Summary

Once AI drives product development costs down, attention—not code—becomes the truly scarce resource. Zara’s view is that AI coding will multiply the number of products, make features easier to copy, and push competitive moats upstream toward go-to-market, distribution, brand, and user community; “everything is ready except a CTO” is becoming a world where the CMO matters more. For investors, the question is not who can ship a feature fastest, but who can acquire, convert, and retain attention at lower cost, continuously.
As AI floods the market with content, “human presence” becomes a scarce brand signal—and founders have to become the CMO themselves. The episode uses OpenAI as an example: Koji believes users often learn about new features from Sam Altman’s Twitter rather than the official account, a point Zara agrees with while stressing that personal and product brands will increasingly become inseparable. Lovable has extended employee social to the entire company, with engineers explaining features and gathering feedback themselves. Zara’s bottom line is blunt: “by default, nobody cares”; building a product and failing to talk about it is the greater risk.
AI products are moving from feature competition to consumer-style competition over identity, aesthetics, and vibe. Cursor Cafe and Anthropic’s thinking cap did not emphasize parameters; they made users feel that “people who use this product are a certain type of person.” When ordinary users cannot distinguish the functional differences among 10 AI coding tools, website design, founder charisma, community, and shareable experiences will determine the choice. Koji believes consumer-marketing tactics still have roughly a 3-to-6-month arbitrage window in tech, but the stimulus will quickly lose its effect.
The “best era for liberal-arts graduates” does not mean technology is no longer important; it means technical curiosity, user insight, and distribution matter just as much. Zara flips the old slogan into “code is cheap, show me the talk”: natural-language tools let people without a CS background build prototypes, while the genuinely difficult questions are what to build, why to build it, and how to explain its value. She preserves one important paradox—“idea is expensive again,” because good ideas come from long-term observation of real situations; but “idea is cheap,” because the cost of execution and imitation has fallen. In the end, advantage comes from turning the execution–acquisition–feedback–iteration flywheel faster.
Positioning is not something you think up; it is something you work out through high-frequency publishing and real feedback. Zara’s Xiaohongshu following grew from roughly 20K at the start of the year to about 180K, but the more important data is nearly 500 posts over more than 2 years, including 6 straight months of daily posting during the cold start. Feedback showed her that “how liberal-arts graduates enter AI” and first-hand resource recommendations were her distinctive supply. Her information-filtering rule is “follow builders not influencers”: builders talk about practice, while influencers often only tell you what model was released next, without answering “so what.”
Content and products are both prediction systems for human behavior; the advantage comes from the number of feedback loops, not one-off inspiration. Publishing content means predicting who will watch and who will share; shipping a feature means predicting who will use it. Updating your sense with data and feedback after every cycle is what produces taste, product intuition, and distribution. Long Cut is the product of exactly this method: instead of crushing a long video into a summary that loses its stories and memorable lines, it marks the highlights on the timeline, preserves the original video, and provides contextual explanations and cleaned-up notes.
The long-term value of content is not advertising revenue, but building in advance the distribution, reputation, and network that future businesses will need. Zara’s Xiaohongshu account and side projects were never built for short-term monetization. She sees “every follower” as future marketing cost saved, and compares free sharing to deposits into a karma bank. Waiting until after starting a company to look for investors, users, talent, and partners is often already “too late.” Personal content is therefore a long-term compounding asset, not an isolated creator-revenue business.
AI ultimately amplifies agency, not intelligence: high-agency people gain superpowers, while people who do not act may lose their existing position faster. Zara went from buying equipment and delaying for a year, to using only her phone to answer 3 frequently asked questions, then learning in public, building products, and changing jobs—all by lowering the threshold to “a 5% change.” Her conclusion is direct: “Ask yourself: who is stopping you? No one is stopping you.” Sustainable action is not the same as grinding; the real goal is to find growth that “feels like play to you but feels like work to others.”

