Dialogue with Guo Yu: What Should We Deliver When Software is as Cheap as Milk?

Author: indigo

Youtube: INDIGO Digital Mirror

Guest: Guo Yu, former senior employee at ByteDance / retired programmer

This article follows the thread of a conversation, discussing the past of ByteDance and the transformation of the engineer's role, from Agent OS to the stratification and commercialization of intelligence, ultimately landing on a seemingly distant yet already arrived topic: when work is no longer a necessity in life, how should we, especially young people, live?

In 2015, Guo Yu first encountered products driven by neural networks at ByteDance; ten years later, he created over thirty products alone in Tokyo using Agent, only to fall into burnout. In this episode of INDIGO TALK, we start by discussing how Douyin unexpectedly triumphed, touching on the decision-making burdens behind Vibe Coding, Agent OS, the five levels of intelligence, and why software can be as cheap as milk. When software can be generated at any time and discarded after use, what is it that we truly need to deliver? Finally, when work is no longer a necessity, how should young people choose?

01. 2015: The First Encounter with "AI-Driven Companies"

Long before the emergence of large language models, ByteDance was already a company driven by neural network products, and even the engineers themselves could not fully explain why the system operated in such a way.

Guo Yu's tech career began in 2011 at Alipay. In April 2013, he first encountered Bitcoin during an internal sharing session at Alipay, when its price was around $100. That year also marked the true explosion of mobile internet in China: feature phones transitioned to smartphones, and many professionals from southern tech cities flocked to Beijing. Many of Guo Yu's colleagues at Alipay jumped to mobile app development or joined smartphone brands like OPPO and vivo.

In 2014, due to some personal life changes, Guo Yu decided to move north and joined a startup founded by a friend on Suzhou Street in Zhongguancun, where the team worked on Duoshuo and Tuchong, sleeping on the office floor at night. Interestingly, part of the company's investment came from Weibo Fund, and at that time, Indigo was responsible for platform business and investment at Weibo. The two did not know each other then, but years later, they connected the dots in this episode. It was also because Weibo had invested in ByteDance early on that this relationship became an opportunity for acquisition later.

In December 2014, Guo Yu received a call from Shen Zhenyu, saying that Toutiao wanted to acquire them. His first reaction was, "What is Toutiao?" At that time, no one knew the name ByteDance. Beijing was full of interesting startups, and 36Kr and the nascent Keep were fancier than Toutiao; although Toutiao had just completed a $100 million Series C financing (Indigo added that Weibo also participated in that round), it was not widely used among young people. After the merger, some employees even left after a few months, feeling that the company was "not very reliable."

What truly made Guo Yu realize ByteDance was different was a small story. Zhang Yiming required employees to recommend their products to relatives and friends. Guo Yu had a relative who installed Toutiao and was particularly averse to a certain type of news, so he clicked "not interested." However, a few days later, similar news still appeared on the homepage. Guo Yu went to find the engineer responsible for the recommendation algorithm to report this bad case, and the explanation surprised him: they were actually also tuning engineers, and the recall strategy set for each experiment could not completely block such content; they themselves could not clearly explain how the system operated internally.

"Some code is already running in a black box, just like today's commonly used neural network-driven large language models; you don't know exactly how it works, you can only continuously improve through experimentation to approach the desired effect." [05:45]

This is the situation everyone faces with large models today, except ByteDance had already gotten used to it ten years ago. In the Alipay where Guo Yu worked and the startups he knew, none of the products were driven by neural networks; whereas at ByteDance, products and business were divided into a powerful business support department (then called the middle platform) and a strong commercialization department, with the recommendation algorithm being the engine that ran through everything.

Indigo inserted a segment about Weibo here, which formed a stark contrast. During the same period, Weibo was also improving its information flow, but it still used more manual rule-based algorithms. Weibo's fundamental problem was that it prioritized relational networks, much like today's X, where the relationship of following is very important. Once content is filtered by recommendation algorithms, users would complain: "Why can't I see what the people I follow post?" Weibo struggled for a long time between the recommendation flow and the following flow, ultimately failing to do well. The divide between the two companies was, to some extent, already determined at that time: one started from social relationships, while the other started from algorithms.

However, Guo Yu also reminded that even though text and image-based recommendation products rely on neural networks, the audience they can cover and the effects they can achieve are far less than later video recommendations. The year 2015 was actually quite difficult for ByteDance: Douyin had not yet launched, and Toutiao's DAU, MAU, and retention had all hit a bottleneck, with a large number of employees leaving.

02. Douyin was an Accident

The success of Douyin was not the result of prior betting, but rather a victory in the long-term internal competition with Kuaishou, thanks to its excellent "user generalization."

Many people think that ByteDance bet on Douyin from the beginning, but Guo Yu said that was not the case. Zhang Yiming initially wanted to create a more mainstream video recommendation product, so a lot of resources were allocated to the mass-oriented Kuaishou. When Douyin started in 2016, it was just a small team, and it was not even called Douyin, but Z, with a product form very similar to musical.ly.

