Dec 27, 2024 · 1h 40m · more-or-less
#79: The More or Less Holiday Special Pt. 1 (AI Highlights) · More or Less Podcast
Sam Lessin · 27m spoken Dave Morin · 20m spoken Jessica Lessin · 15m spoken Guillermo Rauch · 8m spoken Vinod Khosla · 7m spoken Aaron Levie · 6m spoken Brit Morin · 6m spoken
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this 2024 retrospective special, the hosts of the More or Less podcast review the year's defining artificial intelligence breakthroughs, governance upheavals, and macroeconomic debates alongside featured industry leaders. Through curated archival discussions, the panel critically examines OpenAI's platform dominance and boardroom crisis, Apple Intelligence, the unit economics of autonomous agents, and the transformative impact of generative AI on software development and the open web.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. The hosts hold 67.8% of the talking time here. How this is scored →
The hosts as informed peer 6.7 Guest teaching 3.1 Guest disagreement 2.6 The hosts pushing back 3.7
speaking balance: gold is the hosts, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 1:12:16 Aaron Levie mocks Sam Lessin's zero-sum framing
Levie rejects Sam's claim that AI will deflate enterprise software, accusing him of having a zero-sum mindset and joking about his narrow investments in lumber and yard businesses.
Hardest push from the hosts ▶ 1:23:55 Sam Lessin dismantles the automated Tylenol use caseSam forcefully rejects Levie's flagship consumer AI example, attacking its lack of accuracy safety for pediatric medicine and noting that convenience does not create incremental commerce.
Biggest teaching moment ▶ 44:53 Vinod Khosla instructs startups to ignore compute costsVinod reframes the entire cost debate, telling Sam that optimizing software for current compute prices is foolish because market competition will cut model costs by up to 90% within a year.
The host holds their own ▶ 59:35 Sam Lessin details the fatal unit economics of AI agentsDrawing on operational lessons from his startup Fin, Sam delivers an authoritative breakdown of why compounded error rates make real-world agent execution economically unviable.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The hosts as informed peer | Guest teaching | Guest disagreement | The hosts pushing back | Why |
|---|---|---|---|---|---|---|
| OpenAI DevDay and Platform Dominance vs. Startups | 5 | 2 | 1 | 2 | The cohosts review OpenAI DevDay announcements, comparing the release to the early Facebook platform launch. Britt explains how she showed non-technical entrepreneurs on a webinar how to quickly build custom GPT bots. | |
| Monetization Models and Interactive Content Form Factors | 6 | 2 | 1 | 2 | The group explores how custom GPTs will monetize via app store rev-share and examines push versus pull content delivery. Sam and Dave discuss whether feeding book content into a conversational bot adds incremental economic value. | |
| Media Business Applications and Technical Infrastructure | 8 | 2 | 1 | 3 | Sam explains the technical pipeline of vector embeddings and retrieval augmented generation versus context windows. Dave adds details on OpenAI's Assistants API, code interpreter integration, and expanded 128k context windows. | |
| OpenAI Board Crisis and Ideology of AI Fundamentalism | 7 | 1 | 1 | 1 | Jessica outlines the sudden firing of Sam Altman by OpenAI's board. Dave analyzes how the research culture at OpenAI adopted quasi-religious fundamentalism and deified AGI, pointing to rituals like burning an effigy. | |
| OpenAI Governance Mechanics, Capped Profit, and Ideological Flowcharts | 7 | 2 | 1 | 2 | Sam diagrammatically breaks down OpenAI's circular logic: saving humanity from AGI requires building it first, which requires infinite capital. The hosts contrast the ideological founders with newly hired business operators holding secondary equity. | |
| Microsoft's Exposure, Enterprise Trust, and Startup Ecosystem Fallout | 7 | 2 | 2 | 4 | Britt criticizes Microsoft and VC backers for failing to conduct basic board diligence before investing billions. Jessica and Dave emphasize that Satya's real vulnerability was enterprise client trust and thousands of startups dependent on OpenAI APIs. | |
| The OpenAI Board Vote: Reinstating Sam Altman | 7 | 1 | 2 | 3 | The hosts vote hypothetically on reinstating Sam Altman. Sam votes to discard the 'ring of power' to constrain development, while Dave advocates reinstatement paired with a two-tiered governance check modeled on the Harvard Corporation. | |
