Qiao Wang's Rules for Surviving the AI Era: Hold 40% Cash, Concentrate on Google, Beware of the Crypto Bubble

Podcast Source: Empire

整理 & 编译：Deep潮TechFlow

嘉宾：Wang Qiao

主持人：Jason Yano

原标题：Claude Opus 4.5’s Breakout Moment & Investing in 2026 with Qiao Wang

播出日期：2026年1月12日

Key Takeaways

This week, Qiao Wang joined the show to discuss how artificial intelligence will redefine what it means to be an investor in 2026. We delved into Claude Opus 4.5’s breakthrough moment, why Qiao chose to invest in Google, how to build an investment portfolio for 2026, and how to allocate time wisely in the AI era.

Summary of Highlights

• The current market sentiment is overly optimistic.
• “About 40% of my portfolio is in cash.”
• Cryptocurrencies are not very attractive, but opportunities always exist; focus on analyzing assets one by one rather than just looking at the overall asset class.
• Stock and Bitcoin holdings are roughly 50/50.
• The largest investment is in Google; I also hold Tencent stock, which is a very excellent company.
• An overlooked area by the market is AI-driven biotechnology; AI has huge potential in biotech.
• By 2026, we may see some unicorn startups with only one or two people.
• Some of the most successful AI startups are not companies like ChatGPT or OpenAI, but small businesses leveraging AI.
• The essence of company moats has not changed, but moats in the software sector are rapidly weakening.
• Coding itself is no longer the bottleneck; the key is designing the right “prompts.”
• Gemini’s value is underestimated by at least two orders of magnitude. “I’d be willing to pay Gemini $2000 per month.”
• I think Adobe might be this year’s Google because its current valuation is very cheap, and market perception is somewhat similar to how Google was viewed back then.
• Everyone should learn to code, and must learn; automation tools can help optimize parts of our lives and work.
• AI tools will make already efficient and capable people even more effective and smarter.
• In health, the three most important factors are diet, sleep, and exercise.

Qiao’s 2026 Investment Portfolio

Yano: About a month ago, you mentioned the current market situation and expressed some concern. You also referenced the internet bubble era, like 1996, when some predicted the market was about to peak, but it continued rising for three more years.

Additionally, I saw a tweet of yours discussing the relationship between forward P/E ratios and returns over the next decade. You presented some counterpoints, but ultimately said: “I still feel scared.” I’m curious, what are your current views on the market?

Qiao:

The reason I feel scared is because the market sentiment is overly optimistic now. I can’t say everyone feels this way, but you can see on social media many people showing off their investment gains, especially with the recent market volatility. Quantitative data shows that current market valuations are near all-time highs. Of course, some argue that corporate earnings are higher, competitive advantages stronger, and the economic structure more stable, plus monetary easing policies are supporting the market, so valuations are justified. But regardless, we cannot deny that valuations are at historic highs. Therefore, I haven’t fully allocated into risk assets. In fact, about 40% of my portfolio is in cash. While I might miss some opportunities if the market keeps rising, at least I can sleep more peacefully.

Yano: I listened to Drucker and Müller’s interviews. Müller mentioned: “In my 50-year investment career, this is the most unpredictable market moment I’ve encountered.” If even Drucker can’t predict it, how can we ordinary investors possibly do so?

Qiao:

That’s why I focus more on individual stock selection. If you look at the overall market, valuations are high, but by researching individual companies one by one, you sometimes find very attractive investment opportunities. This reduces your fear even if the market declines. During the 2000 dot-com bubble, although tech stocks performed poorly over the next decade, small-cap and value stocks did well, with an average annual growth rate of 10%. I believe this is the best time for stock pickers.

Yano: How do you view investing in stocks versus crypto tokens right now?

Qiao:

Currently, I think cryptocurrencies are not very attractive, but opportunities always exist, like in 2022. Back then, many good investment opportunities appeared. So now I still focus on analyzing assets one by one rather than just looking at the overall asset class. Overall, there are almost no particularly attractive assets besides the dollar. But if you analyze assets individually, you might find some opportunities. As for cryptocurrencies, I still hold some Bitcoin and a small amount of tokens, but the overall allocation is very small.