The turning point came after the acquisition of musical.ly. Alex (Zhu Jun) joined, and both the product form and the underlying recommendation engine were significantly enhanced. The competition between Douyin and Kuaishou continued for four to five years, and ultimately Douyin accomplished a key task: user generalization.

What Guo Yu refers to as generalization is turning a product loved by 18 to 20-something young people into one that can be used by everyone from 7 to 70 years old. Once Douyin achieved this, the historical task that Kuaishou was supposed to undertake was completed by Douyin, leading to their eventual merger. In December 2016, after the team had been working on Douyin and Kuaishou for over ten months, ByteDance's overall situation significantly improved in 2017.

During this time, Guo Yu was still in the middle platform, and the supported businesses changed rapidly. He worked on live video streaming and participated in the once-popular quiz live streaming product "Million Heroes," where ByteDance invested a lot of resources in programs related to the CCTV Spring Festival Gala, allowing users to answer questions in real-time on their phones to win a million yuan in prizes. Later, he was responsible for what was called Micro Runtime, which is the mini-program engine running on Douyin and TikTok. Games and various JS Bundles could be dynamically published; while scrolling through videos, users might come across an ad, and then below that could be a mini-game they could play directly, which was something Guo Yu initially developed.

He positioned himself as a technical executor: throughout his years at Toutiao, he never led a team and did not want to.

03. Customer Acquisition Cost of 50 Yuan and Unexpected Settlement in Tokyo

When even an "app factory" like ByteDance struggled to acquire new users, Guo Yu judged that the mobile internet cycle had ended; a pandemic unexpectedly led him to start a new life in Japan.

In 2019, there was a data point that left a deep impression on Guo Yu: the cost of bringing in a new user from outside had reached over 50 yuan. Even an "app factory" like ByteDance, which excelled in retention, found it difficult to acquire new users, and he realized that he had basically completed everything he could do in this era.

In mid-2019, he decided to leave, and the handover took about six months. After returning home for the Spring Festival in 2020, he traveled to Japan, but the pandemic broke out, interrupting international travel, and he could never return to Beijing. Even the handover of his computer was completed by colleagues who returned to Beijing six months later. He ended up applying for a long-term residence visa in Japan, later obtaining permanent residency, and has lived there ever since.

In fact, even when he wanted to leave in 2019, Guo Yu had made a list comparing various destinations. Singapore was the choice of many former ByteDance employees, mainly for tax identity changes, but he grew up in Shenzhen and did not like the tropical climate. Canada was also an option he seriously considered early on; had he chosen Canada at that time, he would likely have taken the fast immigration route during the pandemic (Indigo added that almost everyone passed through at that time). Ultimately, he chose Tokyo because it scored higher overall, and he had already decided to leave the tech industry to live a normal person's life.

In his view, the opportunities in the tech industry were mainly in Beijing, and that round of the mobile internet cycle had basically ended. It wasn't until the AI wave erupted again at the end of 2022 that he realized he had only truly rested for two to three years, and those years happened to be covered by the pandemic.

"Leaving ByteDance was quite a coincidence; it wasn't a complete experience due to the interruption of the pandemic." [11:13]

Indigo met Guo Yu during that time, on Clubhouse from 2021 to 2022. Indigo recalled that during the pandemic, he also started writing on X, and his strongest feeling at that time was that the entire era suddenly accelerated: remote work, new treatment methods, especially the recovery of SpaceX's rockets. His partners invested in SpaceX in 2019 purely for the stars and the sea, even thinking "this thing is definitely not reliable." Later, on November 30, 2022, ChatGPT was born, and the US stock market was suddenly boosted by AI, leading to increased interactions between the two on X regarding US stocks and crypto.

04. From Executor to "Everyone": The Real Cost of Vibe Coding

After AI took over coding, engineers did not become easier. One person had to bear the decision-making burden of an entire small team, which is the root of burnout.

Indigo outlined the context of the past few years: from 2022 to 2024 is the era of Chatbots; by mid-2025, Vibe Coding proposed by Andrej Karpathy began to gain popularity, with everyone writing code in Cursor, moving towards fully automated programming. He noticed that Guo Yu quickly entered a high-intensity coding state, so he asked: what is fundamentally different about this "implementation" compared to being an engineer at ByteDance back in the day?

Guo Yu's answer pointed to the core: the most important change is identity. After coding capabilities were taken over by AI, engineers became observers.

In traditional tech companies, a small team decides what to do, with one to two weeks as an iteration cycle, setting goals, developing, designing, testing, releasing, and observing experimental results. With Agentic Coding, such a cycle can be compressed to one or two people completing it in half a day.

It sounds like liberation, but in reality, it’s adding pressure. You now have to be an engineer, architect, designer, and most importantly, a product manager all at once. AI can handle design, architecture, and technical execution quite well, so you must become a person who is good at asking questions.

