Author: MapleLeafCap , Co-founder of Folius Ventures
I lived in Shanghai for a year and then in Hong Kong for another year, organizing numerous offline events for investors and projects, but I never organized a hackathon. It wasn't that I didn't want to, but I felt that hackathons at that stage of Web3 were somewhat unrealistic—everyone was building perpetual motion machines in a vacuum, interfering with each other's infrastructure, creating nested protocols, and the resulting products were far removed from the average person's reality. You might develop a protocol in 48 hours, then organize an exchange and distribute it through MM. And then what? Where are the users? Where's the PMF? (Related reading: Folius Ventures: Web3 is on the eve of its application boom, about to usher in a golden age for Chinese entrepreneurs )
I moved to Shenzhen this January. Aside from the Spring Festival and the ten or twenty days of moving, I spent most of Q1 working on AI. I took a detour—starting with Cursor, but after a while I realized the ceiling was obvious, and it wasn't until I switched to Claude Code that I started to get on track. Now, my main focus is on CC (Concurrent Code) for multi-agent scaffolding, running Codex concurrently. I've also worked with claws, and currently I'm working on how to set up skill/harness/scaffolding so that the agent runs autonomously, and then replicate that to do investment research and due diligence.
The last time I programmed was fifteen years ago with VBA and Python. I have virtually no technical background or coding skills. But that's precisely what excites me — I don't need to know it, and neither do you.
Let me first state my assessment of the current stage.
We've gone through the stages of Chatbots, single agents, and Agent Workflow, and are now about to enter the Agent Matrix era—where a single person manages a three-digit number of agents and sub-agents, operating automatically 24/7, collaboratively breaking down complex tasks, and producing product modules or even the product itself on an hourly (or even 15-minute) cycle. From a user experience perspective, the leap from Chatbots to stable and efficient agents really happened only last November and December, and the subsequent two phases emerged within a single quarter.
Behind this speed lies the product of four exponentials, further accelerated by recursive self-improvement:
Raw Compute: With brute force stacking, the total computing power of global AI chips increases by 2-3x annually, and more and better chips are being deployed.
Model Capability: With better training, computing power, and algorithms, each FLOP produces 3-4x more intelligence per year.
Token Yield: Through the exploitation of various prompt, context, memory, harness, scaffold, unhobbling, and orchestration techniques, each token produces 2-4 times the actual smart units per year.
Agent Fan-out: More agents replace human dispatch of more agents, and the high-intensity effective operation time of a single agent increases, allowing a person to mobilize 3-5 times the intellectual units per year in a single time period.
Multiplying by four exponential factors, the number of real intelligence units a person can mobilize in 24 hours is 100x per year. With such high efficiency, it's no wonder that Anthropic's growth rate has maintained a staggering 10x per year—it took only 14 months to go from $1 billion ARR to $14 billion.
An individual standing at the edge of LLM capabilities, capable of fully scheduling intelligent units, and able to ship something within 48 hours, is no longer a toy demo.
The best illustration of this speed is the result of a race:
Anthropic February Hackathon: 13,000 people registered, 500 were selected.
Champion Mike Brown—a California personal injury lawyer with no engineering background—created CrossBeam in 6 days using Claude. Previously, over 90% of California ADU building permits were rejected on the first attempt, taking months and forcing homeowners to pay exorbitant prices to support an entire compliance consulting service. Now? An automated check-up is completed in 20 minutes. A real estate administrative outsourcing industry with an annual output of tens of millions of dollars, extremely bureaucratic and cumbersome, has been transformed into a piece of code with almost zero marginal cost.
Third place goes to Michał Nedoszytko—a Polish interventional cardiologist with zero coding experience. In seven days, during breaks between hospital shifts and on flights to San Francisco, he hand-wrote Postvisit.ai . He transformed medical examination reports, which were originally like cryptic texts and highly prone to causing patient relapses and violations, into a plain-language recovery guide. He encapsulated the extremely rare and expensive "brains of top medical experts" into an API that can be accessed infinitely by all of humanity.
Keep Thinking Award winner Kyeyune Kazibwe—a frontline highway technician in Uganda. Traditional road surveying requires purchasing LiDAR survey vehicles costing millions of dollars. He simply took a regular dashcam costing a few dozen dollars, mounted it on the front of his car, and used AI to perform multimodal visual analysis, instantly converting the video stream into a "heat map of infrastructure repair budget" with GPS coordinates. The million-dollar surveying system that developing countries could not afford at all was reduced to a dirt-cheap SaaS with a Dashcam and a few dollars in API call fees.