Yano: If we try to break down your portfolio, 40% cash, how much of the remaining 60% is in stocks and cryptocurrencies? And what’s the ratio of Bitcoin to other crypto assets?

Qiao:

Between stocks and Bitcoin, it’s roughly 50/50. I still hold a large amount of Bitcoin, but that’s not my ideal portfolio because selling Bitcoin incurs high taxes. The proportion of crypto tokens is very small, less than 1%.

Yano: So, what stocks do you currently hold?

Qiao:

My biggest investment is in Google, and I also hold Tencent stock. Tencent is a very solid company, although its business is relatively low-profile, its fundamentals are very robust.

Yano: Do you think Tencent is better than Alibaba?

Qiao:

Currently, yes. Alibaba has more advantages in AI, but its retail business faces fierce competition. I also hold some Amazon stock. Last year’s market sentiment was very unfavorable to Amazon because all tech giants’ stocks rose except Amazon, which underperformed. I also strongly believe in Amazon’s layout in robotics technology. All these investments are long-term, with a horizon of ten years.

Yano: Amazon is indeed a leader in robotics; they might become the first large company with more robotic workers than human workers.

Qiao:

In the past five years, Amazon’s human workforce has remained stable, while the number of robotic workers increased by 20% to 30%. They’ve significantly improved profit margins over recent years, despite modest revenue growth.

Yano: What about the healthcare sector, like Eli Lilly or others? Any thoughts?

Qiao:

I hold some Eli Lilly stock, but it’s a short-term trade rather than a long-term investment because I have limited knowledge of the pharma industry. I don’t know what will happen in ten years. I know I’ll probably still be using Amazon then, but the pharma industry has many competitors. Eli Lilly has strong patent barriers, but there are many cheap and effective gray-market Chinese peptides competing directly with it.

I believe an overlooked area by the market is AI-driven biotech. While everyone is focused on robots, drones, and AI chatbots, I think AI’s potential in biotech is enormous.

Yano: What are your thoughts on automotive and fintech sectors? Do you hold Tesla or Rivian stocks?

Qiao:

No, Tesla’s valuation is too high, and I don’t know much about Rivian.

Yano: And in fintech? Like Robinhood and Coinbase?

Qiao:

From a valuation perspective, I think Coinbase is more attractive, but neither is a particularly good investment. I hold Coinbase stock, while Robinhood’s valuation is too high.

Yano: Do you think the West will develop a super app similar to “Wanshitong” (All-in-One App)?

Qiao:

Are you referring to a super app that integrates messaging, payments, social, etc., like WeChat? I think in finance, there’s already a trend toward such super apps, like Robinhood. But a comprehensive super app covering all aspects of life, like WeChat, I currently don’t see that happening.

Claude Opus 4.5’s Breakout Moment

Yano: I want to talk about Claude and Opus 4.5. I think the best entry point is the “Oh my God, this is over” moment you mentioned on Twitter. Over the past few years, there have been several moments that gave me the same feeling: one was the release of ChatGPT, another was the first reasoning model, and also Tesla’s FSD V13 (full self-driving). Why do you think Opus 4.5 is such an “Oh my God” moment?

Qiao:

I can’t explain what exactly happened behind the scenes, but I can tell you how I felt as a user. The last time I wrote code was probably a year ago, but seriously coding has been about three years ago. From 2010 to 2017, before entering crypto, I was engaged in quantitative trading, writing code daily—from very low-level C++ to higher-level Python data science code. That was my work for seven years; I also wrote some frontend and backend code. But since leaving Messari, I haven’t written serious code; in recent years, I’ve only done some personal projects, usually dedicating time at year’s end.

What’s different with Opus 4.5 is that previously, you could quickly assemble a demo, but the last 5% was always hard for AI to complete fully, still requiring an excellent engineer to handle details like errors and edge cases. In Opus 4.5, I just need to tell it in simple English what I want to do, providing a very clear specification, and as long as the specification is sufficiently clear and comprehensive, it can complete it in one go.

Claude and Opus might be experiencing a hype cycle on Twitter because I have engineer friends telling me that OpenAI’s latest GPT-5 Pro is almost as good as Claude.