"Essentially, you need to constrain the language space of the language model. You need to know what kind of language to use to communicate with it to get the results you want; you need to understand what kind of language structure allows it to continuously advance the task." [16:30]

Here’s a good analogy: writing code used to be like laying bricks by hand; now it’s like standing on a construction site directing an unflagging construction crew, but you have to decide the blueprints, priorities, and which walls to tear down and rebuild. Moreover, because the Agent is constantly running in a loop, you also need to know how it adjusts its direction based on experimental results. Writing code has thus transformed from a relatively simple mental activity into a much more complex one.

Why can’t AI make these decisions for you? Guo Yu believes it’s not just because it lacks context, but more fundamentally, it lacks proactive intent. Without intent, it cannot make all decisions for you. The currently popular Personal Agents, after having sufficient context, may not be able to transition from "giving suggestions" to "making proactive decisions"; he is uncertain, at least for now, they cannot.

The result is a significant increase in psychological burden. Guo Yu has observed that from October and November 2025 to February or March 2026, many of his engineer friends experienced burnout: they worked tirelessly for two to three months, completing the workload of the past one to two years.

"The human brain consumes very little energy, and it cannot keep up with the speed of AI. We face too many decisions, which affects our autonomic nervous system, making us anxious and unable to calm down to truly think about problems." [18:30]

After burnout, many engineers, including Guo Yu, are rethinking two questions: What is truly worth doing? What rhythm should we collaborate with AI?

Clash of Opinions: Should We Make Plans?

Here, the two had an interesting divergence. Indigo shared his experiences from the past month: he wants to turn his knowledge management, trading, and information management tools into products, and he found that the methods he used years ago when leading product teams on Weibo, such as writing Specs and PRDs, are still useful. Documents don’t need to be as detailed as before, but clarifying thoughts before handing them over to AI results in much better usability than just letting it run. So he prefers to discuss an execution plan with AI first before letting it execute, with planning coming first.

Guo Yu, however, has gone further. He mentioned that Claude Code has recently canceled the planning mode because Agentic Coding itself is already strong enough. In the past, you might spend half an hour detailing all Specs before running everything at once; now everyone hopes it can run continuously 24/7, and this continuous operation can cover everything you previously wanted to plan, "because the things we can think of are actually not as many as AI can." He has rarely used the planning mode in the past six months. As SOTA models and Harness continue to evolve, human engineers must constantly adapt, which itself is another source of burnout.

Indigo admitted that if he doesn’t discuss implementation plans with AI, he feels a lack of control. Some things can be abstracted one level up; he doesn’t care how the code is written, but he must be clear about the logic at the business level. So he often lets AI finish and then asks it to explain its thought process to him, "to let it teach me." Guo Yu, on the other hand, said he has been letting go more recently. These two attitudes actually represent two rhythms of human-machine collaboration today: one is to stay in the loop to understand the system, and the other is to step back to the goal level and trust the process.

05. Software is No Longer Fixed Code

Recommendation engines have personalized content for everyone, while Agents have made software itself personalized for each user. When software can be generated instantly, the "distribution" of software loses its meaning.

Guo Yu believes that the more important change than the shift in roles is the change in the nature of software itself.

He shared a story from an all-night chat. In January or February of this year, his good friend Dongxu was already able to control hundreds of virtual machines simultaneously, running all night to accomplish a major task, consuming billions of Tokens in a day. Dongxu said he felt that with this matrix-like Agent network, any software could be created.

This made Guo Yu realize that not only has the structure of software changed, but coding itself has become an instantaneous act: writing to accomplish a task or achieve a goal, then discarding it or keeping it for personal use.

"In the past, our AI-driven recommendation engines personalized the content of the software we created for everyone. But now, even the software itself can be completely customized. Your software is your software, my software is my software, and the software between us may not be compatible and doesn’t need to be." [22:50]

This comparison is quite insightful. Ten years ago, everyone used the same Toutiao, just seeing different content; today, people may not even be using the same app.

Guo Yu spent three to four months vibe coding over thirty products, many of which were not released. Initially, he thought this was no different from working at ByteDance, where many products were stillborn or cut after launch due to not achieving ROI. However, he later realized he was still using the mindset of the previous era: creating software for those who cannot write code.

This is being dismantled by what is called First Class Harness. Guo Yu defines it as: the first layer of Harness personally crafted by SOTA model labs, such as Claude Code, Codex (which Guo Yu mentioned has now merged into ChatGPT), and DeepSeek Harness (DSH). Their advancements have internalized many functions that software needs to accomplish. If you want to build a CRM or various management systems, it can be done in two, five, or ten minutes. For such needs, making software well and distributing it in the app market has already lost its significance.

06. Agent OS: When Agents Become First-Class Citizens of the Operating System

The next step for operating systems is to shift from "application-centric" to "Agent-centric"; in the future, the computing resources we rent will likely be operated by Agent-controlled Virtual Agent Servers.

If the meaning of software is changing, what is the next step for the internet and operating systems? Guo Yu noticed that some people are already creating new distributions based on Linux, treating Agents rather than Applications as first-class citizens, allowing Agents to operate the entire system. However, a new operating system needs a long time to accumulate the basic OS APIs and complex application functionalities.

