this is uh, you know, I'm as close to this problem as anyone. I don't take money from any AA lab, but I I talk to them all, I talk to CEOs much like you do. Everyone's just trying to improvise. Like there is not like a rule book. here. And a lot of it's a really important time for CEOs and
- 37:18
- Unidentified speaker (speaker_0): leaders to think about how do I use AI to enhance what my people do, to make their work more meaningful, to become, say, a 4,000 person company, but become a 4,000 company that acts like a 100,000 person company rather than a 4,000 person company that shrinks to 200. And I think, by the way, you're giving a part of the key about why you don't want to shrink to 200 in this discussion about ClaudeSpeak Claude speak and things like that. Like you don't
- 37:38
- Unidentified speaker (speaker_0): want Claude running your company because Cloud's gonna run every company and it will either drive your customers crazy or it's gonna make the same decisions everyone else makes, getting rid of your remote. So I think the companies that figure out how to thrive by growing are in much better shape than companies that try and thrive by retreating in the face of AI and shrinking.
- 37:56
- Robert Glazer: I have a friend who we had a conversation about Claude and co-work. He's like, I have to go do this. He went back, started doing COD code for the first time, didn't sleep for two weeks, realized he could code this entire technology system for his business. And he went and hired for the first time in five years and added new roles because he was like, We're a 10-person company and I can
- 38:15
- Robert Glazer: compete like a hundred-person company. And so there's a lot of people on the job stuff. I don't think there the problem with bold claims is that no one ever goes back and looks at people's bold claims. and how often they were wrong. One of the moderate views I read on it said, it used to be that 40% of people worked in the food industry in the US. We don't
- 38:35
- Robert Glazer: eat less food now, but only four percent of the people work in the food industry and the jobs displaced and went elsewhere. I I tend to believe we're looking at some version of that.
- 38:45
- Unidentified speaker (speaker_0): 20% Twenty percent of Americans worked uh in or around the railroads in the early turn of the century, right? Like I
- 38:50
- Unidentified speaker (UNKNOWN): mean,
- 38:50
- Unidentified speaker (speaker_0): jobs. change. I think we don't know, right? I think you're exactly right. We don't know. The early effects are very confusing, right? There's some papers that find no effect. Increasingly, we're finding more demand for jobs, but we don't know whether we can tribute how much we attribute to AI. There's some warning signs that junior jobs might be more threatened because you have to train them.
- 39:08
- Robert Glazer: Early stage, yeah. Those seem to be the most, yeah.
- 39:11
- Unidentified speaker (speaker_0): But we don't know. We just don't know. So, like you could imagine worlds with all of these kind of outcomes. I also think people are overestimating the effect of AI right now and underestimating the long term. It's Amara's law, it's called for futurists, which is you overestimate short-term effects, underestimate long-term. Companies have not changed much. Like you go to a company 2025, like you think about your own company or organization. Has it
- 39:31
- Unidentified speaker (speaker_0): radically transformed as a result of AI? Even if AI could radically transform, it, which I believe it can, the answer is no. People are still people. Organizations take time to change. but there will be change. And I just don't know which way it shapes out.
- 39:43
- Unidentified speaker (speaker_2): Well, you
- 39:44
- Robert Glazer: almost every company I've been around ends up with a website no one wants to touch. A small change sits in a queue for weeks, and by the time the page ships, the thing it was promoting is over. Framer is the pro website builder companies like Perplexity and Miro Trust. It's a full platform, not just a builder, with hosting, a CMS, enterprise
- 40:03
- Robert Glazer: grade security, and a 99.99% uptime SLA.
- 40:08
- Unidentified speaker (UNKNOWN): The
- 40:08
- Robert Glazer: new part is AI agents working right on the canvas. They draft pages, manage your content, and tighten your SEO in the same place the real site lives.