Yano:

I’ve used both, but actually Opus was the first tool that made me switch from ChatGPT. For me, ChatGPT’s memory feature made me rely on it so much that I felt I might keep using it because it knows everything about me, but now Opus 4.5 is better for me.

There’s a new way of working now. I’m not sure whether to call it a chat bot, because the boundary is blurring. For example, I just used Opus 4.5 to do something: we have an eight-person sales team at Blockworks, and usually we need to decide who is responsible for each account and prioritize accounts, like Tier 1, Tier 2, etc. Each salesperson has their judgment, e.g., someone handles 10 Tier 1, 30 Tier 2, and 100 Tier 3 accounts. This allocation is often complex. So I gave this problem to Claude, asking it to integrate relevant info. I told it that if a token’s FDV (Fully Diluted Valuation) exceeds $1 billion, that account gets extra priority; if the project is a public company, even better.

Claude pulled data from sources like CoinGecko and CoinMarketCap, ranked these accounts, analyzed our team’s transaction history with Blockworks over the past eight years, understood the types of products we sell, and based on each salesperson’s likelihood of closing deals with these accounts, assigned accounts to the right person.

Qiao:

During the holidays, I also realized what you said—that the boundary between general chatbots and traditional coding is gradually blurring. When you talk to a general chatbot, it usually generates code instantly based on your needs. Now, when you need to write code, you don’t even have to write it yourself; just express your needs in simple natural language. This makes code assistants and chatbots increasingly integrated.

Yano: Exactly. So why don’t you just use Replit or Lovelace for your Vibe coding work? Why choose Claude?

Qiao:

I think their target audiences are different. I haven’t tried Lovelace’s latest version, but I feel Lovelace is better suited for users who want to quickly build a nice demo or simple app; they perform well in frontend development, though I haven’t used them myself. So I initially used Cursor with Opus 4.5, then tried Claude and Opus 4.5. Honestly, I don’t see a big difference in productivity. For me, they perform similarly, but that’s just my personal experience.

How AI Will Impact Startups

Yano: You’ve witnessed the growth of thousands of crypto startups. How do you see AI affecting these companies?

Qiao:

AI’s impact on startups is huge, especially over the past three years. Basically, I ask each team, especially the technical founders of startups, since 2022 when ChatGPT was released: how much has your productivity improved? Almost every answer is higher than the last. This trend is very consistent. Recently, one team said their productivity increased by about 3 to 4 times.

I think this impact is more pronounced in early-stage startups, and less so in large mature companies, because of a limitation called the “context window” of code assistants. For example, if Claude can handle one million tokens, it can help you very efficiently start a new project, and work well. But if you ask Claude to modify Google’s entire codebase, that’s nearly impossible, right? So these tools are more helpful for early-stage small startups, and less so for big companies.

In large companies, the most effective way to leverage code assistants is by establishing clear inter-departmental abstractions. Breaking complex tasks into smaller parts, narrowing the context window, and inputting these into AI systems makes it easier for the system to understand.

Yano:

But I think calling this “productivity” impact is not entirely accurate, especially for early-stage startups. If you have a company of 100, 200, or even 1000 people, a 3- or 4-fold increase in productivity is a big change. But for early startups, they might not even think about “productivity.” Their mindset is more like: why do we need to hire more people? I see this often in companies I invest in and among friends’ startups.

They think, since AI exists, we don’t need to hire new employees at all. This phenomenon is quite interesting. For example, I recently created two tools for sales: one is a sales commission calculator. Sales teams often ask me how much they can earn. Previously, we had to ask finance to calculate, which was cumbersome. Now, with this tool, that problem is solved.

Another is a sales data dashboard. It can calculate the cost of the dashboard. Before, we needed to ask data teams, and consider whether they used Dune or Gold Sky to index data. If not, we had to pay extra. Now, I’ve developed a dashboard cost calculator based on all the data sources.

What do these small tools mean? They imply we no longer need to hire a dedicated sales support person, so by 2026, we might see some unicorn startups with only one or two people. Such startups may already be in motion, even if they’re not yet valued at a billion dollars, they are growing rapidly.

Qiao:

Exactly. I know many people who are now running a subscription business with annual revenue of $10 million alone. These are often former engineers from Meta and Uber, tired of big company bureaucracy, choosing to start their own ventures.

