Given the impact of generative AI on company formation efficiency, a $1B venture fund will be oversized; for roughly the next 3–4 years from April 2023, an appropriately sized fund for new investments would be on the order of $50M deployed over that four-year period.
“Look, fund four for me was $1 billion. Does that make sense?... For the next 3 or 4 years, no. The right number may actually be $50 million invested over the next four years.”
View on YouTube
Explanation
AI did increase capital efficiency for startups, leading some VCs to raise smaller, more concentrated funds, though headline mega-funds ($1B+) continued to be raised by top-tier firms through 2024-2026 rather than uniformly shrinking to $50M scale.
Fully replacing entire functional teams in startups (e.g., full sales, marketing, or engineering teams) with AI systems will not be possible for at least several years after April 2023; it will not occur within mere months.
“I think we're still a ways away from startups being able to replace entire teams of people... Well, it's in the years I think for sure we don't know how many years.”
View on YouTube
Explanation
As of mid-2026, AI still has not fully replaced entire functional teams at startups; it augments productivity but full team replacement remains an ongoing, gradual, multi-year process.
AI agents will not be able to fully replace good human judgment for several decades (i.e., not before the 2040s–2050s).
“I think that humans have judgment, and I think it's going to take decades for agents to replace good judgment.”
View on YouTube
Explanation
This is a multi-decade claim (through the 2040s-2050s) that cannot yet be evaluated.
Large, sales- and marketing-heavy enterprise software organizations will begin to be materially cannibalized by AI-agent-based competitors, making their traditional go-to-market and sales motions unnecessary over the coming years (no exact year given, but framed as near- to medium-term).
“So I think it's just a matter of time until we start to cannibalize these extremely The expensive, ossified, large organizations that have relied on a very complicated go to market and sales and marketing motion. I don't think you need it anymore. In a world of of agents and auto gpts.”
View on YouTube
Explanation
AI agents have begun pressuring traditional enterprise sales and go-to-market models through 2025-2026, though large enterprise software incumbents have largely adapted rather than been broadly cannibalized.
Within less than one year from April 2023, one-person teams using AI agents (e.g., AutoGPT-like systems) will be able to reconstruct full-stack equivalents of major enterprise software stacks, demonstrating viable end-to-end replacements.
“you actually want to arm the rebels and arming the rebels, to use the Tobi Lutke analogy here would mean to cede hundreds of one person teams, hundreds and just say, go and build this entire stack all over again using a bunch of agents. Yeah. And I think recursively you'll get to that answer in less than a year.”
View on YouTube
Explanation
Within a year of April 2023, one-person teams using AI agents like AutoGPT had not demonstrated viable full-stack replacements of major enterprise software systems; AutoGPT itself proved far less capable than hyped at the time.
Text-to-video / AI VFX tools like Runway will reach visual quality comparable to The Mandalorian TV show within approximately two years of April 2023 (by around April 2025); the gap between AI-generated storyboards and final production-quality output will substantially close within about 30 months of April 2023 (by around October 2025).
“And I said, hey, when would this reach the level that The Mandalorian TV show is? And he said within two years... The difference between the storyboards and the output is closing in the next 30 months, I would say.”
View on YouTube
Explanation
AI video tools (e.g., Runway Gen-3/Gen-4, Sora, Veo) made major visual-quality strides by 2025, closing much of the gap with professional production, though matching a big-budget prestige TV show like The Mandalorian's full production quality remained only partially achieved.
Within the near term (implicit few-year horizon from April 2023), AI tools will enable hobbyists and amateurs to input a screenplay and automatically generate reasonably good-looking animated movies, but achieving full theatrical-quality, fully AI-generated motion pictures will take significantly longer than two years from April 2023 (i.e., not before April 2025).
“So yeah, in theory, you should be able to train the model where you just give it a screenplay and it outputs essentially an animated movie... So yeah, I think we're close to it now... So yeah, I think we're we're pretty close for, let's call it hobbyists or amateurs to be able to create pretty nice looking movies using these types of tools. But again, I think there's a jump to get to the point where you're just altogether replacing.”
View on YouTube
Explanation
By 2025, AI video tools did let hobbyists generate reasonably good-looking short animated content from text/screenplay prompts, while fully replacing theatrical-quality filmmaking remained further off, matching the prediction's two-tier framing.
