Nov 22, 2023 · 1h 1m · news
Jeff Seibert: Why OpenAI Will Become an Infrastructure Play | E1085 · 20VC with Harry Stebbings
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 episode of 20VC, serial tech entrepreneur and investor Jeff Seibert breaks down the competitive dynamics of generative AI, outlining why OpenAI is an infrastructure utility and how Apple Silicon holds a silent advantage. He also offers high-value tactical playbooks on startup decision speed, founder alignment, and the pragmatic realities of angel investing.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Harry holds 35.4% of the talking time here. How this is scored →
Harry as informed peer 3.4 Guest teaching 3.3 Guest disagreement 1.8 Harry pushing back 2.5
speaking balance: gold is Harry, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 58:05 Complete dismissal of competitor tracking
Jeff forcefully rejects watching competitors, telling Harry founders should 'ignore them completely,' and dismisses product managers who monitor competition as lacking vision.
Hardest push from Harry ▶ 14:38 Challenging 12-month pivot runway assumptionsHarry explicitly challenges Jeff's premise that 12 months of runway is sufficient to pivot, walking through fundraising lag times and math to question if six months of actual execution time is enough.
Biggest teaching moment ▶ 56:55 Invoking the Lump of Labor fallacyJeff educates Harry on economic history by citing the late 1800s Lump of Labor fallacy, historical NYT robot coverage, and calculator panics to explain why fears of AI replacing jobs are mistaken.
Harry holds his own ▶ 55:09 Citing China vs Canada emission ratiosHarry counters Jeff's assertion on global climate action by bringing specific comparative statistics on China's annual emissions relative to Canada's multi-decade targets.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Harry as informed peer | Guest teaching | Guest disagreement | Harry pushing back | Why |
|---|---|---|---|---|---|---|
| Childhood Dreams: From Legos to Coding | 2 | 2 | 1 | 1 | Harry asks friendly opening questions about Jeff's childhood Lego ambitions and product philosophy at Twitter. Jeff shares anecdotes about consumer empathy without any friction. | |
| Decisive Cadence: The Critical Importance of Speed | 2 | 2 | 2 | 1 | Harry asks about release cadence and speed. Jeff agrees that moving quickly is critical, contrasting Elon Musk's speed with Twitter's previous sluggish statistical significance cycles. | |
| The Peter Principle and CEO Feedback Loops | 3 | 3 | 1 | 2 | Harry asks why most managers are bad, and Jeff explains the Peter Principle and lack of CEO feedback loops. Jeff outlines trial management periods for ICs. | |
| Accountability and Retrospectives: Anchors and Breezes | 3 | 3 | 2 | 3 | Harry asks about company accountability and gently pushes back on whether weekly retros scale to 100 people and whether celebrating wins breeds complacency. | |
| The Core Dichotomy: The Origin of Digits | 2 | 3 | 1 | 1 | Jeff recounts the founding story of Digits, explaining the contrast between real-time product metrics and delayed monthly PDF accounting statements. | |
| When and How to Execute a Company Pivot | 5 | 4 | 3 | 5 | Harry pushes back on Jeff's recommendation of needing 12 months of runway to pivot, pointing out fundraising timelines. Jeff reframes Harry's assumption about treating pivots as incremental 'experiments'. | |
| True Alignment Over 'Disagree and Commit' | 3 | 3 | 2 | 2 | Harry and Jeff express mutual disdain for 'disagree and commit' culture. Jeff shares insights from working with Benchmark's Peter Fenton. | |
| Platform Shifts and Agile Weekly Sprints | 5 | 4 | 3 | 4 | Harry questions whether LLMs will be commoditized by citing historical precedents where closed systems beat open ones. Jeff reframes using the Apple vs. Android ecosystem analogy. | |