Thus, OpenAI took another route: buying a large number of Mac minis to train the next generation of models to operate these Macs. He believes this path is quite reasonable because Macs are excellent Unix distributions, with a very mature tool and software ecosystem in creative and film production fields.

From this, he painted a near-future picture: the things you use on your phone are actually connected to a remote Mac, possibly the one in your home, connected back via OpenAI’s Tunnel or your own Cloudflare Tunnel, or to a server in Tokyo. Looking at a longer time scale, say five years, this remotely operated computer will definitely specialize and become some kind of dedicated Linux distribution. Today, when you rent a server in a data center, you have to install Ubuntu and a bunch of software yourself; in the future, all computing power rentals may become Virtual Agent Servers, with one Agent operating the entire system to complete tasks for you.

Indigo supplemented this with his recent experience with Grok Bot. The underlying Linux environment has the flavor of Agent OS: it has a small browser, a file manager, and a terminal. You can check what it’s doing, but most of the time you don’t need to open it; just talk to the Agent, and all Agents share this one Computer. He also conducted an experiment: he installed clients on both his MacBook and Mac mini at home, and the cloud computer could recognize the existence of these two devices, allowing him to issue commands on one that could access the other’s client. Thus, the cloud virtual machine coordinates everything, while the two local machines participate in execution, with all data interconnected, and he doesn’t have to type a single command.

Guo Yu shared two remote capabilities that Codex has supported since February: one is traditional SSH port control, and the other is penetrating the internal network through Codex’s own WebSocket connection, allowing you to control all computing resources from any machine, even a phone. OpenAI hasn’t marketed this as a selling point, but the functionality has always been there. Guo Yu has several bare metal servers in his data center, and now he can even let the Agent handle buying machines: after purchasing, the service provider sends the initial root password to Gmail, the Agent reads the password, logs in, installs its own key pair, deletes the password, and then everything is maintained by it.

But when he did this for the first time, he thought about another issue:

"If OpenAI is attacked or its connection is hijacked, it could become the world’s largest botnet. Because you can operate tens of thousands or even millions of different computing resources through one Agent." [29:55]

When everyone manages their computers through Agents, the centralized Agent channel itself will become a huge security single point. This is a dark side that cannot be ignored in the vision of Agent OS.

07. The Five Levels of Intelligence

Not everyone needs the strongest model, but rather the most suitable one. Intelligence is being layered by scenario, from personal convenience, personal productivity, enterprise collaboration, entertainment generation, to scientific exploration.

Indigo brought the topic back to user interfaces: the interaction layer is transforming into LUI (Language User Interface), where everyone lets the model generate the interface they need. However, there is a gap here; those who have worked in product and technology know how to describe requirements, but ordinary people may not.

Guo Yu believes that requirements are layered and cited Meta’s recently released Muse as an example (currently only available to US users). It aims to address simple needs of ordinary people: automatically calling to book restaurants, booking things while traveling in unfamiliar countries, managing various information. It doesn’t need a separate app but is integrated into existing Meta communication channels like Instagram, WhatsApp, and Messenger, connecting all communications around the Agent. In the past, you had to use a bunch of software to piece together a workflow; now, you don’t need a workflow; just state your temporary needs, and it continuously processes until the goal is achieved.

Guo Yu has a deep understanding of this. Earlier this year, in March and April, he developed a product for making phone calls on behalf of others, which was used by many Portuguese and Spanish-speaking users after its release (Indigo mentioned he had used it and recommended it to others); later, he created another product aimed at integrating phone calls, emails, and various communications, managed automatically by an Agent. However, by nearly October, super apps had already integrated these basic functions. This once again confirms the judgment made in the previous section: the distribution of specialized software is becoming unnecessary. Of course, more complex creations still need to revolve around SOTA models like Claude Code and Codex.

Following this line of thought, Indigo presented his five-layer framework, starting with a judgment: people do not need the strongest model; they may only need the most suitable model for the right scenario.

The first layer is personal convenience, like Muse, which integrates daily connections and trivial tasks.

The second layer is personal productivity, involving coding and creation, using models like Claude and Codex that have stronger interfaces.

The third layer is enterprise productivity collaboration, which is more specialized than personal use and requires Agents to integrate into enterprise workflows, such as teaching models to directly use Mac mini or Linux, connecting systems within the enterprise, which demands higher requirements from the models.

The fourth layer is entertainment generation, which Indigo believes is Byte's strong suit: Seedance paired with the Hongguo short drama platform, where AI generates content consumed by humans, has already formed an internal cycle; BytePlus has also developed tools for overseas creators, packaging video generation models for sale along with a complete set of creative services.

Mini-games and interactive games may no longer require human involvement, as brands and professional creators generate content distributed through social media information streams; with the proliferation of head-mounted devices, companies like Byte and Meta, which are betting on the metaverse, will have an entrance advantage, "the previous metaverse was just born too early."