If AI capabilities continue improving on a roughly 48–72 hour cycle, then by roughly six months after April 14, 2023 (i.e., by mid-October 2023), the effective progress in AI will be comparable to 10–12 years of progress at traditional technology innovation rates.
“And this is a perfect example where when you start to compound technology at the rate of 24 hours or 48 hours, which we've never really had to acknowledge, most people's brains break and they don't understand what six months from now looks like. And six months from now, when you're compounding at 48 or 72 hours is like 10 to 12 years in other technology solutions.”
View on YouTube
Explanation
This is a vague, difficult-to-measure metaphorical claim about compounding AI progress rates that cannot be objectively verified.
If the United States creates an FDA-style regulatory body for AI in the near term (before clear technical standards exist), then U.S. innovation in AI will slow substantially and other countries that do not impose equivalent constraints will advance their AI capabilities faster than the U.S. and surpass it in AI leadership.
“And what we will do by racing to create a new FDA is destroying American innovation in the sector. And other countries will not slow down. They will beat us to the punch here.”
View on YouTube
Explanation
The US avoided creating a heavy centralized AI regulatory body akin to the FDA, and the Trump administration explicitly pursued a lighter-touch, competitiveness-focused AI policy, though whether this measurably slowed or sped US AI leadership relative to competitors remains debated.
Over time, AI regulation will evolve into a set of domestic regulatory bodies in major jurisdictions (US, EU, Canada, Japan, China), analogous to FDA/EMA, whose AI safety guardrails and standards will significantly overlap and share substantial commonality rather than diverging completely.
“I think you need to have a domestic organization that protects us. And I think Europe will have their own again. FDA versus EMA Canada has its own, Japan has its own, China has its own. And they have a lot of overlap and a lot of commonality in in the guardrails they use. And I think that's what's going to happen here.”
View on YouTube
Explanation
Major jurisdictions (EU with its AI Act, China, and others) did develop their own domestic AI regulatory frameworks with some overlapping principles, though the US notably pursued a much lighter-touch approach diverging from the EU/Canada model, undercutting full convergence.
Broad, coordinated global regulation that effectively stops or tightly restricts AI model development and deployment worldwide will not occur; even if the U.S. imposes strong regulations on AI models, many other countries will not follow in lockstep, and advanced AI models will continue to be developed and exploited competitively outside the U.S.
“If the US tries to regulate it or the US tries to come in and stop the application of models in general or regulate models in general. You're certainly going to see those models continue to evolve and continue to be utilized in very powerful ways that are going to be advantageous to places outside the US. There's over 180 countries on Earth. They're not all going to regulate together... to try and get coordination around the software models that are being developed. I think is is pretty naive.”
View on YouTube
Explanation
No coordinated global regulatory regime effectively restricted AI development worldwide; AI models continued to be developed and deployed competitively across many countries despite varying degrees of national regulation.
David Sacks
Partly Right
01:07:43
aigovernmentpoliticseconomy
If a new centralized AI regulatory body is created to approve AI models/apps, the approval process will primarily advantage politically connected incumbents, and the overall rate of AI innovation (e.g., number and diversity of new entrants and products) will slow dramatically compared to the preceding permissionless-innovation period.
“This will be beneficial only for political insiders who will basically be able to get their projects and their apps approved with a huge deadweight loss for the system, because innovation will completely slow down.”
View on YouTube
Explanation
No single new centralized AI approval body was created in the US, so the specific 'political insiders' capture scenario did not clearly materialize, though this makes the prediction largely untested rather than confirmed or refuted.
As AI advances, law-enforcement agencies will adopt AI-based tools and copilots that make large-scale malicious activities such as mass phishing-site creation relatively easy to detect and counter, reducing the advantage that AI-empowered criminals might otherwise gain.
“So there will be new tools that law enforcement will be able to use. And if somebody is creating phishing sites at scale, they're going to be probably pretty easy for law enforcement eyes to detect. So let's not forget that there'll be copilots written for our law enforcement authorities. They'll be able to use that to basically detect and fight crime.”
View on YouTube
Explanation
Law enforcement agencies have increasingly adopted AI-based tools for detecting large-scale phishing, fraud, and cybercrime patterns through 2024-2026.