| Model Specialization and Domain-Specific Fine-Tuning | 3 | 3 | 1 | 2 | Harry and Jeff discuss model specialization versus generalized LLMs. Jeff explains how fine-tuning models requires high-quality focused data rather than massive data volume. | |
| The Hype Cycle: Thin AI Wrappers vs. Deep Workflows | 4 | 3 | 2 | 2 | Harry criticizes flimsy 3-week-old AI wrappers. Jeff agrees, comparing prompt-engineering wrappers to basic database forms and predicting OpenAI will become infrastructure like AWS. | |
| Apple's Silent Advantage: On-Device AI and Privacy | 4 | 4 | 1 | 2 | Harry brings up compute pricing pressure on OpenAI. Jeff points out Apple's hidden advantage with custom silicon and on-device privacy-focused AI. | |
| Consulting Services vs. Disruptive Upstarts | 4 | 4 | 3 | 3 | Harry argues AI consulting services will be a massive monetization market. Jeff dismisses consultancy value, advocating for native workflow automation, and explains why AI adoption curves will outpace mobile. | |
| Google's Cannibalization Dilemma: Killing the Golden Goose | 3 | 3 | 2 | 2 | Harry asks what Jeff would do as CEO of Google facing search query cannibalization. Jeff argues Google must aggressively kill its own golden goose before competitors do. | |
| Data Acquisition and Proprietary Value | 2 | 4 | 1 | 1 | Jeff explains the value of proprietary datasets like Digits' 100M financial transactions. He surprises Harry by pointing out how Google Photos was built to harvest vision training data. | |
| Amazon's AI Play and Vulnerable Incumbents | 4 | 4 | 3 | 3 | Harry pushes back on the claim that 90% of early-stage AI investments will go to zero. Jeff corrects him by citing actual data from his own portfolio of 97 angel investments. | |
| Liquidity Planning, Secondaries, and Lockup Windows | 3 | 4 | 1 | 2 | Jeff explains how secondary market valuations are down 80% and details the pain of holding stock through 6-month post-IPO lockup windows while valuations collapse. | |
| Lessons Learned in Angel Investing | 5 | 3 | 1 | 3 | Harry shares strong VC insights about check size discipline and warns against buying into temporary consumer traction spikes like Fast or Clubhouse. | |
| Jeff's Biggest Hit: Alchemy and Founder Grit | 3 | 3 | 1 | 2 | Jeff describes turning a $10k angel check into a 200x return via Alchemy's pivot. Harry brings up Fast checkout as a counter-example on taking secondary liquidity early. | |
| Investors Asking Founders for Cash Back | 3 | 3 | 1 | 2 | Harry asks when investors should ask founders for cash back. Jeff outlines scenarios where asking for capital return is justified versus backing founder grit. | |
| Talent Density in Down Markets vs. Hot Markets | 4 | 3 | 2 | 3 | Jeff argues down markets increase startup talent density. Harry pushes back, noting high big-tech cash compensation prevents talent from moving to early-stage equity. | |
| Quick-Fire Round: Runway Climate Change and Reality | 5 | 3 | 3 | 5 | Jeff makes a bold quick-fire claim that runaway climate change is under 10 years away. Harry forcefully pushes back citing China vs Canada emission stats and economic equity. | |
| Quick-Fire Round: AI Impact on Jobs and Historical Technology Fallacies | 4 | 5 | 4 | 4 | Jeff argues AI won't cause net job losses, invoking historical economic fallacies. When asked about competition, Jeff aggressively says 'ignore them completely,' prompting Harry to push back. | |
| Quick-Fire Round: Best Piece of Advice & 24-Hour Hypothesis | 3 | 3 | 1 | 2 | Jeff shares the 24-hour hypothesis rule for executive decision-making. Harry agrees that activity drives learning, and they wrap up the interview smoothly. |
Statements from this episode (74)
Seibert: Google is the most vulnerable tech incumbent to AI disruption
“I think Google's by far the most vulnerable.”