Yano:

But I find an interesting phenomenon: some of the most successful AI startups are not companies like ChatGPT or OpenAI, but small businesses leveraging AI. These companies are reluctant to reveal their core business publicly. Usually, if you have a startup doing well, with rapidly growing revenue, you want to tell the world, raise funds, or promote on social media.

Do Moats Still Exist?

Yano: How do you view company moats? Do you think the definition of moats will change?

Qiao:

The essence of moats hasn’t changed, but moats in the software sector are indeed weakening rapidly. For early startups, there are almost no moats. Companies like Facebook, Google, Microsoft, and Apple still have strong moats. AI code assistants can’t destroy these moats. For example, Apple’s moat is its developer ecosystem, and Microsoft’s is the high switching cost for users moving from PC to other platforms. In cloud services, AWS, Azure, and Google GCP’s moats are the high switching costs for customers. Also, platforms like YouTube with massive proprietary data can develop very powerful video models. Similarly, Microsoft’s enterprise software is a critical tool. Sure, you can copy Office’s functions, but will enterprises really switch from Office to other software? These tools are too important, switching costs are high.

But I recently noticed an example that might be one of the biggest undervaluation cases in the current financial markets: Adobe. Photoshop and Adobe’s creative suite are very well known. There’s a view that the latest video and image generation models will replace Adobe’s products. But I think this is completely wrong because Adobe’s moat lies in its enterprise-level integration. Many companies using Adobe’s creative suite store their images and videos in Adobe Cloud. For creative professionals, switching from Adobe Cloud to other services is very costly.

Moreover, many creative workers have used Photoshop for many years, and their muscle memory is well ingrained. Switching to other tools is very difficult for them. So Adobe’s current P/E ratio is only 12, which is astonishingly low for such a high-quality company.

Yano: Suppose you work at a company, say a startup that’s been around for ten years with 500 employees. How should the company respond? Many young entrepreneurs are familiar with these technologies, but for a more mature company, it’s not so easy to adapt. Employees might just throw things into ChatGPT and send automated emails.

Qiao:

I think you shouldn’t force companies to adopt new tech; you must let them discover the impactful applications of AI themselves and see the huge potential. This is actually what we’ve experienced at Alliance over the past three years. We knew AI would be a major trend three years ago, and we aimed to become an “AI-first” organization. But we didn’t try to impose AI everywhere.

Instead, we use AI to automate specific workflows. For example, every year we receive thousands of applications. Three years ago, I had to personally review all of them—about 5,000 per year. That was very exhausting. Now, we’ve automated about 50% of that work with AI.

Yano: How did you do that? Did you use Opus 4.5? How is the system built? What are the inputs and outputs? Why only 50% automation? Why not 99%?

Qiao:

Yes, this software was developed by our engineers. I believe the code itself is no longer the bottleneck; the key is designing the right “prompts.” I wouldn’t call it a secret weapon, because everyone knows some traits of successful founders.

I simply translated these rules of thumb about excellent founders or startups into prompts to filter applications. Currently, this system mainly filters out obviously unqualified applications, rather than directly identifying the best candidates, because I believe that still requires human judgment. AI hasn’t reached that level yet, but I believe by the end of this year, AI and AI-driven VC will outperform humans in this area.

Yano: I think the prices of these tools are ridiculously low, like when Uber launched, a $5 ride from Wall Street to the Upper East Side. Back then, it seemed obviously wrong, but they subsidized with venture capital to attract users.

How valuable do you think these tools are? How underestimated are their current prices? For example, how much would you pay for Opus 4.5?

Qiao:

As for Opus, I’m not quite sure yet; it depends on what I can build with it. But in terms of Gemini, I think its value is underestimated by at least two orders of magnitude. Right now, I pay only $20 for the professional version, and I haven’t even upgraded to Pro Plus.

Yano: If Gemini charged $2000 per month, would you be willing to pay?

Qiao:

Yes, because its capabilities are very powerful. In fact, I’ve said on Twitter that it’s a research assistant, junior researcher, basic coding helper, a pretty good medical advisor (can verify doctors’ advice), and a pretty good legal assistant. Combining all these functions, $2000 is a bargain.