A lawyer, a cardiologist, a Ugandan road worker. Of the top five, only one has a programming background.
YC Winter 2025 : 25% of batches , and over 95% of the codebase generated by AI . This percentage was almost zero last year.
Then there are some examples outside of hackathons: Peanut (Chen Yunfei) – an economics major, former PM at a major company, who can't write a single line of code. He created "Kitten Lighting" in one hour using Cursor, a lighting app specifically for taking pictures of cats. One of his Xiaohongshu posts garnered 1.18 million views, 73K likes, and 30K downloads, and remained at the top of the App Store's paid charts for over a month.
There's also Zach Yadegari , 18 , from Long Island . Two high school students created Cal AI— a calorie-measuring app based on photos—with 90% accuracy. It has 15 million downloads and generates $40-50 million in annual revenue. It was acquired by MyFitnessPal this March .
When these cases are viewed together, the pattern is very clear:
The winner isn't the one who writes the fastest code, but the one who understands the problem best.
500 programmers lost to a lawyer.
To take it to an even more extreme level, what can you actually do in 48 hours?
Let's do some calculations using someone at the very forefront of this field. Boris Cherny, creator of Anthropic Claude Code: By the end of 2025, his output in one month is equivalent to the workload of a senior classical programming engineer over two and a half years. His own pace is 20-30 pull requests per day, with almost no hand-written code.
90% of Anthropic's internal code is now written by Claude Code, not by humans.
Compress this number into a 48-hour all-out sprint: one weekend equals half a year.
Consider this in historical context. In 2010, Systrom and Krieger spent eight weeks creating Instagram's MVP, attracting 25,000 users on its first day. Two years later, it was sold to Facebook for $1 billion. That eight-week undertaking could be accomplished in a single weekend by someone of Boris Johnson's caliber.
Going back further, WhatsApp was acquired for $19 billion after five years of work by 32 engineers. Based on a four-fold projection to the end of 2026—one person, 48 hours—could almost match the entire five-year engineering investment.
Looking at the two states together:
Classical: One person 48 hours ≈ Toy demo
Currently: 48 hours for one person ≈ the workload of creating an Instagram MVP
One year later: One person in 48 hours ≈ the total amount of work done from creating WhatsApp to its acquisition at $19B.
Examples like this abound in 2026:
Completed in 8 hours using Claude Code, winner of the Anthropic Hackathon Grand Prize
Base44: Maor Shlomo completed Base44 in three weeks while backpacking in the Philippines and Thailand, and sold it to Wix for $80 million in cash six months later.
Anything : A Vibe coding platform created by two former Google employees, it generated $2 million in revenue within two weeks of launch and is valued at $100 million.
Pieter Levels: Created a flight simulator MMO in 3 hours using Cursor, earned $1 million ARR in 17 days, zero employees, and $3.1 million annually across all product lines.
Google Principal Engineer Jaana Dogan confessed that Claude Code completed her team's entire 2024 distributed agent orchestration system iteration in just one hour.
Sixteen Claude agents were used in parallel to produce a 100,000-line Rust C compiler in two weeks, passing 99% of the GCC torture test and capable of compiling the Linux kernel, at a cost of $20,000.
METR Controlled Test: Opus 4.6 has a 50% chance of independently completing a software task that would normally take a human 12 hours. Sixteen months ago, that figure was 21 minutes.
Karpathy—the inventor of the term "vibe coding " —recently confessed: " I've started to notice that my ability to write code by hand is slowly atrophying. " Even he himself can't go back to that level.
Ryan Dahl, the creator of Node.js, said something chilling: "The era of humans writing code is over."
The bottleneck in creation is no longer in the code.
A problem worth solving, a low-cost way to reach users, and the drive to keep going. Code? Code is Free.
The world’s smartest people are learning from each other how to push this boundary.
The information flywheel has already started spinning in the English-speaking world: Karpathy posts a thread, and the entire industry discusses it. An unknown developer's demo gets millions of views through tiered quotes. A single sentence from Altman, Karpathy, or Dario instantly becomes discourse. I shuttle back and forth between Twitter, GitHub, and YouTube in long-form conversations—this is my daily routine.