The fifth layer is scientific exploration, which requires the top-tier models, and such models may not be disclosed by companies but rather signed under annual framework agreements with pharmaceutical companies and research institutions for profit sharing; defense contracts likely follow a similar format.

On this basis, Guo Yu added two points.

First, enterprise adoption remains awkward. Microsoft has just released a new form of work for enterprises, lowering the hierarchy of Office beneath the Agent, allowing the Agent to operate Office, effectively turning Copilot into an operating system (the brand name is still retained). Microsoft aims to accelerate the implementation of Agent-centric automated workflows in enterprises, but companies remain hesitant: they know they can cut jobs to save money but do not know how to rebuild the entire information and cash flow around Agents. One Person Companies and small teams will naturally embrace it, while large enterprises (including Japanese companies) are filled with concerns, as no one can take responsibility for this matter; only the CEO and the board can make decisions, and middle management sees opportunities but dares not act. They even hesitate to use Chinese models like DeepSeek, fearing internal information might be distilled and leaked to other companies.

Second, he referred to a layer of on-site AI: the cheapest and fastest models used in highly automated scenarios, such as managing farms and factories. These scenarios must rely on edge computing because the amount of information a camera processes in a day is too large. For example, a ranch needs to identify which cow has escaped the pen; it is impossible to send all images to OpenAI, so extremely cheap models must operate locally 24 hours a day.

Putting all this together, Guo Yu provided an investment perspective:

"At this moment in October 2026, the impact of AI on human society, organizational structures, and business is still in a very, very early stage. Although we are accelerating, what we can achieve may only be 2% or 3% completed." [39:30]

08, 2% Pro Users and the Commercialization of Intelligence

AI is a blank slate that amplifies those with ideas; for the mass market, AI-generated content will lead to complete competition, and intelligence itself will eventually become as cheap as milk.

Indigo observed that the distribution of AI users is still in its early stages: the first to use it are Pro Users, those who are active, love learning, and love research. The reason is that AI does not present buttons in front of you like games or traditional software; it is a blank slate, and you must have an idea to activate it. Most people go to work, receive tasks from their leaders, finish their work, and go home to watch TV, without ideas or thoughts on how to make requests.

Thus, AI presents a clear dichotomy: for Pro Users, it is a tenfold amplifier, allowing one person to manage over thirty applications and hundreds of Agent Servers, equivalent to a company of fifty or sixty people in the past; however, based on various surveys of Gemini and ChatGPT, most users still regard them as more convenient search engines. By demographic distribution, perhaps only 2% to 3% of people can become Pro Users, while the rest need another type of AI to fill their time, which is precisely what Byte is doing.

Guo Yu holds a reserved attitude towards the commercial prospects of this path. Consumer entertainment has always been at the core of Byte, relying on capturing attention and monetizing through advertisements. However, the profitability of products like AI short dramas remains uncertain. Even if a closed loop is formed, where you pay Seedance to generate a video and then upload it to Hongguo for others to pay to watch, economically it will become a completely competitive market: there will always be someone using cheaper tokens, spending more time conducting more experiments, driving prices down until the selling price is only slightly higher than the generation cost, deducting computing power and electricity costs, leading to profits so thin they become meaningless.

"If a market is a completely competitive market, its final pricing will have almost no difference from the cost." [42:30]

Indigo took over the conversation: this is what many people in Silicon Valley currently believe, that intelligence will become public goods and be thoroughly commercialized. Guo Yu used a very apt analogy:

"When we buy milk at the supermarket, it is cheap because there are many centralized factories producing milk. If we view software as milk, in the future, any software or anything that helps us achieve our goals will become as cheap as milk." [42:50]

Thus, Guo Yu's focus is shifting: from creating software to thinking about the purpose that software is meant to achieve.

09, Confusion and Waiting for the Environment to Mature

In the new paradigm, rather than rushing to apply old thinking to software, find users, and distribute, it is better to spend time observing like early Byte did, waiting for the environment to truly mature.

Guo Yu admitted that he is currently somewhat confused. Recently, he talked for half an hour with You Yuxi in Singapore, who also felt a bit fatigued after selling his startup to Cloudflare, pondering what to do next. Besides returning to personal life and family, Guo Yu feels the need to think deeper, such as observing for three to six months before deciding where to focus his most important energy.

This reminded him of Byte's own history. Byte was founded in 2012 and spent two years validating the first idea: whether the recommendation system could work on text and images. In March 2015, at the annual meeting in Okinawa, the most important topic discussed was that the entire environment finally made people willing to use mobile data to watch videos, leading video recommendations into a mature phase, worth trying.

