Consider investing in companies solving "hard problems" in physical-world sectors like manufacturing, energy, and chips, as their specialized knowledge provides a strong defense against AI. The analysis highlights TSMC and ASML as high-conviction investments because their complex chip-making processes cannot be easily replicated by AI. This strategy is time-sensitive, as the arrival of Artificial General Intelligence (AGI) is anticipated within the next 2-3 years, which will fundamentally reshape markets. Conversely, be cautious with the traditional SaaS sector, as many software business models face a significant risk of commoditization. Before investing, determine if a company has a durable advantage that will survive in a world where competitors can be created with a simple AI prompt.

Artificial Intelligence (AI) as an Investment Theme

• The speaker believes we are in an extremely fast-moving and confusing time for technology, driven entirely by AI. He suggests that he can no longer see 5-10 years into the future, but only three weeks or less.
• He posits that AGI (Artificial General Intelligence) is "extremely likely" to arrive in the next 2-3 years. Startups and investors should be planning their strategies around this fact, not just the capabilities of today's models.
• The rise of AI will affect both sides of the market. It's not just startups building new products; enterprise customers (the "buy side") will also use AI to accelerate their own adoption and buying decisions.
• A key question for any AI-related investment is whether it's better to build a new "AI native" product from scratch or retrofit an existing product that already has distribution. The speaker believes the winner may vary by industry.
• Economic pressure will drive progress in AI alignment. To build agents that can work for long periods (a day or a week) without human intervention, companies will need to solve trust and control issues, making the models more economically viable. The speaker is bullish on this creating a positive feedback loop for alignment research.

Takeaways

• Investors should focus on companies that are not just using AI, but are thinking about a world where AGI exists in 2-3 years.
• The most important question to ask about a company is: "What's your moat?" or "What makes a durable advantage in a post-AGI world?" If a startup can be replicated with a simple prompt to a future AI model like Claude 7 or GPT 7, it is not a durable investment.
• Be cautious about companies whose only advantage is being early to the AI trend. An investment thesis should be based on resilience against future, more powerful AI models.

Defensible "Hard Problem" Sectors (Infrastructure, Energy, Manufacturing, Chips)

• The speaker argues that the best "moat" or durable advantage in a post-AGI world is solving hard problems, particularly those in the physical world.
• AI models like LLMs are great at tasks involving information on the internet, but they do not have the "tacit knowledge" locked up inside specialized companies that solve complex physical challenges.
• He specifically highlights infrastructure, energy, manufacturing, and chips as sectors that will likely remain "hard" and defensible even in two years.
• Robotics is noted as lagging behind AI language models, which means physical manufacturing and infrastructure problems will not be easily automated away soon.

Takeaways

• Consider investing in companies that operate in complex, physical-world industries like manufacturing, energy, and infrastructure.
• These companies often have decades of specialized, in-house knowledge that cannot be easily replicated by a general AI model, providing a strong competitive advantage.
• Look for businesses where the core challenge is not just software, but involves atoms and complex physical processes. This is presented as a key defensible strategy against commoditization by AI.

Taiwan Semiconductor Manufacturing Company (TSMC) & ASML

• The speaker explicitly mentions TSMC and ASML as prime examples of companies with a defensible moat.
• Their advantage comes from the immense difficulty and tacit knowledge required to build cutting-edge semiconductor fabs. This knowledge is kept in-house and does not "leak out" onto the internet for AI models to learn from.
• He states as a fact: "Frontier LLMs do not know how to build a cutting edge semiconductor fab." This makes their business models highly defensible.

Takeaways

• The speaker presents TSMC and ASML as embodying the ideal investment thesis for a post-AGI world: solving incredibly hard physical problems with deep, proprietary knowledge.
• The sentiment is highly bullish on these types of companies as their competitive advantage is not threatened by the current trajectory of AI development.

Software as a Service (SaaS) Sector

• The speaker raises a critical question: "Is software going to fully commoditize?"
• He presents a possible future where it no longer makes sense to be a SaaS provider. Enterprises might simply build all their software in-house using future AI tools, making it as easy as writing a single prompt.
• This could lead to a world where consumers no longer download apps but have them generated on-demand by their personal AI.
• The counter-argument is that AI could also be used by dedicated teams to raise the quality bar for software so high that in-house, on-demand apps can't compete. The outcome is uncertain and may depend on the specific vertical.

Takeaways

• The long-term viability of the traditional SaaS business model is questionable. Investors should be aware of the risk of commoditization due to AI.
• When evaluating a SaaS company, ask if its product offers a level of quality and specialization that would be difficult for a customer to replicate on their own, even with powerful future AI tools.
• The pressure of commoditization will be more extreme for tasks that have an "intelligence ceiling," meaning once the AI is "good enough" for that task, there's no further competitive edge to be gained.

Blockchain

• When asked if blockchain could be a solution for building trust, the speaker prefaces his answer by stating, "I'm a huge blockchain doubter."
• Despite his personal skepticism and refusal to buy any, he acknowledges that "the price keeps going up."
• He concedes that in a world where trust is paramount, ideas from the blockchain space are the "right set of ideas."
• He mentions potential use cases where blockchain could be valuable, such as mediating a Universal Basic Income (UBI) to prevent it from being controlled by a central government, or creating trust between AI auditing systems.

Takeaways

• The speaker's sentiment is skeptical but open-minded about blockchain's future role.
• While he personally doubts it, he admits it could provide a necessary trust layer in a world dominated by AI agents and centralized power.
• This suggests that blockchain's value proposition may become more apparent as AI becomes more integrated into society, potentially serving as a decentralized trust mechanism. The investment insight is speculative but points to a potential future catalyst for the technology.