| Apple Intelligence and the Paradigm Shift in Human-Computer Interfaces | 6 | 5 | 3 | 4 | Vinod Khosla joins to discuss Apple Intelligence, arguing computers will finally learn humans rather than vice versa. Jessica presses whether Apple is trying to rebuild an App Store by routing to multiple model providers. | |
| Apple's Ecosystem Trifecta and Local Model Developer Restrictions | 0 | 0 | 0 | 0 | Host monologue segment. Dave Morin outlines Apple's trifecta across technology, platforms, and products, noting developer disappointment that local on-device models remain wrapped behind Swift rather than directly accessible. | |
| Frontier Models vs. Small Models and the Economics of Compute | 7 | 6 | 6 | 7 | Sam suggests compute costs will favor small on-device models over expensive frontier models. Vinod flatly disagrees, advising startups to ignore compute costs entirely because competition will make PhD-level intelligence dirt cheap within a year. | |
| Venture Capital in AI: Seed Pitfalls vs. Vertical Defensibility | 6 | 6 | 5 | 5 | Sam claims AI seed investing is a fool's errand where small checks cannot move the needle. Vinod counters with portfolio examples like Symbolica and Hedrol, asserting that disciplined seed investing succeeds when avoiding crowd-following. | |
| OpenAI's Mega Funding Round, Exclusivity Demands, and Secondaries | 7 | 2 | 1 | 3 | The panel discusses OpenAI's massive funding round and requests for investor exclusivity. Sam explains the dual motivations behind secondary sales, balancing investor appetite with employee liquidity. | |
| Model Strategy Divergence: Deep Reasoning vs. High-Speed Search | 8 | 1 | 1 | 2 | Dave reports on OpenAI shifting product strategy toward slower, higher-reasoning models while competitors like Perplexity and Google battle over low-latency search models. | |
| The Economic Reality and Error Rates of Autonomous AI Agents | 9 | 1 | 1 | 2 | Sam draws on his experience building Fin to explain the core economic failure mode of AI agents: even a small error rate in real-world actions creates downstream costs that zero-margin software models cannot absorb. | |
| Google's Browser Advantage and Gemini Deep Research Mode | 8 | 1 | 1 | 2 | Britt highlights Google's browser moat with Project Mariner, while Dave shares hands-on impressions of Gemini Deep Research synthesizing hundreds of sources into spreadsheets. Sam questions how the open web survives when AI destroys publishing economics. | |
| Aaron Levie on the Trillion-Dollar Enterprise AI Opportunity | 7 | 5 | 6 | 8 | Aaron Levie argues AI will unlock a trillion dollars in enterprise software automation. Sam pushes back aggressively, arguing that automation is deflationary and threatens to wipe out legacy enterprise value like Box's own stock. | |
| The Personal App Revolution and No-Code Software Construction | 7 | 4 | 2 | 3 | Dave asks how Box is adapting to an explosion of non-engineers building bespoke personal software. Aaron explains Box's new no-code app builder, which allows users to dynamically configure workflows without rendering probabilistic UI. | |
| Debating AI S-Curves, Query Volume, and Web Monetization | 8 | 5 | 7 | 9 | Sam challenges Aaron on model accuracy S-curves and asserts that AI collapses ad inventory. When Aaron suggests an AI agent ordering Tylenol for a sick child demonstrates new query growth, Sam dismantles both its clinical safety and incremental economic utility. | |
| Guillermo Rauch on AI Search Summaries and Frontier Content | 7 | 6 | 4 | 6 | Guillermo Rauch explains that AI search overviews strip away backlink traffic from generic factual content, incentivizing personal 'frontier content'. Sam argues that destroying monetization for factual reporting breaks the internet's machinery for truth. | |
| Democratizing Software Development and the Value of Taste | 8 | 7 | 6 | 8 | Guillermo argues AI code generation will expand developers tenfold, but Sam counters that trivializing code creation commoditizes Vercel's business. Guillermo defends Vercel's defensibility around operational infrastructure and embedded design taste. | |
| Guillermo Rauch on Polymarket and Invisible Infrastructure | 6 | 4 | 2 | 2 | Guillermo highlights Polymarket as a breakthrough product because it abstracts away crypto plumbing to deliver consumer value. Sam enthusiastically agrees, asserting that great infrastructure must always become invisible. |
Statements from this episode (56)
Lessin: Launching an AI startup is a bad idea due to OpenAI
“I think the idea being that OpenAI has now swiped all these AI startups for sure, and being an AI startup is a bad idea, which has been my line for a while.”