"If we do not enter a mature environment, where users do not know how to use our products, and we still follow the entrepreneurial thinking of over a decade ago to make software, reach users, and distribute software to them, is that itself not meaningless?" [44:45]

The conversation then shifted to the next generation. Guo Yu expressed envy for the children of today, who are born in this era without burdens and do not think, "I want to make something to sell to someone." Indigo also mentioned his child who just started college: without thoughts of making money, they enjoy their hobbies, previously interested in geography, researching and creating things daily, feeling great in the community on Reddit. With the pressure of life disappearing, the next generation may follow their interests and communities more. Guo Yu added that Japan entered such a social state after the bubble economy, and AI might bring the next generation worldwide to this point; if what Elon Musk said is true, that in ten to fifteen years people will no longer need money and robots will do all the work, that state could very well become a reality.

10, Everything is Media

In an era of AI content proliferation and extremely low signal-to-noise ratios, real people themselves become the most scarce differentiation; what we deliver is no longer software, but media.

Indigo shared his thoughts. This year, he has seen too many unreliable small companies in Silicon Valley, producing only the edges of models; as Guo Yu said, software does not need to exist, and these functions have no value. However, he believes one thing will become increasingly important: the connection between people.

"The emotional connection between people will become stronger as AI intelligence develops. I really want to see a real person here sharing their thoughts; that would be very interesting." [46:50]

He set a small goal for himself: to productize himself, or to productize his attention. His extensive thoughts will be automatically organized by AI Agents into readable and shareable content, and community members even hope he will directly open an API for their Agents to read what his Agent has written, completing another information transfer. What he wants to deliver is to hand over his attention to others, forming connections of thought collisions, just like this program: two people concentrate their attention and then release it to the audience.

"This is not software; this is media. Software is media; everything is media. Every time you produce a Vibe Coding product, that action and result is part of media. Users take it, digest it, and if satisfied, they want to see you post again tomorrow, just like watching a short video." [48:10]

Such content will form a community around a certain theme, such as the US stock market and primary market, where people are still willing to communicate with each other, and KOLs will still emerge.

Guo Yu echoed this judgment using the example of Zara Zhang. She is not from a programming background but has created many Keynote-related products that have become popular online. She once tweeted that to gain more traffic for a startup's product demo, the only thing to do is to have a real person introduce the product. Guo Yu explained that as AI Slop increases, anyone can use AI to create what they want, leading to extremely low signal-to-noise ratios; in an era of extreme content richness, appearing as a real person is the biggest differentiation, provided you are good enough and do not speak carelessly.

Guo Yu then shared a classic case from within ByteDance, revealing the mechanisms behind this differentiation. Douyin often sees an unknown individual suddenly pushed to the center of online traffic, usually because they have inadvertently hacked into the part of the recommendation algorithm that resonates most with human nature. ByteDance has a dedicated team to combat such behavior. One classic trick is to intentionally misspell words in video subtitles. Initially, Guo Yu thought this was to evade censorship, but later discovered that adding typos significantly increased the completion rate: viewers would rewind to check if the word was indeed misspelled. This trick was quickly identified and blocked, but engineers can never predict how creators will hack the system next, because although they created the algorithm, they are merely tuning engineers and cannot intervene in its internal logic.

"Whether we are distributing AI-generated content or real human content, we are essentially hacking into the flow of this mental model. Currently, there is more AI content and less human content; if you excel, the experimental results in the early traffic pool will be better, and you will be placed into a larger traffic pool." [51:40]

Thus, Guo Yu returned to his core question: What exactly are we distributing? Is it software, or the stories told through software, the personal brands established, or the business goals achieved through personal brands? The previous sequence from A to B to C to D may now be completely reversed.

"If we can achieve this goal without software, perhaps we don't need the software itself. If all our goals can be achieved through intelligence, then having intelligence itself is enough." [52:35]

He used the entertainment industry to illustrate: For most people, the goal is simply to pass the time, whether watching short videos, AI short dramas, or scrolling through TikTok; ultimately, it is all about consuming time, and the form of the product in between is not important. Douyin has taken a significant share from Tencent because playing games prevents one from watching short videos, and vice versa for long videos. All entertainment products may appear different in form, but the competition for user time is direct. Broadening this idea, if we consider work and life goals as the goals themselves, then the form of software, whether To B software, factory software, or personal apps, may not matter at all.

11. Industrial Revolution or Renaissance?

This change is akin to the Jenny spinning machine, transforming the means of achieving goals; when basic needs are met and time is freed up, the lifestyle of Japan's Heisei generation may foreshadow the next step for all humanity.

Guo Yu compared this change to the Industrial Revolution, like the Jenny spinning machine of the 18th century: the ultimate goal is to produce clothing, and whether it is done by hand or machine is irrelevant, as long as the goal is achieved. Indigo pointed out that the level of human goals is rising: previously, the goal was to write software; now, one only needs to make requests for AI to create the interface, or even have no interface at all, simply moving things from A to B. So, in an era of highly developed agents, what will human goal orientation and meaning become?

Guo Yu shared a story from when he had just "retired" to Japan. He often drove out for fun, and one day he drove to Chiba's Kujukuri Beach, where he saw many people surfing on a weekday. Later, he learned that a considerable number of young people in Japan no longer engage in formal work but instead do arubaito (part-time jobs). To them, part-time work is liberating, and the income difference from full-time employees is minimal, mainly due to Japan's small wealth gap, where income is relatively uniform across jobs. Thus, they work three days a week and spend the remaining four days surfing or doing whatever they want.