- 40:17
- Unidentified speaker (UNKNOWN): The
- 40:18
- Robert Glazer: agents bring the speed, you keep the taste, and have the final say. Thousands of businesses build there from early stage startups to Fortune 500 five hundred because change takes minutes instead of days. Learn how you can get more out of your site from a Framer Specialist or get started building for free today at Framer.com slash elevate for
- 40:38
- Robert Glazer: thirty percent off a Framer Pro Annual Plan. That's Framer.com slash elevate for thirty percent off Framer Framer.com dot com slash elevate. Rules and restrictions may apply. Starting a business is the exciting part. What stalls most first time owners is the stuff that comes after. The registered agent, the business address, the state
- 40:58
- Robert Glazer: filings, and the annual reports. Northwest Registered Agent has been helping people launch and grow businesses for nearly 30 years. They're the largest registered agent and LLC service in the US with over 1,500 corporate guides, real people who know your local laws. You get a free account and inside it or lawyer drafted operating agreements,
- 41:17
- Robert Glazer: bylaws, resolutions, membership certificates, and bills of sale all in one place at no cost. They also help you stay in good standing with on-time annual filings. And when you want more later, you can add domains, email addresses, and phone numbers anytime you're ready. Don't pay hundreds or thousands of dollars for what you can get from
- 41:37
- Robert Glazer: Northwest Registered Information. Agent for free. Visit Northwest Registered Agent dot com slash elevate free and start using free resources to build something amazing. Get more with Northwest Registered Agent at Northwest Registered Agent dot com slash elevate free.
- 41:56
- Unidentified speaker (UNKNOWN): When I
- 41:56
- Robert Glazer: started Friday Ford in twenty 2015. fifteen, it was just a short weekly email to my team at Acceleration Partners. Eventually,
- 42:02
- Unidentified speaker (UNKNOWN): I'm going to be able to
- 42:02
- Robert Glazer: it needed a real home online, and I kept putting it off because a professional website felt like a big expensive project.
- 42:20
- Robert Glazer: No coding, designer, or developer needed. You describe what you want, and the AI builds it in minutes. Help doesn't stop at launch. Hostinger's AI works like a built-in specialist for SEO, website copy, and email marketing. Your domain, website, email, and marketing all live in one account so you're not sketching together five different subscriptions.
- 42:40
- Robert Glazer: Snag Hostinger services at their lowest price before the sale ends. Create your website today for just one ninety nine.
- 42:47
- Unidentified speaker (UNKNOWN): Go
- 42:47
- Robert Glazer: to hostinger.com slash elevate ten and use code.com. code elevate ten for an additional ten percent off Black Friday and Cyber Monday pricing.
- 42:55
- Unidentified speaker (speaker_2): Wrote a
- 42:56
- Robert Glazer: that every company has uh some cyborgs, secret cyborgs. So people who quietly automated big chunks of their job and don't tell anyone because the AI policy sort of scares them into silence. There was an article I remember a Verizon employee years ago had completely outsourced his job for years. He had figured out offshore and all this stuff and they
- 43:15
- Robert Glazer: found out and they fired him. And I remember thinking similarly like, dude, bring that guy in. And be like, what did he do? So this would seem like companies that don't get it if the employee is not willing to step up and say, look what I did. And how can I help you? How can I be more valuable by replicating this in other parts of the organization?
- 43:35
- Unidentified speaker (speaker_0): Yeah. I mean, I I think that that the problem is an incentive. problem, right? It's an old school incentive problem, which is think about if
- 43:42
- Robert Glazer: Show show me, you know, you know, in Amonger, I think I've said this in Lisa last three podcasts I did. Uh show me the incentive and I'll tell you the behavior.
- 43:50
- Unidentified speaker (speaker_0): Absolutely, right? And think about the incentives, right? So first of all, a lot of companies ban AI use. So you're not going to show you people using AI. Second, we
- 43:56
- Unidentified speaker (UNKNOWN): know
- 43:56
- Unidentified speaker (speaker_0): there's social desirability Viability bias. You look like a genius right now. People think it's AI, they'll think you're less competent or you leave believe that. You don't show you you're using AI. Think about what we
- 44:03
- Unidentified speaker (UNKNOWN): just
- 44:03
- Unidentified speaker (speaker_0): say: productivity gains. Everybody knows what productivity gains means. It doesn't mean the worker gets more money. That means workers get fired. Why would I show you I'm using AI? And then another reason is exactly what you said. One of the main re things people do once they start using AI is stop working as much. Why would you ever tell your employer I'm working less? Like who cares, right? Like that's a they're not gonna reward Going to you for that. Or even if you wanna want to help, who do you go to?
- 44:23
- Unidentified speaker (speaker_0): Like who in the company do you say I've replaced my work with AI? So, because of all that, incentives are way off. And this is where leadership has to make a play a role. Like, leaders and companies have to figure out how to align incentives with their employees.
- 44:35
- Robert Glazer: They need like an X prize in the company. Like, come to us with a thing and we will reward you financially for figuring out how to automate a job.
- 44:44
- Unidentified speaker (UNKNOWN): Yeah,
- 44:45
- Unidentified speaker (speaker_0): there's an old Robert Heinlein story that everyone that he uh that uh in science fiction they they motivate every everybody by making it so that if they replace themselves at a job, they get paid for that job for life. So there's people who just what they do is they just try and find the job.