Seibert: OpenAI will evolve into an infrastructure platform like AWS
“I view OpenAI probably evolving more into an infrastructure company like AWS.”
Seibert: Apple's custom silicon provides a massive advantage for on-device AI
“What very few people I think are paying attention to is Apple, because again, they control the silicon. Imagine they're able to pioneer small models that run on device, and then they do custom silicon to make them run. The performance could be outlandish.”
Seibert: Designing software features for the average user is a mistake
“So the product managers who were super data-driven started designing and building features for the average user, because that's what the data told them. And they actually believed that there was something such as an average Twitter user. It's such a huge mista…”
Seibert: Emotional attachment to a problem is a founder superpower
“I think it's so much easier to build something that you personally feel than have to sort of try to interpret other people's thoughts and beliefs about it. So no, I would not be worried about getting too emotionally involved. I think that's a superpower.”
Seibert: Legacy Twitter required two to three weeks for statistical data
“Particularly at Twitter, we need to get stats, SIG data, which always took two to three weeks.”
Seibert: Elon Musk moving fast at Twitter beats standing still
“Agree or disagree with some of Elon's decisions, but he is moving quickly in a direction. That is way better than sort of standing still.”
Seibert: The vast majority of startup founders are bad at running companies
“The vast majority of founders are simply bad at running companies. And I'm sorry, but it's true.”
Seibert: Startup CEOs face severe feedback isolation from internal employees
“I think there's a ton of challenges, and it's even worse for CEOs, because who in your company is going to give you really crisp, blunt feedback on what you did right and wrong?”
Seibert: Investors are too removed to critique CEO meeting behavior
“And your investors aren't involved enough in the day to day to really know. So they can give strategy advice, but they don't know how you're behaving in meetings.”
Seibert: Imposing a paranoid founder mindset on teams is demotivating
“If you have that mentality with the team, it's very demotivating, right? It's like, what are we trying to go to? Like, when are we going to get somewhere? And so it's really important to celebrate small wins.”
Seibert: Crashlytics scaled to 300 million phones in 12 months
“We scaled from zero to three hundred million phones in 12 months.”
Seibert: Crashlytics reaches 5 to 6 billion active monthly devices
“Today Crashlytics is on five or six billion MAU, roughly every active smartphone on earth.”
Seibert: Twitter, Slack, and YouTube originated as completely different products
“Twitter was a podcasting startup. Slack was a game. YouTube was a dating website.”
Seibert: Founders attempting a pivot need 12 months of cash runway
“You need at least a year of cash left, because if you don't have a year of cash, you're not going to have time to see this pivot through. And so it's really like founder grit and enough cash. If those aren't there, return your capital.”
Seibert: Great founders build true alignment rather than 'disagree and commit'
“I will not disagree and commit. I think this is the hallmark of great founders is you need to be able to convince your team and have the trust of your team to go all in on a new direction.”
Fenton: Accounting startups win just by making the product somewhat enjoyable
“When Uber was going against the taxi industry, the NPS scores on taxis was so bad, right? Like even if Uber was mediocre, it would still be way better. And he said accounting gave him the exact same vibes. Like the status quo is just so bad that if you can mak…”
Seibert: Large language models will inevitably become commoditized
“I certainly think we will. And this may not be a popular position. Obviously, opening eyes, charging ahead, sort of leading the way right now. I think the market forces at work I mean, there's just immense energy to have an open source equivalent. Meta appears…”
Seibert: Open-source LLMs will become the 'Android' to OpenAI's 'Apple'
“So the closed may win in terms of having the most advanced model, but I think the Apple versus Android comparison is exactly accurate. So you're going to have something proprietary that might be best because it can be fully vertically integrated. They have, th…”
Seibert: A dominant open-source base LLM will emerge with fine-tuning tools
“My sense is there'll be a very common popular open source sort of base LLM and then tools to allow folks to fine tune it easily.”