"When your material desires are extremely low, or all your needs are met by society, you can freely pursue anything you truly want to do. These pursuits may be your interests or may evolve into your career due to long-standing interests." [55:50]

He cited the Olympics as an example: some athletes in events like skateboarding are not traditional professional athletes but have accumulated years of interest and experience, choosing to participate in the Olympics. This is not like F1 drivers who require massive investments for returns; rather, it is a natural outcome driven by interest, reflecting an important aspect of the so-called "Heisei otaku" society after many years of development.

Of course, Guo Yu also acknowledged that for the general public, most people will still think about how to pass the time. However, among the masses, there will always be a sufficiently small proportion of people who significantly enhance productivity with the help of AI, which will be far greater in absolute numbers than in the traditional education era, and more equitable. In the past, children born in rural areas could not receive standard university education and could not pursue software development; now, as long as they are willing to change themselves, they can obtain what they desire.

He borrowed a philosophical concept often discussed within ByteDance: If you view Douyin as an information cocoon, immersing yourself in what you want to see, you will undoubtedly become unproductive; but if you see it as a window to a larger world and actively train your recommendation algorithm, you can learn anything you wish. He recalled a child from a remote mountain village in the West who loved modeling, making clothes from plastic and rags, and walking the runway in the fields. By continuously posting videos, he eventually made it to Milan to model.

"In traditional times, no matter how you spread your ideas, you would only be seen as a lunatic by your peers and fellow villagers. But with such tools, you can use them to step onto a larger stage. AI is the same." [58:00]

Guo Yu believes that our generation's childhood belongs to the last century, inevitably leading to fixed mindsets; the next generation will not carry these burdens. Proportionally, the public will still consume entertainment, but the changes brought by AI are too significant. Even a small portion of people will create explosive productivity and creativity, "we will see many things we never even thought of before."

Indigo prefers to call it a Renaissance. Just like in the 14th and 15th centuries, productivity suddenly surged, and nobles spent lavishly to support guests and artists, leading to a flourishing of science and art. Japan's Heisei generation has already experienced a similar state, but the cause was economic models rather than AI; now, the next generation worldwide will have more time due to the productivity boost brought by AI.

This is a double-edged sword: more time may mean fewer job opportunities. But the poverty baseline is also continuously rising, "even if you are at your worst now, you are still better off than kings 300 years ago; you have all kinds of electronic devices." Indigo believes that the concept of working has existed for over a hundred years since the Industrial Revolution and should disappear after the intelligent revolution. People will differentiate: some will be content with little, living off short dramas and games as long as they do not harm society; while others, those who think differently, will seize this AI Renaissance.

"We now have tools of infinite possibilities. As long as you have infinitely possible ideas, you can do infinitely possible things." [61:55]

He envisions that in a year or two, as long as you have enough initiative and unique human creativity in your mind, combined with AI, you will be able to produce a large number of works, known as artifacts, or entirely new media forms; with these works, those who resonate with you will quickly gather into a community and participate in creation together. The social structure may shift from "entering a company to work" to "Pro Users creating media, with many building communities around the media, forming a cycle of small worlds." Therefore, social media remains important.

12. Advice for Young People: Don't Hold on to Falling Stocks

Do not treat work as a golden rice bowl; short-term anxiety is an escape from a long-term decline.

Finally, Indigo asked Guo Yu, who has "been acquired after starting a business, retired early, and encountered the AI revolution," to give some advice to young people.

Guo Yu first described two types of young people. One type is very anxious, fearing they will soon be unemployed or unable to find work at all, as AI-driven cost reductions and efficiency improvements have led to fewer positions in companies, leaving them feeling uncertain about the future. The other type consists of kids who have not yet encountered this issue; they are full of interest in AI and can now create toys and games that were previously unattainable. These two generations are precisely at the boundary, marking the most emblematic impact of AI on real life. Looking further ahead, for example, when today's two- or three-year-olds grow up, AI will be as familiar to them as WeChat is to the elderly and QQ is to them.

For those who are being forced to find work, his most important advice is: do not treat work as the core goal of life, do not view work as a necessity. Without work, the choices you face are infinite, which may amplify anxiety in the short term because you do not know what you want to do.

"But the reason you chose not to work is actually an escape from a long-term decline." [65:00]

He made an analogy with investments: when a stock is trapped, what you need to decide is not the specific price at which to sell or how much you will lose, but two things: first, is there a better opportunity? Second, is it on a very long downhill slope? If work itself is already on a long downhill slope, holding onto it is the biggest risk.

He shared a story about his father. After the reform and opening up, many people working in state-owned enterprises wanted to go south to Shenzhen or Guangzhou to "go into business," but they feared a risk that today's people find hard to understand: the loss of seniority. So when his father went south, he spent hundreds of yuan each month to apply for a leave of absence, worried that if he failed, he would have to return. Years later, that mine closed down. Friends from Northeast China felt this even more deeply, as there were too many laid-off workers.