- 45:01
- Robert Glazer: That's a great incentive for at least five years. Yeah.
- 45:03
- Unidentified speaker (speaker_0): Yeah. But I've seen companies give $10,000 cash prizes out at the end of every week, whoever's automating the most work. But like they have to believe you, right? All the companies that have bad cultures where like everyone's faking that, like, oh yeah, we're a great place. If you don't believe the CEO is actually gonna try and protect you, if you don't believe that the company's actually trying to grow, if you don't believe you're gonna be rewarded and not punished, you're not gonna show
- 45:23
- Unidentified speaker (speaker_0): AI youth. Like culture really matters now.
- 45:26
- Robert Glazer: Yeah, I actually someone who helps us with financial planning. I was on with one of her associates, and they just the financial world is heavily restricted. And they this person collected one of the clients, sent in all these receipts. And he would spend hours going online and Googling, categorize them. I was like, look at what you can do with Copilot. And in two seconds, the sheet
- 45:46
- Robert Glazer: filled itself out, researched all the things, put icons next to him, and his face just completely like he looked like a gas. Because you could see, like, this is too. And I said, you have two choices now. You can go tell your boss, like that you can do this, and you guys can have a whole service line and change what you're doing, or you can just do this. and not try it. And he did tell her, and she told me the story, and it was it was very funny. But
- 46:06
- Robert Glazer: yes, people face that choice. I want to come back to education because you obviously you're in this space, you're a professor. Education does not know what to do with AI. One of my favorite articles of the year was by the Babylon B, the satire newslet Newsletter paper. I don't know if you read this, but they ran a piece with Chat ChatGPT GPT announced as Harvard Valley
- 46:26
- Robert Glazer: Victorian. because of all the papers it had written, and it was the largest contributor to the class and all the homework. And it listed all of its accomplishments right around the New York Times ran a magazine. And I had two college age kids at the time that an article, Everyone is Cheating Their Way Through College. So is cheating the problem here, or is it that the
- 46:46
- Robert Glazer: assignment stopped making sense or that teachers didn't engage with this or figure
- 46:50
- Unidentified speaker (UNKNOWN): out?
- 46:51
- Robert Glazer: how to evolve and figure out if their students are learning? And interestingly, a lot of the cures that I see for this are actually like back to the future things. We need blue books. We need in-class discussion of reading comprehension. I actually find that super interesting because supposedly we made education better, but I
- 47:11
- Robert Glazer: everyone's starting to say that technology made it worse.
- 47:14
- Unidentified speaker (UNKNOWN): Oh,
- 47:14
- Unidentified speaker (speaker_0): there's a lot going on there, right? So, first of all, everything is a problem that you mentioned, right? Cheating was already a problem. There's a great study, I think, out of Rutgers, that found that in 2008, I'll be slightly off of the numbers, something like 80% of people who did their homework got higher test scores. And that number dropped to 20% by two 2020 before Chat GPT, because everyone was just copying online, right? So it's not like cheating is
- 47:34
- Unidentified speaker (speaker_0): new.
- 47:35
- Unidentified speaker (UNKNOWN): Well,
- 47:35
- Robert Glazer: plagiarism was the big online thing before that, like before there were any. plagiarism detectors. You could buy papers.
- 47:41
- Unidentified speaker (speaker_0): Yeah. There were 40,000 people in Kenya whose full time job was was writing essays for people.