Seibert: LLM fine-tuning requires surprisingly little data if high quality
“While it creates, it takes so much time, data, energy, money to train an LLM, fine tuning it can actually be done relatively straightforwardly with a surprisingly small amount of data, as long as your data is very high quality and focused.”
Seibert: Apple Silicon prioritizes energy efficiency over pure compute power
“And look at Apple Silicon, right? They've done an amazing job. Like, yes, they're fast, but instead of pushing the bounds on pure compute, they're pushing the bounds on energy efficiency, which for Apple's use case is critical.”
Seibert: AI R&D will focus on smaller, specialized models
“And so I think there will be a lot of really interesting R and D on how do you make these models maybe smaller and perform it in certain use cases.”
Seibert: AI startups that are thin wrappers will get killed
“Startups are gonna get killed because they're very thin wrappers.”
Seibert: Most angel-stage AI deals are thin OpenAI wrappers
“Most of what I'm seeing on the angel side right now are these very thin wrappers on top of open AI.”
Seibert: Products built in 2-4 weeks scripting GPT are thin wrappers
“If your primary product value is, is scripting GPT, that's a thin wrapper. If you've built it in two to four weeks, that's a thin wrapper.”
Seibert: In 5 years, AI wrappers will look like MySQL forms
“It's really important to me that people view this as a technology. It is like your MySQL database. MySQL was very popular 20 years ago, right? Like it was super cool what it could do. And it's still cool, but if you try to go raise money on you've built a form…”
Seibert: OpenAI won't build direct Notion or Salesforce competitors
“I would doubt if opening eye goes and builds like an ocean competitor or an HR or Salesforce competitor or so on, they probably want to stay at the more generic level”
Seibert: Horizontal AI wrappers OpenAI will solve are bad investments
“If you're working on a use case that's pretty horizontal, that like OpenAI is going to need to solve within five years in order to scale. That's not a great investment, and that's not a good use of your time as a founder.”
Seibert: Apple was caught flat-footed on AI and is way behind
“Apple was caught flat footed for sure. They're way behind at the moment”
Seibert: On-device Apple LLMs could make OpenAI obsolete within five years
“I bet Apple can and will. And so if you project forward five years, if they get to the point where they can run a sufficiently large LLM on your iPhone, then OpenAI is out of the picture. You don't even need to hit their servers. It's just on your phone.”
Seibert: Enterprise AI adoption will follow cloud's path as privacy concerns ease
“And so I think this will go in the same direction. And like you'll, you'll have very clear guidelines around how the companies use the data for model training and it's off limits and so on. And sort of that trust will be overcome.”
Seibert: IBM ridiculously charged enterprises for useless blockchain consulting
“And like IBM's at fault. IBM was consulting, like charging for services to consult on how to adopt blockchain into your enterprise. That's ridiculous.”
Stebbings: AI implementation services will be a major category within years
“I think that actually AI implementation services will be one of the biggest categories in the next few years.”
Seibert: Enterprise AI consulting products will be mediocre and ineffective
“Like, yes, there'll be a lot of consulting to help enterprises adopt AI, and the products won't be that great, and they won't really make that much of a difference.”
Seibert: AI adoption will happen radically faster than previous tech waves
“And so I actually think the adoption curve here will be radically faster and industries will be disrupted probably way quicker than prior tech waves just because of the barriers are so low.”
Seibert: Google must go all-in on AI to prevent existential disruption
“They need to go all in on it. I don't think they have a choice. I agree with you. I think it's existential for them because if AI replaces search, their golden goose has been killed. And so it is way more effective to kill your own golden goose than let and wa…”
Seibert: Top AI models are bottlenecked by memory bandwidth, not just compute
“It's very clear the top models are memory bound as well as CPU bound. So you can't just make the CPUs, the GPUs faster. You need to increase memory bandwidth in sort of on par with that.”