AI hackathons are popping up one after another in the US: Anthropic, with 13,000 participants, was just the beginning. Various regions, companies, and communities are intensively organizing Vibe coding competitions, allowing builders to experiment with where the boundaries are.
Where is all of this in China?
Model capabilities are indeed catching up. The new "Four Little Dragons" (Minimax, Zhipu, Moonshot, and DeepSeek) plus ByteDance's Doubao, with DeepSeek even driving the price war down to 1/20th, directly altered the industry landscape. However, Q1's overall focus was on model releases and the Spring Festival sales battle. ByteDance was busy with Doubao 2.0 and Seedream 2.0, while Tencent spent over 4.5 billion on Yuanbao red envelopes. No major company held a flagship AI hackathon in Q1.
The information flow is also fragmented. Jike is too niche, Zhihu is too cumbersome and slow, Bilibili is for video consumption rather than social interaction, and WeChat is a closed island—public accounts are one-way broadcasts, group chats are information black holes, and external links are directly blocked.
China has the capability to catch up with the world's leading AI models and the world's largest AI user base, yet there's nowhere for builders to learn from each other how to use them. This is an absurd mismatch.
China urgently needs a public square for AI builders.
Then came a discovery that surprised me quite a bit.
Xiaohongshu (Little Red Book)
Looking back at all the successful hackathons—lawyers, cardiologists, highway technicians—the common thread is: an exceptional understanding of the problem + an exceptionally proactive approach to solving it.
Once we agree that the real winners in vibe-coding/agentic engineering are not programmers but domain experts, we can see that Xiaohongshu occupies a unique position.
The reality for Chinese developers today is this: if you create a product and want to discuss, release, get feedback, and communicate with users locally in China, you have to switch between 4-5 platforms: where there are discussions, there are no users; where there are users, there are no discussions. There's a wall separating the builder and the end user.
I've reviewed all the major platforms in China and have a question: Which platform allows a single AI builder to simultaneously...
Reaching discerning users ✅
Find peers ✅
Found by someone with malicious intent ✅
Cash out directly ✅
Get real product feedback ✅
Only Xiaohongshu has all five features.
Jike Builder has high user concentration, but its scale is too small (millions of MAU vs. 350 million for Xiaohongshu). It lacks action and a closed business loop—builders communicate on it, but users are not.
Douyin has a large user base and monetization potential, but users come to watch short videos not to find solutions.
WeChat has everyone, but public accounts are one-way broadcasts, group chats are closed silos, and external links are directly blocked.
Zhihu has search and high-quality users, but it's too cumbersome and slow, and there's almost no builder community.
Douban has the highest standards of taste, but the platform itself is barely surviving. Weibo has fast dissemination speed, but its audience is entertainment, not a product.
What about overseas? Instagram and Pinterest are theoretically similar in positioning, but the minds of AI builders have long been occupied by X/Twitter—Karpathy is there, Anthropic is there, and all hackathon results are spread there. Instagram and Pinterest do not and will not have AI product communities.
Because there isn't a Chinese version of "AI Twitter," Xiaohongshu has the opportunity to occupy this position in a completely different dimension.
By the way, ByteDance is quite interesting. Its Doubao ecosystem is massive, and Coze Space, with 4.58 million MAU, is developing a builder platform. Its modeling capabilities are catching up the fastest (Seedream/Seedance series), and its overall execution is undoubtedly the strongest in China. However, ByteDance's advantage currently lies in the tool and model layers—it's an arms dealer providing builders with tools, not a user distribution platform for them.
Xiaohongshu's irreplaceable nature lies in the fact that it simultaneously has builders and 350 million end users coexisting on the same platform. ByteDance doesn't have this, and there probably isn't another company like it in all of China.
What would happen if Xiaohongshu took this seriously?
Xiaohongshu's 350 million users represent China's most discerning consumer group: 90% are under 34 years old, over 65% live in first- or second-tier cities, over 50% have a university degree, and their family income is 30-50% higher than the urban average. The conversion rate is 5-12%, with some brands reaching 21.4%—the highest among Chinese social media platforms. This isn't a mass market platform; it represents the most concentrated group of tasteful, affluent young users in China who are willing to pay for quality products.