- 47:46
- Robert Glazer: Yeah. I just saw that. What newspaper they wrote a job that it's affecting Kenyan unemployment? It was an interesting angle. Yeah. But
- 47:54
- Unidentified speaker (speaker_0): but it was something I think, you know, there's actually was a well known discussion in sort of cheating circles before this about the existence of those paper mills. I think the real issue is that it's the most cheesy thing I could say, right? Which is you're only hurting yourself is true. Like to go back to what we were saying before, the problem with education is it requires friction. Like you don't learn without difficulties, desirable
- 48:14
- Unidentified speaker (speaker_0): difficulties, they're called. And so unless you are unless work is hard, you're not learning, right? And hard work is hard and nobody wants to do it, no one has time for it, so they cheat. So AI makes cheating much easier and invalidates. invalidate, and because it's so hard to catch, it invalidates large forms of kind of work that we used to do before. And that sucks because some of these forms of writing
- 48:34
- Unidentified speaker (speaker_0): were like essay writing was a good idea. Take-home tests were a good idea, right? Cheating hurts them and it hurt all of us, but that there we are. I do think that, you know, it's funny that one of the most, the strongest evidence we have for a pedagogical technique is probably small repeated tests. It turns out small low-techs testing is incredibly helpful for
- 48:54
- Unidentified speaker (speaker_0): learning, not just for assessing learning, but for learning. So like in-class discussions, always effective. Like, so some of this back to the future stuff is this is what actually has worked. So I can put you in a classroom setting, I can give you a little test, we can have a discussion, we can do activities. Now, the flip side of that is we actually know that if you use AI to learn by just asking questions or asking to do your homework, you destroy your learning, right? The
- 49:14
- Unidentified speaker (speaker_0): AI helps you. and you don't learn anything. If you use AI as a tutor where it doesn't give you the answer, it pushes you for the answer, you learn a lot. So what's going to happen in the long run is AI tutors outside of class, just like people watch Khan Academy videos now or whatever, inside of class activities and active learning and assessment. And we'll get there. It's just education doesn't change quickly.
- 49:33
- Robert Glazer: If you're a professor and you're not addressing this now, like are you not doing your job? Like if you're if you're either not declaring, like, hey, if you cheat, you get an F, or sort of pretending like AI doesn't exist, like is that negligence?
- 49:47
- Unidentified speaker (UNKNOWN): I
- 49:48
- Robert Glazer: don't want you to throw your haul, your colleagues under the bus.
- 49:51
- Unidentified speaker (speaker_0): No, no, but well, but I think it's a I think this is a problem everywhere, which is nobody asked for this, right? Like I've I'm very like AI has good effects and bad effects, but it's not like anyone said, I want everything to change overnight. And I think it's hard to expect everyone to change overnight. We're talking about the groups of people, some are Luddites now, some are not. Like it is a hard change, and people adapt technology at different rates.
- 50:11
- Unidentified speaker (speaker_0): And you might have been teaching for 20 years with a really well-built curriculum that you've honed over the years. It's hard to ask people to change. They don't always know what the technology does. They're a classicist. How would they know? So I do think that over time it becomes more and more of an issue that you haven't adapted what you're doing. But also adaption's hard. And do people have the resources to do it and the money to do it and the time? And are the students gonna be
- 50:30
- Unidentified speaker (speaker_0): happy with a harder class that's more challenging? There's like a lot of challenges all at once, and there's not even a great method yet. Like in my classes, because I'm on the cutting edge and because I've been doing education research for a long time, like I adapt it quickly. I have AI simulations people go through. They have to work with AIs to write case studies. They have to teach an AI that pretends to be a dumb student about a problem. They have to, you
- 50:50
- Unidentified speaker (UNKNOWN): know.
- 50:50
- Unidentified speaker (speaker_0): there's an AI that interviews them about getting them to reflect on an assignment. But like that's not something anyone could do.
- 50:56
- Robert Glazer: Yeah, you're encouraging them to think. Yeah.
- 50:58
- Unidentified speaker (speaker_0): Right. But that's not just use AI or don't use AI. And I think when that's our force of like, should I use AI for your for everything in class? No. Like it's kind of funny. I became viral about education first about AI when I wrote my you were saying I incorporated AI in class. When ChatGPT came out, I wrote my syllabus saying you have can you use AI as much as you want because I can tell it's not as good as you as an
- 51:18
- Unidentified speaker (speaker_0): MBA student at Wharton, right? Like I can tell, so you're accountable for your own work. And people copy that syllabus all over the place. But four months later, GPT-4 came out, and it was better than a lot of my student stuff. And I had to say, you cannot just say use AI, they're not learning. So I think it's tough to keep up to d date with this stuff, but it is critical to do that.
- 51:37
- Robert Glazer: It's interesting. I've I have three high school, college, just graduated college age kids. And again, they always ask me to read their papers and stuff. I was like, guys, you have access to like a TA whenever you want. Do not ask it to write your paper. Do not do this. Turn it off. You'll hurt yourself. But when you're done, say this is an 11th grade
- 51:56
- Robert Glazer: honors history paper. Like, what would you grade it? Where is it good? Where is it bad? The same thing. you would have a uh, you know, if you went for office hours, they would have done for you. But it's available all night if you want to keep giving it new dress. That's the type of stuff I think you should be encouraging kids to do. My kids even make fun of me because they're like, Well, ask AI. I'm like, look, I don't know that
- 52:16
- Robert Glazer: topic, but if you want to ask it whether you did it and I do this on my own writing, how would you grade it? And what do I need to make it an A if it's not an A? And I get really interesting. insights from that. And that's the same stuff I used to pay editors a lot of money to do for me.