Morin: OpenAI DevDay is most impressive platform launch since Facebook Platform
“Like this is one of, I think the most impressive developer platform launches. Since maybe Facebook platform, if I do say so myself.”
Sam Lessin: Conversational bots are merely content with a different index
“It basically is just content with a different index into it.”
Lessin: People Will Not Want to Consume News Through Chatbots
“Do I think it's how people are going to want to consume their news? No”
Lessin: The Information transitioned its internal search function to GPT
“Yeah, the information search function already, we moved over to like using some GPT based stuff.”
Dave Morin: AI researchers treat AGI with religious deification
“The research side of the AI community has, you know, is largely a secular you know almost always talks in a deeply religious tone. They've made the, you know AI, in particular the AGI technology, into almost a sacred you know, they've almost deified this techn…”
Dave Morin: Fundamentalist ideologies become cults and cause grave miscalculations
“All fundamentalisms, and we've seen quite a, even, you know, in the last few months, we've seen other forms of fundamentalism make grave miscalculations in terms of the actions that they take in the real world. And fundamentalisms in all of their forms are the…”
Lessin: Sam Altman is an AGI fundamentalist, not just a business pragmatist
“Like, you know, and it's like, so it's very interesting to me that the story's become Ilya, The fundamentalist versus Sam, the pragmatist who just wants to build a business, because the reality is Sam Altman, the founder, even as a, you know, private conferenc…”
Lessin: Reporting shows Sam Altman knew he had board problems before ouster
“I mean, I just have so much reporting pointing to the fact that Sam knew he had problems with the board, and then a lot of people were like, well, why didn't he do anything about it?”
Morin: Governance requiring a flow chart starts from a bad position
“I think anytime you need a flow chart to explain how your board works and is hired, you're probably starting from a bad position.”
Britt Morin: Microsoft, Sequoia, and Thrive failed OpenAI board diligence
“Frankly, I'm surprised that Microsoft and Sequoia and Thrive, like didn't do more board diligence when they were investing all this money to figure out like this board was like not adequate for the type of company that this has become.”
Lessin: An $11 billion OpenAI loss is irrelevant to Microsoft
“Eleven billion dollars is completely irrelevant to Microsoft. Right. It's a three trillion dollar company, right? Like they can spend a living billion dollars.”
Morin: OpenAI shifted in one year to lead all of Silicon Valley
“Like this went from being a nonprofit research organization with a lot of really, you know, interesting ideas, definitely fundamentalists, to being the tip of the spear, not just for Microsoft and their customers, but for all of Silicon Valley.”
Sam Lessin: I would not vote to reinstate Sam Altman as OpenAI CEO
“If I'm in the seat, I don't bring him back, right? I say it's the ring of power. I threw it away.”
Britt Morin: I would reinstate Sam Altman with working checks and balances
“I think I bring him back, but I create a new set of checks and balances that actually work to try to meet as many expectations as possible”
Dave Morin: I would vote Sam Altman back into OpenAI with two-tier board
“My vote would be to put him back to vote him back in. And I would do that with the creation of a You know, if it's going to stay a non-profit or, you know, whatever, I would do a two-board structure.”
Khosla: Computers Will Learn Humans Instead of Humans Learning Computers
“This idea that computers will learn humans instead of humans needing to learn computers. And I think we were seeing the beginnings of that.”