If your product is ugly, offers a poor user experience, or doesn't address a real problem, tell us directly in the comments section. This isn't a bug; it's the world's best free product feedback, and currently the only place where builders and end users naturally collide on the same platform: discover needs, create products, distribute them, and they go viral—the entire process happens within this platform.
Supporting this viewpoint is the spontaneous explosion of AI and geek content on this platform. Tech content has grown by over 100% year-on-year, the number of creators has increased by over 200%, there are over 50,000 active independent developers, and over 1.1 million pieces of "Build in Public" related content. This wasn't driven by Xiaohongshu; builders flocked in on their own—because they discovered that this platform offered users, feedback, and distribution opportunities.
For example, Xian Xinglang, born in 2008, is a high school freshman who transferred from Shunde, Guangdong to Hong Kong. He came across posts about AI programming on Xiaohongshu (a Chinese social media platform) and started self-studying using Cursor. He noticed the platform was flooded with posts expressing the anxieties of working-class people – they had an outlet, but no tools – so he created "Ox and Horse Clock" to help them calculate their earnings in real time. Another emotion management app, EmoEase, was completed in 18 hours and reached number 5 on the App Store's paid tools chart. He posted on Xiaohongshu that he couldn't afford the $99 annual developer fee, and Xiao Hong, the founder of Manus, saw it and directly sponsored him.
Note this chain: browsing AI content on Xiaohongshu, understanding emotional needs on Xiaohongshu, distributing demos on Xiaohongshu, and finally being seen by users on Xiaohongshu. The entire process would take more than a year, before even finishing high school.
Referring to the case of Anthropic, where a lawyer defeated 500 programmers—in the end, the one who stood on the platform wasn't the one with the most beautiful code, but the one who made the most precise move.
Engineer/Product/User Flywheel:
Builder in Xiaohongshu (Little Red Book) → Build in Public
350 million users directly tried and reviewed the product →
Comment section = most efficient user research →
Product iteration →
Bestselling products attract more builders →
The flywheel accelerates.
As this flywheel spins, Xiaohongshu will become not just "China's largest product recommendation platform," but also China's only integrated platform for the discovery, verification, and distribution of AI products.
It seems that Xiaohongshu itself has also discovered this opportunity.
A couple of days ago, I saw that they'll be holding a 48-hour AI hackathon in Zhangjiang, Shanghai in early April. I did some digging into the participants' profiles and saw a group of Gen Z geeks, which was even more exciting than I expected. Why? Because successful people are getting younger and younger, so young it's making me anxious.
Here's a recent glimpse into real life from across the ocean in March: the two 17-year-old high school students from Cal AI, Zach and Henry, mentioned earlier:
Zach Yadegari taught himself programming at age 7. He started teaching programming at age 10 and beat college students in a hackathon at age 12.
In my first year of high school and college, I created "Totally Science"—a website that helped classmates bypass the school's WiFi restrictions to play games—and sold it for $100,000 when I was 16.
After selling it, he didn't take a gap year. Instead, he teamed up with his old friend Henry Langmack, whom he had met at a coding summer camp, to come up with a new idea.
The motivation was extremely simple: Zach started working out to pick up girls. He was fed up with having to manually enter calories every time he opened a calorie app. So the three of them decided to create something that could calculate calories simply by taking a picture. Startup capital: $2,000, all of which was poured into social media testing.
Zach was 17 years old when he founded Cal AI in 2024.
The company generated $28,000 in revenue in its first month, $115,000 in its second month, and projected $30 million in revenue by 2025. Zach projected revenue to reach $50 million by 2026. Ultimately, it was acquired by industry giant MyFitnessPal.
A traditional SaaS team that hires dozens of senior engineers and takes 7-10 years to complete a $50 million ARR + M&A exit can be done by two 17-year-olds in a year during breaks and weekends.
What an absurd yet sexy rate of commercial compression!
Why are successful people getting younger and younger these days?
The "atomicization" of skill thresholds: AI is bridging the experience gap.
In the past, starting a software company required years of programming experience and a large engineering team.
Now, AI has become an "external brain." An 18-year-old geek, with good prompt engineering skills and architectural intuition, can write code that previously required a team of 10 people.
Conclusion: Experience is no longer the moat; imagination and the speed of using tools are. Young people are not burdened by "old technologies," and learning AI is as natural as breathing for them.
The intuition of "digital natives": They understand the pulse of algorithms better.