Deep dive

1. Action Before Understanding: 2025 Was the Year She Went from Observer to Builder

Zara defines 2025 as “a year of action.” She had realized the year before that AI was a once-in-a-generation opportunity she had to seize, but mostly remained in the mode of watching videos, following news, and studying. Eventually she admitted that “watching videos and studying are actually a form of laziness”—often a way of avoiding action because she was afraid to act.
Her correction was not to stop taking in information, but to change the order: make content, build products, try vibe coding, and then return to papers, interviews, and tutorials with concrete questions. Ideas she had failed to understand before suddenly became meaningful. “The best way to learn is learn by doing.”
Over the year, she broke through in both Chinese- and English-language content, moved into a product role, launched her first side project, and collected user feedback. She also turned more ideas into working products. Koji joked that listeners could shut off the podcast and go take action immediately, but the shared conclusion was that listening is fine—as long as it does not replace doing.

2. AI Lowers Development Costs—and Pushes Customer Acquisition Higher Up the Difficulty Curve

Zara is less interested in the next model leaderboard than in the structural change in go-to-market: as design and coding costs fall, startups and AI products are appearing everywhere, and users are being bombarded with new launches every day. “There’s no way to use them all.”
She therefore reorders the scarcity hierarchy. Code used to be expensive, making it the scarce resource; now products can be built faster and more cheaply, while attention has become “an even scarcer resource.” Getting users to choose your product over 10 similar alternatives is harder than building the product itself.
Marketing, distribution, brand, and user community can no longer be treated as remedial work after the product is finished. Koji summarizes the old paradigm as “everything is ready except a CTO.” Zara’s response is that today the CMO, storytelling, and the ability to create connections are becoming the real moat.

3. “Human Presence” Is the Scarce Signal in an Age of AI Content Surplus

Zara is not opposed to using AI to write content. What she opposes is content with no real experience or distinctive point of view—and creators who do not even bother to remove the obvious AI voice. In a sea of homogenized generated content, only work that makes people feel there is a specific person behind it can stand out.
She summarizes the future of marketing in one phrase: “human presence.” A voice with imperfections, language that is not completely polished, and a creator willing to appear on camera can all help audiences form a connection. Podcasts are especially well suited to this because listeners can directly sense the two people’s voices and live interaction. “You can tell immediately it isn’t AI.”
Koji uses OpenAI as an example, saying he can barely remember when the official account last posted something influential. Users often learn about new features from Sam Altman’s personal Twitter. Zara agrees and goes further: personal brands and product brands will become increasingly inseparable.

4. An Entire Company of CMOs Can Turn the Organization into a Distribution Network

Zara’s idea of founder-led marketing goes beyond the founder posting on X. It means “everyone is the CMO”: when an engineer or product manager builds a feature, they explain it, promote it, and read the comments directly. Distribution and product iteration become one connected loop.
Lovable operates both founder social and employee social; whenever it launches a feature, the entire company reposts it to build momentum. Koji compares this with the old promotions at Jumei, when employees would change their WeChat profile pictures to the same “301” or “901.” In essence, it was an early version of company-wide social.
Arc’s weekly changelog did more than list 10 updates. It used doodles, graphics, and employee avatars to show who built each feature and explain it clearly. Users could see the builders behind the product, while employees gained ownership—“I made this.” Even after the team shifted its attention to Dia, Arc’s updates retained that personality.
Smaller companies may actually have an advantage in building in public. Large companies have complicated PR policies; small teams can let employees speak directly on the basis of trust. Zara’s warning is simple: “by default, nobody cares.” If you build a product but do not talk about it, the work effectively disappears from the market.

5. Positioning Is Defined by the Audience After Hundreds of Posts

Zara’s Xiaohongshu following grew from roughly 20K to about 180K in 2025, but she believes the more relevant metrics are nearly 500 posts in total and 6 months of daily posting during the cold start, “without missing a single day.” The result came from a sufficiently large sample of practice, not from choosing the perfect niche at the outset.
She did not have her positioning figured out early on either. Later she summed it up this way: “Positioning isn’t something you think up; it’s something you work out.” After dozens or 100 posts, the content that gets watched and the content that gets ignored train your content sense. Your eventual positioning is not how you describe yourself, but “what the audience thinks your positioning is.”
The actual positioning shaped by feedback was to share AI trends, personal learning, and growth from the perspective of someone without a technical background. It was not exactly what she had initially imagined, but it filled a gap outside the dominant technical narrative.