Decision to Invest in Google

Yano: How much time do you spend daily using Gemini and Claude? How do you carve out time from your daily activities? For example, reducing Zoom meetings or spending less time on social media?

Qiao:

Actually, I analyzed my time allocation six months ago, which prompted me to invest in Google. I checked my iPhone usage records and found that the top three apps I use are Chrome, YouTube, and Gemini—all Google products.

Of course, I was initially worried that ChatGPT might threaten Google’s search business, so I discussed it with my wife. She gave me some insights: she uses Google Search mostly for shopping, and in the short term, ChatGPT can’t fully replace that function. Then I did some research and found that more than half of Google’s search revenue comes from shopping ads, which reassured me that Google’s core business remains very solid.

In the short term, no one can shake its position. Plus, Google has GCP (Google Cloud Platform), TPU (Tensor Processing Units), and other technological advantages, which I only realized later. I believe these factors make Google’s moat very strong. That’s why I decided to invest in Google last year—almost my only big investment.

Yano: On New Year’s Eve, I had dinner with a Google employee, and we discussed these things. Everyone was talking about their favorite tools. That Google employee mentioned that many people overlook a fact: Google controls vast amounts of shopping data, and the market has not fully realized this. Are there other similar opportunities?

Qiao:

I think Adobe might be this year’s Google because its valuation is very cheap, and market perception is somewhat similar to Google’s back then. I’ve never used it myself, but I know it has many use cases.

Yano: No new users will start using Adobe. Like Google, my most used app is YouTube, which I use very frequently, and Gemini. But Adobe? I don’t use it at all, so I think new users won’t choose Adobe either.

Qiao:

But all new users will use Google. I think a common misconception in the market is that Adobe is not a consumer-facing product but an enterprise product.

Yano: Enterprise products are true, but my company would never consider using Adobe; Figma is more popular.

Qiao:

Figma and Adobe actually serve two different markets. For new project demos and website development, Adobe has lost out; Figma dominates this space. What about Canva?

Yano: Canva targets the low-end market, more suitable for amateurs, while Adobe serves high-end enterprise users. So I agree with your point that Adobe’s new user growth has stagnated for a long time, but its pricing power is very strong, allowing it to keep raising subscription fees.

How to Allocate Time Wisely in 2026

Yano: As a Google user, I want to discuss investment, but more importantly, how should we allocate our time in 2026? How do you think AI will change our time management?

Qiao:

Well, I think everyone should learn to code, and must learn. If not, you risk falling far behind. But here, “learning to code” doesn’t mean traditional programming, but using automation tools to optimize parts of your life and work. I believe many B2B SaaS tools like Gmail, Zoom, etc., will still be paid services. But at the same time, everyone’s workflow will have very personalized needs, which third-party software companies might not develop specifically.

Yano: For example, we can develop a commission calculator at Blockworks. That’s a good example—no one would develop a dedicated SaaS just for that.

Qiao:

And when I say “everyone should learn to code,” it’s not meant to scare people. Because current technology has already made it unnecessary to write complex code manually. You just need to talk to systems in natural language to automate tasks.

So, I strongly recommend everyone, not necessarily to use complex tools like Claude Opus, but some simpler, more accessible ones. For example, Replit. When I first realized AI’s potential, it was through Replit. That’s when I understood AI can help you build anything. This technology will change our world profoundly. I highly suggest everyone try Replit—it’s really amazing.

Yano: How do you think this gap will affect the labor market?

Qiao:

I think it will be like the spread of the internet—these tools will make capable, efficient people even more effective and smarter, while those who are less efficient will fall further behind. AI is an extremely powerful tool that can significantly boost productivity, but ultimately it depends on how much you’re willing to leverage it.

Using AI Models for Investing

Yano: In a way, from the opposite perspective, do you think the Warren Buffett stock tracker you built is really worth the time?

Qiao:

Yes, it took me quite a bit of time. Although the results are good, more importantly, I gained a deeper understanding of the relevant technologies during the process.

This tool can be seen as a digital clone of investing masters like Warren Buffett, Charlie Munger, Howard Marks, Peter Drucker, and Bill Miller. It scans thousands of stock codes regularly and deeply analyzes each stock, trying to simulate Buffett and Munger’s investment logic.