Established entrepreneurs find it difficult to replicate the success of companies like Cal AI, which leveraged social media for a cold start.
Reason: Gen Z grew up immersed in algorithms. They naturally know what kind of video can hold viewers' attention in the first 3 seconds and what kind of UI screenshots can go viral on Xiaohongshu.
Advantage: They don't need to hire expensive marketing consultants because they themselves are the target users. This "product sense" is an innate social intuition.
The "small team, high leverage" model: Rejecting the mindset of large companies.
While traditional elites are still pursuing entry into large companies and reporting through multiple layers, young people have already realized that a one-person SaaS can also be a company worth tens of millions of dollars.
Mindset: Young people are more inclined to become indie hackers (independent developers). They aim for $50 million in ARR (Annual Recurring Revenue) rather than managing 500 employees.
Result: This extreme flexibility allows them to complete product iterations within 24 hours based on feedback from communities (such as the Xiaohongshu comment section), while large companies need to hold meetings for three months.
Extremely low tolerance for error: They dare to "fire while on the move".
Veteran mindset: Consider compliance, architectural stability, and 100% accuracy.
Gen Z thinking: Cal AI's initial accuracy wasn't 100%, but Zach dared to launch it directly. He knew that AI would evolve and users would provide feedback.
The logic is: while they're young, they can fail 10 times, but if they hit a pain point like Cal AI on the 11th try, they'll achieve financial freedom directly.
In the past, starting a business was like running a marathon, requiring physical strength, endurance, and decades of experience.
Now, AI startups are more like 'ball-throwing machines'. As long as you react quickly enough and your technique is accurate enough, every ball could be a multi-million dollar ARR opportunity.
The world is rewarding young people who are more curious, more willing to get their hands dirty, and more discerning about the aesthetics of algorithms.
The 16-year-old who won the WWDC award seven times on Xiaohongshu; the youngest AI creator born after 2010 with 300,000 Xiaohongshu followers; and a group of junior high school students who have made their own robots are definitely the key targets of our primary investment. In the AI era, the success of Zach Yadegari (18 years old) and Henry Langmack is by no means accidental, but a complete reshuffling of generational competitive advantages.
The inflation of coding skills has finally delivered the ticket to the business world to the hands of the original generation who are still in high school.
Can Cal AI's success be replicated in China?
Based on the successful path taken by Zach and his team, I have summarized three underlying methodologies:
Frictionless product intuition
Instead of building complex systems, it solves a single, extremely annoying little problem. MyFitnessPal failed because manual data entry was too tedious, while Cal AI succeeded with a simple click. It reduces the number of clicks a user needs to make from generating a request to receiving a result.
"Ghost Teams" and API Leverage
Formula: Mature API + Smooth UI + Powerful Hook = Hit App. Zach didn't train the underlying large model himself. He focused all his energy on user experience (UX) and growth by calling OpenAI or other CV (computer vision) APIs.
Matrix-style "stealth promotion"
Cal AI has hired hundreds of nano-influencers to bombard users with "native content." Instead of hard-sell ads, they create everyday videos like "What I ate today, and AI says this meal has 800 calories." This is a traffic-generating project, which can be easily scaled up as long as there's a standard operating procedure (SOP).
Then you'll find that every step of "The Zach Way," for some reason, can be perfectly supported by underlying data on Xiaohongshu:
This table shows more than just that "Xiaohongshu is suitable for AI products." It shows that Xiaohongshu is currently the only platform in the world where every step of the Cal AI-style success path has a native platform capability to support it.
Therefore, the biggest Wildcard of this hackathon is most likely a middle school student born after 2010. Instead of building an aircraft carrier, they simply used the most mature APIs and leveraged their native internet savvy to successfully run this traffic matrix on Xiaohongshu, creating a product that adults might not even understand, but which immediately went viral in the comments section upon launch.
Subsequent closed-loop simulation: 48-hour product → Direct exposure on Xiaohongshu (algorithm-driven traffic, no cold start required) → User feedback → Iteration → Traffic/resource support (last year's competition champion received 150,000+ in promotion) → Integration into e-commerce seeding chain → Direct monetization.
The value of this matter from another perspective:
If you're a developer—this is the most efficient product validation platform you'll find. You can market yourself, validate ideas, gather feedback, iterate your product, and enrich user needs anytime, anywhere, in front of 350 million users with actionable intent. No website, no paid advertising, no cold start. Your users are in the comments section. Who knows which post will be the next dark horse?