6. A Nontechnical Perspective and First-Hand Builder Content Are Her Differentiated Supply

The breakout topic that surprised Zara most was “how liberal-arts graduates enter AI and learn AI.” Industry discussion is usually led by researchers, engineers, and technical founders, but the vast majority of people in the world do not have a technical background. She eventually realized that this audience needed someone who could translate the wave without pretending to be a technical authority.
Another category that generated strong feedback was simple resource recommendation. She initially thought she had to write polished, insightful summaries. Then she found that listing screenshots from 5 long YouTube interviews, explaining who was speaking and why each was worth watching, already met a large demand for first-hand information—and carried more “human presence.”
What determines the value of a recommendation is the filter, not the format of “5 videos.” Zara’s rule is “follow builders not influencers.” Watching a high-density long-form interview from beginning to end is like having a free video meeting with a leading Silicon Valley practitioner.
Builders speak from having done the work themselves. Influencers often repeat that “OpenAI released another model, Google released another model,” without answering “so what”; sponsorships can also compromise the authenticity of recommendations. Koji adds that the show found a Chinese builder with the world’s highest usage through the third-party token leaderboard for Claude Code. That practical episode became the year’s most commented-on, bookmarked, and reposted episode.

7. The Most Shareable Vibe-Coding Prototypes Came from the Creator’s Own Frictions

Zara realized that people freeze in front of a camera because one-way output runs against the most natural state of dialogue. She therefore built a video-recording tool in which Gemini’s real-time voice capability plays the role of a podcast host, asking follow-up questions based on the user’s answers and turning camera-facing monologue into two-way communication.
Because the user does not speak while the AI is asking a question, the tool also added one-click “remove silence,” automatically cutting the questions and pauses and exporting a clean video of the user speaking continuously. The feature was not invented in the abstract; it followed directly from the real recording workflow.
She also turned the Xiaohongshu livestream experience into an “AI hype squad.” The interface simulates a livestream platform, while a group of AI viewers asks questions, offers praise, and provides positive feedback in real time. Koji agrees with the underlying insight: humans are social animals, and without a response they freeze; healthy feedback can materially lower the barrier to expression.

8. Learning in Public Can Rewrite a Person’s Identity

Zara believes she became AI-native not simply by learning a certain amount of technology, but because she kept sharing her practice publicly until the outside world first began to see her as AI-native. That perception then fed back into how she defined herself.
She sees this as an identity-rebuilding hack: if you want to escape an old label, start by publicly creating new evidence of behavior. Koji relates it to his own experience overcoming motion sickness—he repeatedly told himself, “I am someone who doesn’t get carsick,” and eventually changed both his physical and psychological response.
In her view, the technical/nontechnical binary is already outdated. She prefers “technically curious”: whether someone follows new models and APIs and is willing to try them has no necessary relationship with whether they studied CS at university. Some excellent engineers do not care about the latest AI, while nontechnical people may be far more enthusiastic.
She stresses that technical curiosity is a learnable skill. Every new technology wave forces everyone to learn again; even a degree in AI cannot make anyone rely indefinitely on old knowledge. Natural-language tools also let nonprogrammers build prototypes. Zara is careful in saying that liberal-arts graduates “should” have stronger insight into people, users, communication, and distribution—and may therefore outperform some programmers on ideas and distribution.

9. Ideas Are Becoming Both More Expensive and Cheaper; the Real Asset Is Loop Speed

The “expensive” side of “idea is expensive again” comes from distinctive, long-term understanding of real situations. Zara’s two-way recording product was easier to conceive only after she had livestreamed, published hundreds of videos, and experienced the awkwardness of being on camera herself. A few casual user interviews usually cannot produce that depth.
The “cheap” side of “idea is cheap” comes from the simultaneous decline in development and imitation costs. An idea by itself is no longer defensible. You have to build it, attract attention, collect feedback, and generate the next idea. Competitive strength comes from how quickly and how many times you can turn the “execution–attention–feedback–iteration” loop.
Zara once had imposter syndrome because of her nontechnical background in internet and early AI discussions. Later she inverted “talk is cheap, show me the code” into “code is cheap, show me the talk.” She uses the evolution of Chinese characters as an analogy: writing characters on turtle shells was once an elite skill for a small number of people, while writing and typing later became basic capabilities. Programming may undergo a similar migration.
Technology creates value only when it becomes a product, reaches users, and is genuinely used. This is where liberal-arts graduates can answer “what to build” and turn insight into a prototype; they can also answer “how to sell,” using storytelling, community, and communication to start the flywheel. That is why Silicon Valley is beginning to hire heads of storytelling.