Seibert: AI will not alter established SaaS pricing models in specific industries
“I don't think it'll change how people price in specific industries. Like if your market does per seat pricing, that'll probably stay. If your business and product does consumption pricing, that'll probably stay.”
Seibert: Tech companies are locking down data, reversing 20 years of open APIs
“You're seeing Reddit, Twitter, et cetera, shut off APIs, put in more strict rate limits, et cetera. And so the whole world is starting to lock down data, which was counter to the trends over the past 20 years, when everything was being pushed more and more ope…”
Seibert: Google Photos is one of the most strategic products ever launched
“So Google Photos, Potentially one of the most strategic products ever launched, because it allowed Google to collect the world's largest library of photos ever assembled. And that's how they were able to train their early vision models and so on.”
Seibert: Scale-ups adding AI features like Dropbox are just riding the hype wave
“They really are. And one of the funnier ones to me is Dropbox. Now Dropbox has a built-in AI, and I don't know, I mean, I'm sorry, Drew, but I just want Dropbox to store the files. Bandwagon ride the hype wave aspect.”
Stebbings: Complete startup failures going to zero are rarer than believed
“Actually, I think this is a gross over exaggeration on the mortality rate of startups, which is like going to zero is actually rarer than people give credit for. It could be half your money back. It could be a one X, whatever it is, but it's actually rarer tha…”
Seibert: 30 of my 97 angel investments have failed outright since 2014
“So since 2014, I have angel invested in 97 startups. And I just did the count so far, 30 have failed outright. So about a third, another 19 are still at one X. So basically haven't gone anywhere. And if you look at the overall portfolio, really, it's like 10 o…”
Seibert: Angel investment liquidity timelines exceed a decade
“The time range on seed investing on the angel side is just, is like a decade plus.”
Seibert: Startup book values from 2020-2021 are far from reality
“A lot of them are basically holding at their 20, 21 valuation whenever the last round they raised was, and it's nowhere close to the reality.”
Seibert: Secondary market data shows startups trading down 80%
“I do get data from the secondary markets, and so I can see with some of them like where they're trading versus where their last preferred round was, and it's a bloodbath. It's like down 80% for many of them.”
Seibert's early IPO investments crashed near zero during six-month lockups
“A couple of my early investments went on to IPO. And so we're extremely successful, but then during the six month lockup window went almost to zero. And I couldn't exit almost anything.”
Seibert finds secondary pre-IPO share sales more lucrative than post-IPO exits
“I've actually had more success selling on the secondary market pre IPO than I have had actually waiting the whole time.”
Seibert argues small angel investors shouldn't face six-month post-IPO lockups
“I mean, I understand why, but I don't understand why small time angel investors should be locked up for six months.”
Seibert: Two oversized angel investments I felt certain about went to zero
“So on two of them, actually two of my biggest failures, I was so convinced. I was like, this is a no brainer. This is going to be a huge home run. I'm going to put in way more than I usually do. Usually I try to stay pretty disciplined on check size and both o…”
Seibert: Angel investors should keep check sizes identical across deals
“So my advice would be just like, be very disciplined. It's like do a bunch of deals because you need a portfolio and put the same amount in every time and average it out.”
Stebbings: Varying conviction across early-stage investments is wrong
“The idea that you have more conviction in one versus another at an early stage in particular, totally wrong.”
Seibert: Enterprise software traction is more sustainable than consumer traction
“On the consumer side, I'd say real traction on enterprise software. Potentially more sustainable.”
Seibert exited his $10k Alchemy investment at a 200x return
“A college friend of mine was prototyping what was gonna be a new teenage social network, and I invested 10 K. I was like, the guy's smart. Let's see what happens. He eventually pivoted that into down to lunch, which did well for a bit and then failed, unfortun…”
Seibert's single 200x win paid for all his other angel investments
“Yes. So that is why on paper and like, and in terms of cash returned, I'm actually in pretty good shape, but yeah, it's remarkable how, how the outliers sort of outclass the rest.”