The coding part is actually simple, but designing the prompts took me months of iterative adjustments. I created a very detailed prompt set to mimic Buffett and Munger’s thought process in evaluating potential investments.

This process involves six steps: first, a deep research model gathers information on six key aspects they might focus on; then, it calls another API with a reasoning model for analysis. Deep research and reasoning are two separate modules.

The deep research model excels at collecting facts and data, while the reasoning model, though sometimes generating errors, performs better in logical inference. Once accurate data is fed into the reasoning model, its analytical capability far surpasses that of the deep research model.

Next, I invoke the reasoning model in the second phase to simulate a “digital Buffett and Munger” investment committee, analyzing whether a stock is worth investing in. It then outputs a specific recommendation, completing the process.

Have you heard of a theory? Some believe Renaissance Technologies discovered large language models (LLMs) long before others, but kept it secret, which might explain their extraordinary returns. My design of the Buffett and Munger models aims to avoid competition with Renaissance. They are very skilled at short-term trading, like day trading or weekly trades. In such short timeframes, cutting-edge AI models are hard to beat.

Yano: Are you competing with firms like Susquehanna?

Qiao:

Exactly, but my models lean toward long-term investing. Currently, almost no one in the market has the patience to hold a stock for more than 5 minutes, which is where language models can be useful.

Yano: How do you combine advice from different investing legends? For example, Howard Marks and Peter Drucker might have completely different views, and their thinking can be opposite to Buffett’s.

Qiao: I weight their advice and synthesize it.

Yano: Do they all recommend buying Adobe?

Qiao:

Yes, and an interesting phenomenon is that if you run the same prompt multiple times, the model might give different answers each time. If you run the same question repeatedly and average the results, and each time it recommends buying a stock, then the confidence in that recommendation is greatly increased.

It has recommended about ten stocks, four of which are already in Berkshire Hathaway’s portfolio—like Chubb, an insurance company, and Google.

How AI Will Change Branding and Distribution

Yano: How will AI change branding and content distribution? For example, we mentioned Delphi before. Have you noticed OpenAI recently launched the Sora tool? With Sora, you can clone someone’s image and embed it into your videos. This makes me think about marketing and branding changes. In the near future, branding marketing will become more personalized.

Qiao:

This trend has already begun. A few days ago, I saw an ad on LinkedIn that said: “Hey Jason, as a co-founder of Blockworks, I really think you’ll like Rippling.”

Yano:

AI will make their ad algorithms smarter, making ads highly targeted. Actually, a few days ago I searched online for a term—High Rocks. It’s a CrossFit-like marathon fitness event. I’ve been training for High Rocks, so I searched for High Rocks fitness apps in the App Store and found some specialized apps for High Rocks training. But High Rocks is a niche activity, so there shouldn’t be so many dedicated apps.

I found it strange and wondered whether these apps are truly designed for High Rocks or just targeted ads for users searching “High Rocks.” Sure enough, these apps are just general fitness and nutrition apps, but their ad targeting is very precise, so I think this kind of personalized advertising will become more common in the future.

Health and Longevity

Yano: What do you think are the most effective changes in health?

Qiao:

Ultimately, the three most important factors are diet, sleep, and exercise. Back in 2021, I was very focused on optimizing every detail—trying various health supplements, sauna routines, etc. But after four or five years of personal practice, reading tons of research, and listening to countless podcasts, I realized that nothing beats getting enough sleep (8 hours), maintaining a healthy diet, and sticking to regular exercise.

Yano: How do you arrange your diet and exercise?

Qiao:

Honestly, I no longer obsess over extreme diet optimization. That creates too much pressure. I just try to eat healthily without overburdening myself.

Yano: That reminds me of a meme about neurology, showing someone preparing complex meals daily and taking 17 supplements every morning. Your current attitude is “Eat healthy, don’t overthink it.”

Qiao:

Exactly. I find that if I try to optimize everything to the extreme—like Brian Johnson (note: Brian Johnson is known for extreme health management)—I feel a lot of pressure. Stress raises cortisol levels, which is not good for longevity.