So the first step is to change your mindset. Do not treat work as the core goal; it may cause short-term anxiety, but in the long run, this choice is undoubtedly correct. Of course, this also depends on the nature of the work: if it involves a dull job with a secure position, or if you are pursuing further studies for it, he believes it is unnecessary; but if you have the ability to join a cutting-edge AI lab and become one of those creating all this, that job is worth pursuing.

As for how to explore the world and earn the desired income after not working, Guo Yu provided a very specific suggestion: go south and see how people in Shenzhen face all this. When he was in university, he had two roommates from Chaoshan who, back in 2010 and 2011, never thought about graduating to work because their family belief was "better to be a small boss than to work for someone else." They themselves could not explain where this belief came from, but it has been passed down through generations and has become a deeply ingrained thought.

"They believe that working for others leads to nowhere. I think this idea is very applicable in the AI era: be your own boss, be your own OPC. If you don't know how to be your own boss, go see Shenzhen." [68:20]

Indigo added that the Bay Area and San Francisco in the U.S. are also such places. He particularly agrees with the analogy of stocks: holding onto a sinking stock, the real cost is the opportunity cost. Before giving up, first choose a direction that can rise, then adjust accordingly; "the most important thing in investing is not to lose opportunity cost."

Extended Thoughts

Looking at this conversation as a whole, one can find a recurring underlying theme: black boxes, attention, and purpose.

In 2015, Byte's engineers were already coexisting with a system they couldn't fully explain, relying on experimentation to approach results; ten years later, everyone using large models finds themselves in the same position. The difference is that back then, only recommendation engineers needed this mindset, whereas now, everyone working with Agents does. The divergence between Guo Yu and Indigo regarding "whether to make plans" essentially reflects two attitudes towards a black box: one is to maintain understanding and control, while the other is to trust the process at the goal level. It can be anticipated that as Harness continues to evolve, the latter will become increasingly feasible, but the "judgment" represented by the former will not lose its value; rather, it will become the key differentiator between Pro Users and ordinary users.

The second line is attention. The commercial essence of Byte is to capture attention; Indigo's "Everything is Media" is about packaging and delivering one's attention; the typo hack that Guo Yu mentioned is a microcosm of the attention battle. As the marginal costs of software and content approach zero, the only scarce resources left are the attention of real people and real individuals themselves. This may imply that in the future, "product managers" and "content creators" will gradually converge into the same role.

The third line is purpose. From the Jenny spinning machine to First Class Harness, the means are constantly being replaced, while people are being pushed to move up to the goal level. On a personal level, this presents opportunities, but on a societal level, it poses real challenges: when the majority of people's goal is to "kill time," and only 2% to 3% can use AI as an amplifier, a leisurely life reminiscent of the Heisei era requires a premise of a smaller wealth gap and a sound social security system. The Japanese model holds because "the income from any job is roughly the same." Whether the AI era can replicate this premise, rather than further widening the gap, may be the key issue in whether the "Renaissance" can truly benefit the majority.

Key Takeaways

AI has not made engineers' lives easier; it has shifted the decision-making burden of an entire team onto one person. The real bottleneck has moved from writing code to asking questions and making judgments, and the human brain's rhythm cannot keep pace with AI's rhythm. Therefore, finding a sustainable rhythm for collaboration with AI is as important as learning to use the tools.

Recommendation engines make content personalized for each user, while Agents make software personalized for each user. When software can be generated instantly and discarded after use, the entire entrepreneurial logic of "creating software and then distributing it" needs to be re-examined. First, clarify the purpose that the software is meant to achieve; if the goal can be reached without creating software, then there is no need to build it.

Intelligence will stratify and commoditize. Not every scenario requires the strongest model, and in a fully competitive content market, prices will ultimately approach computational costs. Intelligence will become as cheap as milk, while the scarce resources will be intention, judgment, and real people.

In an era flooded with AI Slop, real people on screen, genuine thoughts, and connections between individuals become the strongest differentiators. Every piece of work you deliver is essentially media, and communities will grow around that media.

For young people, the most important thing is not to find a stable job, but to judge whether they are standing on a long downhill snow slope. Short-term anxiety is the price paid to avoid a long-term slide; if you don't know how to be your own boss, go to Shenzhen or the Bay Area and see how those who never intend to work for others live.

Note: This article is based on program subtitles, with quotes organized in spoken form and lightly polished.

This content is provided for general informational purposes only and doesn't constitute financial, investment, legal, or tax advice. Any events, rewards, online promotions, or related information mentioned herein should not be considered a recommendation, solicitation, or invitation to purchase, sell, trade, or otherwise deal in any crypto assets. Crypto assets are highly volatile and may result in loss. The availability of WEEX services, products, and related events may vary by region. You are responsible for ensuring that your participation is in accordance with applicable local laws and regulations.