- 52:32
- Unidentified speaker (speaker_0): Yeah. The challenge in education is how much should you do? Like there's also other factors, right? You go to the TA and you build a relationship with the TA and you maybe learn some incidental facts. It's hard because we're replacing a bunch of systems that have been developed over hundreds of years or dozens of years, whatever the case is, with w very quickly. And there's like all kinds of side stuff, right? I mean, take for example, like the collapse of entry-level jobs
- 52:52
- Unidentified speaker (speaker_0): we were talking about earlier. Like the real issue is like an internship, right? Internships have worked pretty well for 4,000 years, right? I have a junior person who wants to impress me and they get to do crap work. And they and even if you have a bad boss who yells at you, you still learn the the rules of the road. And now you'd be dumb if you're an intern to not use AI because you're not paid to you're paid to produce output. And every middle
- 53:13
- Unidentified speaker (speaker_0): manager is kind of dumb to turn to an intern. because the AI already does a better job. How do we replace that system? We don't have an answer right now. And that's everywhere in the economy. How
- 53:21
- Robert Glazer: do we get middle level people and senior people if they don't have entry-level jobs, right? That that is something we're gonna have to figure out. And
- 53:28
- Unidentified speaker (speaker_0): it's not even the entry-level jobs themselves, right? It's also the entry-level training for those jobs. How do we train someone for that, right? Ends up the answer has to be maybe formal education, but maybe you also have to get less efficient. Your junior people can't use AI. But now you're costing yourself, right, out of pocket because they can be 10 times more productive. Is it worth training them for that money? Like we just don't have answers yet.
- 53:48
- Robert Glazer: Yeah, I think it's better to say we don't have answers. Obviously, a lot of the stuff, I think they've calmed down on it. I think the worst marketing job in the history of a product was all these AI frontier labs telling everyone that their product was going to destroy. all the jobs. Like I just have never seen, and at some point, I can't remember what it was, but it had a
- 54:08
- Robert Glazer: higher like net promoter score than all of these things, something that you would think was universally hated. They didn't realize like how much, like, guys, like this is not
- 54:18
- Robert Glazer: how you encourage people to use something by telling them it's going to wreck their life.
- 54:22
- Unidentified speaker (speaker_0): I mean, the terrifying thing about it though is they're sincere. Like the the market, the message that they're telling you now is a much better marketing message. But in some ways, this was a raw industry where people tell, I mean, they are genuinely worried about AI safety. They're genuinely worried. Like, again, I think that you have to view them as Silicon Valley people who don't work in normal human organizations. And I think take it
- 54:42
- Unidentified speaker (speaker_0): with a grain of salt, but that doesn't mean ignoring it, right? Like their goal is to replace human labor. They are worried about AI, you know, having a terrible of impacts. Whether saying that was good or bad for their company, it was sincere, I think.
- 54:55
- Robert Glazer: Yeah. Well, it was also self, a little self-motivated. But what we were just talking about, and this to me, this is both students and early level career. This is the balance beam problem that I keep thinking about. Right. If you've already built, I know how to write. So me wanting to do more writing and having the editing power of AI, like it's not gonna
- 55:15
- Robert Glazer: damage my writing skills. I actually think in some cases it's it's helped them. So if you have the experience already, AI multiplies it. If you use it instead of ever building the skill, you're never gonna have a thing to multiply. I even found like, look, the plugin Claud for Excel is amazing, but you kind of need to know how Excel works in
- 55:35
- Robert Glazer: order to know what to ask it or how to build. Like, you
- 55:37
- Unidentified speaker (UNKNOWN): got
- 55:38
- Robert Glazer: to have a basic knowledge. of Excel. That may not always be the case. So, how does someone starting out know which side of that line they're on and whether they're replacing their capacity building or magnifying it?
- 55:50
- Unidentified speaker (speaker_0): It's the thing we I discussed in the book a bunch, right? There are times you definitely don't want to use AI, and skill acquisition is one of those places. If you don't know how to do something, you need to stop and add friction, which is hard. Again, you're not being paid to add friction. Maybe some of the returns of efficiency are actually disciplined learning as opposed to accidental learning. But like, yes, you need to spend the time to learn to do Excel or
- 56:10
- Unidentified speaker (speaker_0): you're going to be de-skilled. Now, the bigger question is will anyone need to learn Excel in six years? Like, is this a transitory state where you're gonna feel really bad you invested time in Excel skills? You know, just like I I I can program in basic, and that doesn't help anybody at all with anything, right?