Khosla: Apple Partnership Establishes OpenAI Leads LLM Competition
“Obviously from an open AI perspective, it clearly establishes where open AI is in the competition for the best LLMs.”
Khosla: Siri Marks the Beginning of Evolution Toward Agentic Interfaces
“And Siri is the beginning of what will evolve over time into agentic sort of interfaces for humans.”
Dave Morin: Apple uniquely executes tech, platforms, and products simultaneously
“Apple kind of, or not kind of, they really showed that they have the trifecta down there of the ability to build technology The ability to build platforms and the ability to build products all at once. Most of the big tech companies can only do two of the thre…”
Dave Morin: Apple denies developers direct access to on-device models
“At the same time, the one thing that I was disappointed about is that you cannot access these local models directly. It's something that a lot of folks aren't talking about. They've put a pretty intense wrapper around it, at least on the developer front. You k…”
Khosla: Small models will not replace large frontier models for intelligence
“If you want human level intelligence from a model, it'll be a big model, and there'll be a couple of them around, some better than others, some a year or two ahead of others, but I don't think small models substitute
For big models”
Khosla: Frontier AI compute costs will drop 80-90% within a year
“My bet is, a year from now, it'll cost one-fifth to one-tenth of what it does today, and my advice to all our startups is ignore the cost of compute, because any assumption you make, any dollar you spend on optimizing your software will be worthless within the…”
Khosla tells his portfolio startups to completely ignore AI compute costs
“My advice to all our startups is ignore the cost of compute, because any assumption you make, any dollar you spend on optimizing your software will be worthless within the year, and so forget about it.”
Khosla: Open source AI models will not be viable due to training costs
“It is costing an order of magnitude more to train, and that's why I don't think open source models will be viable because the training costs.”
Khosla: OpenAI executes better than Google despite Google's great tech
“I think OpenAI has been very good in execution, but Google has great technology and technologies. They just haven't been as crisp about their execution.”
Vinod Khosla: Khosla Ventures has never taken a dollar in management fees
“By the way, I, since I started the firm 20 years ago, I've never taken a dollar in fees.”
Lessin: AI seed investing is a fool's errand
“I think most seed investors getting into AI are playing a fool's errand because it's not a place where one to three million dollars really can move the needle in terms of value, in terms of what the game is.”
Khosla: Khosla Ventures followed a $2M Symbolica seed with $15M
“And I said, fine, I'll give you a two million dollar seed. And if you prove this out, then we can talk more. They came back in March of this year, not very long ago, and said, here's the results that I promised. I said, don't go out. I'll write you a fifteen m…”
Khosla: Automatically funding anyone leaving Google DeepMind is a bad strategy
“Anytime somebody leaves deep, deep mind, give them two million dollars, that's a bad strategy.”
Britt Morin: OpenAI's exclusivity demand will send investors to competitors
“Like, I think it actually sends people to the competitors.”
Dave Morin: OpenAI has visibly stopped investing in SearchGPT
“They've pretty visibly stopped investing in this search GPT thing.”
Lessin: OpenAI's revenue will ultimately come mostly from businesses
“Like the bulk of OpenAI's revenue, I believe will ultimately like come from businesses”
Sam Lessin: Cheap AI agents fail economically due to downstream error costs
“If it costs zero dollars to use the agent and there is any error rate and there will be an error rate, right? Dave Morin: Absolutely. Sam Lessin: You have no margin to pay back for the error. So the reason that DoorDash works, for instance, is an economic reas…”
Lessin: Fin Exploration warned its investors upfront it would lose their money
“And so, years later, Cortina, who founded Venmo, and I went off, and we just said, look, we're going to start a company called the Fin Exploration Company. And we basically raised some money and told the investors we literally were going to lose all the money …”
Britt Morin: Chrome gives Google an AI agent advantage over OpenAI and Apple
“And I think that, like, plays to Google's advantage in the AI race that they have Chrome. Like, OpenAI doesn't have a browser that they can install this into, and Apple's certainly not, like, ahead in this race with Safari.”