Stebbings turned down a 50x angel return offer in Fast
“I had this offer and I was like, ah, it's like, it was like a 50 X on an angel check. And I was like, no, I buy it. Like I'm going to stay in here. Stripe have just come in. There's going to be a partnership agreement there. No, I should have fucking been fear…”
Seibert: 10-15x paper valuations often fail to deliver actual cash returns
“So I had the opportunity to take some money off the table in a deal. So I got a three X return, right? Returned cash in the bank, three X return on paper. It's gone on to do another 10 to 15 X, but I'm worried that valuation's fake. I don't think it'll ever re…”
Seibert: Investors should avoid asking motivated founders for cash back
“I, so I am biased obviously by the founder side, but I think it's very scenario dependent. So if the founder is still like excited, the team is there, they have grit. Like, I don't think you can ask for cash back. I think like, You want to bet on the founder? …”
Seibert: Startup valuations will crash dramatically with major layoffs in Q1-Q2
“A lot of valuations will still crash from here dramatically. And there's probably going to be huge layoffs going into Q one, Q two as well.”
Seibert: Cash-strapped startups will struggle to raise follow-on funding next year
“What'll be really interesting is the number of companies that raised and have sort of struggled through 20, 23, but now we're down to six months or so of cash. I think next year could be very tough because they're not all going to be able to raise follow on ro…”
Seibert: Many tech employees are trapped at startups with underwater stock options
“A lot of folks, you're right, are trapped in companies with high valuations where their options are likely underwater.”
Seibert: Down markets enable higher startup talent density than bull markets
“I actually love sort of down markets because I think in the very hot markets, it's too easy to raise capital and you peel off all the sort of like fantastic second engineers or product managers or designers or whoever they might be, and they go get funding and…”
Seibert: High late-stage salaries prevent talent migration to early-stage startups
“And that's probably the counterforce is all of these big companies who raised huge rounds in 20, 21 have high salaries. And if you switch, you'll get more equity, but way less cash. And so that's probably keeping folks where they are.”
Stebbings: Frequent childhood moves are a common trait among successful founders
“One that I see is often the most successful founders moved a lot in early childhood.”
Seibert: Customer obsession drives founder success over technical skill
“I'd say looking at my portfolio, it's not even necessarily the most techie founders that are successful, but the most sort of customer obsessed and just like really deeply personally understand the market and really want to solve it. And usually have some patt…”
Seibert predicts runaway climate change is less than ten years away
“Runaway climate change is less than 10 years out.”
Seibert: I would choose to be OpenAI's CEO for a day
“Well, this is for one day, so I would go be CEO of OpenAI, just so I can see the roadmap, and then I'll know what to do from there.”
Seibert: AI will boost productivity rather than replace human jobs
“Ah, so I would say actually, on a good light, AI won't replace many jobs, I think, actually. It'll drive productivity, not replacement, and it'll be like other technologies that have come out, like phones didn't replace people, now you just use them and you ca…”
Seibert: Product managers focus on competitors when lacking product vision
“I think a lot of product managers get too distracted watching the competition. And they don't have a vision for their own product.”
Seibert: Startup leaders must answer team questions within 24 hours
“So when your team comes to you with a question, you need a decisive answer within 24 hours. And if you're slower than that, you're not moving fast enough.”
Shorts cut from this episode
▶ Runaway climate change is less than 10 years away 🌍 | Part (@55:13)
▶ Runaway climate change is less than 10 years away 🌍 | Part (@54:22)
▶ Most Surprising Take on AI 😲 · 20VC with Harry Stebbings (@56:36)
▶ OpenAI Prediction: What Happens Next? 👀 · 20VC with Harry S (@0:04)
▶ How to Promote ICs to Managers 📈 · 20VC with Harry Stebbing (@7:05)
▶ Google's Smartest Move Ever · 20VC with Harry Stebbings (@39:05)