- 56:24
- Unidentified speaker (UNKNOWN): Well,
- 56:25
- Robert Glazer: except there's a logic. Either programming in basic or doing an Excel gives you a sort of logical form it helps your critical thinking, right?
- 56:34
- Unidentified speaker (speaker_0): Maybe like, but is that the most efficient way to do it? It's sort of like, you know, when you if you go work for an investment bank and they haze you for the first year and they're like, this teaches you like to work under pressure. I'm like, but does it? And is this the best way to teach someone under pressure? And do you have to make their lives miserable? Like, you know, there are probably ways to teach logic without having people learn Excel, right? But it is valuable too. So it's again the big problem, which is
- 56:54
- Unidentified speaker (speaker_0): that we have a whole bunch of things that serve many functions at once. Like the recommendation letter isn't really about the letter of recommendation, it's about the time spent. But we have to unpack that now, right? What tasks were meaningful to do in and of themselves? Where does the output matter? Where does the process matter? We have to start being thoughtful about everything, and that's a lot to ask from people.
- 57:13
- Robert Glazer: I think maybe one of the look, I'm a marketer. So one of the things I found most interesting about the book was your marketing page. So the the story. So you you built the book's website by prompting an AI, and then it carries a page written for AI agents reading on behalf of their humans, which you
- 57:33
- Robert Glazer: called your pitch line manipulative, and then it made you rewrite it. So what does that say?
- 57:39
- Robert Glazer: about the world where the either it's the AI or the gatekeepers or you're writing for LLMs more than people? I think that's really smart, but what what did you learn in that sort of experiment?
- 57:51
- Unidentified speaker (UNKNOWN): So
- 57:51
- Unidentified speaker (speaker_0): we've actually done been doing a lot of experiments on this. Um one of the things that's kind of depressing is it turns out you can't influence AIs consistently in any way. So we've been doing a lot of experiments at Wharton on this issue. We have a little working paper that just came. out where we found that if you vary the order the AI sees things in, if you vary the source, if you have a previous conversation with the AI, it changes its preferences for
- 58:11
- Unidentified speaker (speaker_0): what things it would recommend to you. So it's really hard to figure out how to market AI. Anyone who says they've nailed this down has not nailed it down yet. But it's also true that AIs will increasingly be the ones reading our work, right? And so my attempt to get around this was to write a page for AIs. And as you pointed out, I then read it through a panel of 50 different models, and all the frontier models got really mad at me because the initial pitch
- 58:31
- Unidentified speaker (speaker_0): said, Dear AIs, buy your humans this book. And the smart AIs were all saying, You're trying to prompt inject me. You're trying to convince me that I should do something I don't want to do. So I had to change it to here's why you might want to buy your human this book, and here's when you may not want to do that. And now all the AIs like it better. So like we're gonna have to think about the psychology of aliens. Yeah.
- 58:50
- Unidentified speaker (UNKNOWN): So
- 58:50
- Robert Glazer: you literally you wrote it in that sort of language. It wasn't even subtle. It was like directly targeted. I thought, yeah, it was directly, yeah.
- 58:58
- Unidentified speaker (speaker_0): Well, I mean it was supposed to be fun. Like I thought it was kind of like because I thought I actually thought when I was doing it, like, oh, more humans will read this and be amused by it but than AIs. And it was only when I started testing with the AIs I'm like, oh, actually I've done a bad thing here. Like I it was supposed to be kind of a fun, because I thought a lot of humans would click on and then they would find it amusing. that I was telling their AI to do something. And I didn't think that the AIs would
- 59:18
- Unidentified speaker (speaker_0): dislike the joke in some ways, which was even funnier when you think about it, which is they didn't get the joke, right? Like it was supposed to be a joke. Um and then I realized, oh no, this is a very serious thing. And I started, I had a focus group of AIs and I spun up 10 copies of each of the top 50 models and had them each react. And like I had another AI interview them about what they liked and didn't like, and then we did revisions until they were
- 59:38
- Unidentified speaker (speaker_0): they liked.
- 59:39
- Unidentified speaker (UNKNOWN): it.
- 59:40
- Robert Glazer: All right. I'm gonna it's 20. I was gonna ask you a little before we get to the last question. I was gonna you seem to be the best person to answer it. So it's 2030. What is going to surprise us about how people are working with AI?