If you're an investor—this is the first roadshow for the next takeoff product. Anthropic hack released CrossBeam, and the Google × XHS summer hack released PlanCoach (App Store 4.8 rating, over 200,000 likes). Discover projects here—it's 100 times more authentic than watching slides on a pitch deck.
A bolder prediction: A new generation of YC will definitely come, and it will be completely different in form.
The timeline has been absurdly compressed—48 hours to ship production, 3 weeks to 1 million ARR, 6 months to 80 million exit. The traditional 10-year pipeline is a product of a different era.
The problem is that two fundamental assumptions of the traditional VC model have been broken: you need to have a company, and your equity needs to have value. But a solo builder might not even have a company; the product just emerges and then explodes in popularity.
Therefore, the new paradigm may not be equity investment, but rather: direct investment in individuals, tied to ARR (Average Revenue Sharing); escrow of funds, programmatic draw, and exclusive use for computing power and AI-native marketing. What investors can truly offer is not money, but token subsidies, agent matrix to help MVPs upgrade to volume-ready, and knowledge of how to run growth strategies.
This direction is already being pursued: Calm Company Fund is offering Shared Earnings Agreements, TinySeed focuses on bootstrapped SaaS, and Station Fund's F/ai arm, in conjunction with Meta, Google, Anthropic, and Mistral, is directly providing $1M+ in credits without equity—compute as investment is already underway. AWS offers $1M in credits, Google $350K, all non-dilutive.
Sam Altman and Dario have both publicly bet on when the first one-person billion-dollar company will emerge. Dario publicly stated that there is a 70-80% probability that the first one-person billion-dollar company will appear in 2026. Lovable took 12 months to grow from $1 million to $200 million ARR, with a valuation of $6.6 billion. Its speed in reaching $100 million ARR is faster than OpenAI , Cursor , Wiz , and all software companies in history, and it is already calibrating this probability.
Interestingly, on the Chinese side: the Qianhai OPC Mavericks Program was just launched on March 20th—the same day as the Xiaohongshu Hackathon. It boasts "eight zeros": 200 sq m of free office space for two years, 50 sq m of free housing, 50P/year of free computing power, free large-scale model trials, unsecured loans, a highly fault-tolerant seed fund, and a talent reward of 600,000 RMB/year. It's open to solo builders globally, with no equity disclosure requirements.
China's policy response speed may be faster than that of many venture capitalists .
Finally, let's talk about mindset.
To be honest, I'm very anxious. This round of AI paradigm shift is quite similar to the DeFi Summer. I'm feeling a lot like I did during Crypto 2019/20: something huge is about to hit us, and the rest of the world is still a little asleep.
Those familiar with Web3 probably share a similar feeling. Long-term exposure to Web3 offers advantages in cold start intuition and non-linear growth—if this hackathon is worth paying attention to, or even signing up yourself, it might be an unexpected entry point.
Now, the first thing I do every morning is check Twitter for new breakthroughs, then browse GitHub repositories, and watch long-form breakdowns on YouTube, but I feel it's still not enough—at least I sincerely hope that China also has a high-density AI information aggregation portal, allowing me to see ideas and products that I can't see on English platforms. Perhaps Xiaohongshu can fill this gap to some extent, or perhaps not. But at least for now, it's the most likely option I see.
What Xiaohongshu currently lacks most, in my opinion, is an open developer API—similar to Twitter v2, allowing third-party developers to integrate search intents, analyze content trends, and create tools. The core reason Twitter has become an information infrastructure is its API layer, upon which the entire ecosystem of tools can grow. Another crucial element is the openness of content embedding and creator analytics—once implemented, builders and investors can more systematically track what's trending and why.
I'm returning to Shanghai in early April, hoping to chat with friends who are involved in primary investment in the AI sector. Previously, I was the one talking nonsense about Web3, but now I'm truly eager to learn and see firsthand what Vibe Native can accomplish in 48 hours. The crayfish event that ZhenFund and Tencent recently held in Shenzhen was a huge success (because there were so many people), which also shows that everyone's enthusiasm for AI applications is real.
We don't necessarily have to talk about AI ; we can talk about how our generation can avoid being left behind in this window of opportunity, haha.