Dave Morin: Gemini deep research handles complex queries no other model can
“And I've been shocked at the kinds of queries that it can do that none of the other models can.”
Levie: Enterprises and government agencies are not at peak digitization
“But like, I don't, I just don't think we're at peak software. Like, if you go to the businesses and government agencies that I've been to, We're just, we're, you know, we are not at like peak digitization.”
Levie: Incumbents like Salesforce, Oracle, and Box will capture enterprise AI revenue
“Now it doesn't mean that it's all startups. It's going to be Salesforce and it's going to be service. Now it's going to be Oracle. And it's gonna be box.net which is not our official brand. We, you know, we will capture collectively all of that.”
Levie: Role-specific AI agent startups will probably outperform AI CRM startups
“I think that the startups that are doing something more disruptive, i.e., you know, pure agents that, you know, go after a particular role or function, like those will probably work better than trying to be like the AI CRM company.”
Levie: Companies will blow billions training models instantly supplanted by Meta's Llama
“There'll be lots of companies that just blow billions of dollars doing, you know, training runs of a model that get, you know, supplanted instantly by Lama.”
Levie: AI will cause 10x more software to be created
“I totally subscribe to the thesis that, that you just said there'll be way more software, 10 times the amount of software that gets created.”
Levie: AI's current capabilities plus 20% would justify all VC funding
“If I'm wrong and it taps out sooner, then I think you would just have exactly what AI is good at today, plus, you know, 20% or whatever, and then fully extrapolate it out. So, so that, that to me is, that, that pays all the, for all the VC funding.”
Levie: Discovery efficiency gains consistently expand consumer commerce volume
“Every time there's been a, you know, any efficiency gain in people discovering things, they just buy more stuff, they transact more, they do more commerce.”
Sam Lessin: Generative AI collapses ad inventory rather than expanding it
“AI today, even if you're a really big bull and you think that Gemini and open AI is like better at answers is not actually going to expand inventory, right? If anything, it collapses inventory because you're doing fewer searches, right? It just gives you the a…”
Sam Lessin: Most internet usage is concentrated around a tiny set of activities
“The internet is very large, but the actual used internet is actually quite small. And so I do wonder to what degree that this is a little bit of navel gazing of all these great uses for I'm going to do all these great things when the reality is, Is like, peopl…”
Rauch: Traffic for generic factual content will stay inside AI front ends
“It used to be a valid content marketing strategy and like marketers in the values to do this, like create a lot of content so that when people search, you'll hit the keywords and whatever, but increasingly a lot of that traffic will just stay in the AI front e…”
Lessin: AI search will break the internet's truth machine
“And the thing I think is so fucked about this whole AI move is, like, in the one boxes in the direction is, like, You're just going to break the entire machine for truth, right? Because you just get a reflection of whatever the economics are of the system. And…”
Rauch: Google estimates there are four million React developers globally
“For what we do, which is React, which is like the front end, there's four million is the best estimate we get from Google.”
Rauch: Generating software is AI's killer application
“Generating software is actually the killer app of AI.”
Rauch: Copilot gave Microsoft conviction in OpenAI after broader bets
“I was interviewing Nat Freeman today for our internal Vercel podcast today, and he was talking about how much conviction Microsoft got into OpenAI because of the effectiveness of Copilot. Like when Satya saw Copilot, when they were doing the initial iterations…”
Rauch: AI will expand React developers from four to fifty million
“There's four million React developers. I can move that from four million to fifty million trivially, because now like you just speak English to that thing. Even if like it doesn't do a hundred percent of the job, it's just like this like infinitely available a…”
Lessin: Vercel will struggle to monetize commoditized AI software creation
“If the marginal cost of building, it goes towards zero. And like, that's the general trend of history. My big concern for your company will be, you're not going to get paid enough for it. Right. Cause like it'll be commodity.”
Rauch: Crypto apps should treat decentralized infrastructure as an invisible implementation detail
“What I saw in Polymarket and starting to see in a lot of other companies is give value to the customer and use decentralized infrastructure under the hood, but don't lead with the decentralized infrastructure. That's just an implementation detail.”