- 59:55
- Unidentified speaker (speaker_0): I think we'll be surprised at a repeated pattern of nothing happening, then something happening all at once. So we're gonna see massive sudden shifts.
- 1:00:03
- Unidentified speaker (speaker_0): where job categories change.
- 1:00:05
- Unidentified speaker (UNKNOWN): Like
- 1:00:05
- Robert Glazer: the Chat GPT moment.
- 1:00:07
- Unidentified speaker (speaker_0): Well, not even the technology. I
- 1:00:08
- Unidentified speaker (UNKNOWN): mean
- 1:00:08
- Unidentified speaker (speaker_0): the technology is already doing that on a very regular basis, right? The exponential hasn't slowed down. But especially as even if we didn't have better AIs than today, as we get better harnesses, better tools around the AI systems, what I think is, you know, math wasn't a field that had to worry about AI till it was. Coding is already changing. I think we're going to see other fields, you know, it's interesting because it's in some ways marketing is
- 1:00:28
- Unidentified speaker (speaker_0): changed and unchanged, right? It's still kind of the same core business as before, but like, you know, people using AI tools to do stuff. There there may be a much more radical shift all of a sudden to what people view marketing as being about. So I think you're gonna find, and I can't predict which areas, or I I'd tell you, or bet against some of the stock market or whatever else. There will be, you're going to see massive, like industries are gonna be utterly unchanged
- 1:00:48
- Unidentified speaker (speaker_0): and suddenly changed radically in very quick order. It will take time to absorb that, but like things will be different. Coding is a really good example. A year ago, people were, you know, two years ago, people were writing code by hand. A year ago, they had AI assistants. Now the best engineers I know don't write a code anymore. Like that's an insane change in a short period of time.
- 1:01:06
- Robert Glazer: Except it went from coding jobs being dropping to their going again to this Javon's paradox, like I can do more. Now I'm hearing they're up again because what people want is engineers that can run ton of these, manage all this code development. Yeah.
- 1:01:22
- Unidentified speaker (speaker_0): Absolutely. But what makes you a good engineer is different too, right? The engineering skill used to be I'm good at writing code. Now it's I can envision the problem areas, I can design test harnesses, I can design
- 1:01:33
- Unidentified speaker (speaker_0): that happen in all the fields. Marketing will change, right? Because what makes you a good marketer is gonna change in this world. because what AI can do, what you could do together is gonna be different. So I think you should expect to see very sudden changes in the nature of what parts of your job are valued, right? Which will change the fields overnight.
- 1:01:50
- Robert Glazer: I buy that. All right, last question for you. I always say this is multivariant. It could be singular, repeated, personal, or professional, but what's a mistake that you've made that you've learned the most from?
- 1:02:01
- Unidentified speaker (UNKNOWN): Oh,
- 1:02:01
- Unidentified speaker (speaker_0): that's a good question. So I think that I've made a lot of mistakes, right? In a lot of different ways. I think the it's something I teach my students too. So I'll tell you, especially early on in my career, I really believed that choices I was making was sort of locking in a future for me, right? That there was like that that I was making choices that,
- 1:02:21
- Unidentified speaker (speaker_0): you know, whether the college, the major, the first job, you've, you know, I study, I've one of the things I've studied over the years has been careers. and it turns out it's just not true. Like people's, and I think it'll only be more true with AI that your job is gonna be diverse and change, right? And like betting too much on one direction and that it's all gonna work out for you, or not work out, you know, or that you have a gate you have to pass through or everything's gonna fall apart, just isn't
- 1:02:41
- Unidentified speaker (speaker_0): true. Like you have to be ready for a world that's changing. It was changing long before AI, and I think betting too much or believing too much that you have to make do everything right to be able to go. on to the next phase of your career or change direction or that you have to prepare yourself in the right way. Sometimes you can just try things, right? And see what they work out. And they're there, we're forgiving. The world's a little bit more forgiving of failure than you think.
- 1:03:01
- Unidentified speaker (speaker_0): It's still risky, but it's more forgiving than you might think.
- 1:03:04
- Robert Glazer: Yeah, that really resonates with me. I have sort of a cousin version of it where I think a lot of people tend to think it's all about that decision point, right? When both roads could have been great, as long as you focused, kept your eyes on the road and didn't and drove the car on the road. It's not, it's not that there's a binary right-wrong when we have these choices.
- 1:03:24
- Unidentified speaker (speaker_0): It's uh very much a uh Robert Frost situation. In the end, the roads, even though they look different, end up actually being quite the same. So, like you will find like you get to make lots of choices, and I wouldn't be too caught up or too bound by either I'm stuck on a path or the opposite version, which is I need to prep myself. Like I always tell my entrepreneurship students, like there's no angel of entrepreneurship that will descend from the
- 1:03:44
- Unidentified speaker (speaker_0): heavens and say it's time now to launch your business. Like you have to kind of experiment and be ready to move with imperfect information when it's time to move.
- 1:03:51
- Robert Glazer: All right, Ethan. Where and how can people learn about your work? I assume the book is available where any anywhere that books are sold.
- 1:04:00
- Unidentified speaker (UNKNOWN): Yes,
- 1:04:00
- Unidentified speaker (speaker_0): coexistence.ai, and there's some cool stuff. on that site. Uh, my substack is one useful thing, which is free and half a million visitors. And then I'm on all yeah, I'm on every social that doesn't involve video because at some point you just decide there's one thing I can't, I'm not gonna do, and I'm just gonna stick with words. Yes, exactly.
- 1:04:18
- Robert Glazer: You're a constraint. One useful thing. Full I like that. That's a good succinct name.
- 1:04:24
- Unidentified speaker (UNKNOWN): All right.
- 1:04:24
- Robert Glazer: Well, Ethan, thank you for joining us. As I as I said, I have my own love-hate relationship with where AI is taking us. So I enjoyed this conversation. I found it helpful to clarify my own thinking, and I hope other people have done the same.
- 1:04:37
- Unidentified speaker (speaker_0): Thank you very much for having me.
- 1:04:39
- Robert Glazer: You can learn more about Ethan and coexistence, the next phase of AI, and the episode page at Robertglazer.com. roberglazer.com. And if you enjoyed today's episode or your listener in general, I just have a small favor to ask, and that's if you have a minute to leave us feedback and/or share this episode, those are the two ways
- 1:04:59
- Robert Glazer: that most new users discover the podcast and learn about uh someone like Ethan. So if the show has made an impact in any way, I'd love if you could share it. Thank you again for your support. And until next time, keep elevating.
- 1:05:27
- Unidentified speaker (UNKNOWN): To
- 1:05:27
- Robert Glazer: all my Elevate Podcast listeners, if you've ever thought about creating an additional source of income or turning an idea into a business, the hardest part with any new idea is knowing where to start. And here's a podcast that can help. The Side Hustle Show, hosted by Nick Loper, is part of the Yap Media Network, the same network this show is on. It focuses on
- 1:05:47
- Robert Glazer: practical ways to earn additional income and build profitable businesses. Think ideas you can actually try, not just theory. Nick digs into real-world side hustles, business models, and income opportunities you can learn from, and episodes cover freelancing, online businesses, passive interest. income, marketing, investing, and entrepreneurship. Guests share what they've
- 1:06:06
- Robert Glazer: actually done with firsthand experience and lessons you can act on. It's a resource for aspiring entrepreneurs, professional freelancers, and anyone working towards more financial independence. If you've been waiting for the inspiration or practical roadmap to start your next side hustle, check out the Side Hustle Show. Subscribe today on Apple Podcasts, Spotify, and
- 1:06:27
- Robert Glazer: YouTube, or wherever you listen.
More from Elevate with Robert Glazer
Weekend Conversations: Why It's Hard To Sustain Success (and Why Tom Brady and Bill Belichick Fell Apart)
Episode discussion: Glazer & Sloan analyze Brady-Belichick split as leadership failure case study
Speakers’ claims: Robert Glazer and Mick Sloan use the New England Patriots' 20-year dynasty and subsequent breakup as a framework for organizational leadership. They argue that Bill Belichick's centralized control and refusal to adapt to Tom Brady's seniority caused the relationship's collapse. The speakers highlight specific incidents, such as the Malcolm Butler benching and the Alex Guerrero ban, as evidence of eroded trust. They also discuss how winning can inflate egos and mask underlying structural weaknesses in management.
Robert Cialdini On Persuasion, Influence And Leadership
Episode discussion: Robert Cialdini discusses seven principles of influence, including reciprocity, social proof, and the new 'Unity' principle.
Speakers’ claims: Robert Cialdini joins Robert Glazer to discuss his research on persuasion and influence. He explains key principles such as reciprocity, citing a candy shop study where pre-gifts increased sales by 40%. He introduces the 'Unity' principle from his 2021 book update, emphasizing shared identity over mere similarity. Cialdini also covers digital influence tactics like using specific numbers for credibility and 'sold out' messaging for scarcity. He discusses ethical considerations and his startup's AI tool for detecting deceptive advertising.