10. Offline Pop-Ups Are Packaging AI Products as an Identity

Cursor once took over an old San Francisco coffee shop and turned it into Cafe Cursor, where users could code for a day while the core team answered questions and listened to requests. It was simultaneously a user event, a product-research venue, and a place for builders to connect with the community.
Anthropic took over a coffee shop in New York and handed out thinking caps with only the word “thinking” on them. In the face of AI slop, it did not advertise how powerful Claude was; it implied that “people who use Claude think independently.” People lined up for hours, some even flying to New York specifically to get a hat. What was being spread was the type of person users wanted to become.
Zara says the highest form of marketing is not promoting the product, but promoting the user: “Using your product is a symbol of identity.” When a nontechnical user builds a first website with vibe coding and immediately posts it on Xiaohongshu, it is because “I can code too; I’m also a Cursor user” feels cool enough to share.
Koji believes these campaigns, though familiar in consumer products, can still become global hits when transplanted into tech. Zara adds that for now—and perhaps for the next 3 or 6 months—these pop-ups may remain effective marketing leverage, but the stimulus will not work forever. Offline experiences also matter because they create face-to-face human connection and let builders see users’ reactions directly.

11. The Best Marketing Is Written into the Product Experience

There are at least 10 AI coding products, and ordinary users struggle to articulate the functional differences or have the patience to compare them one by one. They often choose the product their friends mention most, the one with the better-looking website, the founder whose energy feels right, or the one whose marketing happens to reach them. Differentiation has shifted from features to brand, design, storytelling, and vibe.
Zara uses shampoo, Evian, and VOSS as analogies: consumer products can be highly similar in function, yet buying one still expresses identity. Offline events should likewise turn the experience into shareable material, giving participants reasons to take photos and post them on Twitter or WeChat Moments. A few hundred attendees can become much broader online reach.
Poke builds shareability into onboarding. Users must persuade an AI over text message to admit them into the beta; the AI challenges them, mocks them, and negotiates a price with no fixed value. Koji recalls that it once opened at $996 because he had made the 996 podcast. The privacy implications may be uncomfortable, but screenshots of the negotiation naturally encourage sharing.
When Koji was CEO of Tangdao, he also built a temple on Shanghai’s Yuyuan Road for a cat-belly pillow, where visitors could worship the god of sleep and take amusing photos. The offline activation reached only a few thousand or 20K people; the real amplifier was that every participant wanted to prove they were someone who could “always discover something new.”

12. Repetition Is Not a Lack of Creativity; It Offsets Communication Loss

Zara’s marketing rule is “repeat, repeat, and repeat again.” When creators feel they have talked about something until they are sick of it, the audience may only just be starting to hear it. You may think you transmitted 100 points, while someone else received 50—and many people did not notice at all.
On Xiaohongshu, roughly 99% of people encounter content by chance on the recommendation feed rather than visiting a profile to check whether it is repetitive. The reality is that “people don’t read”: long articles are often reduced to an AI summary, and most videos receive only the first few seconds. You cannot assume the audience will consume the whole thing.
She explains the path to downloading one of her products through touch points. The first time someone sees a product video, they think it looks good but are busy with something else. The second time they bookmark it and forget. The third time a friend recommends it, they still do nothing. Only after the fifth touch do they actually download it. Moving from awareness to action may naturally require 5 reminders.
Quantity therefore matters as much as quality, and the content threshold must be low enough to support daily posting. Even excellent founders or investors with demanding jobs may post 10 tweets a day. When “only 1% of people get 10% of your message,” high-frequency engagement is itself a distribution strategy.

13. Long Cut Refuses to Flatten High-Quality Long-Form Content into a Distorted Summary

Zara’s product Long Cut is the opposite of Short Cut, built around the idea of “take the long cut.” After a user pastes a YouTube long-video link, the model reads the full transcript, finds the most distinctive and valuable passages, and adds titles and highlights along the timeline.
Clicking a title does not show an AI retelling; it jumps directly to the corresponding point in the original video. She insists that “even if you watch selectively, you should still watch the original,” because the first things an ordinary summary loses are often the guest’s small stories, concrete examples, verbal rhythm, and memorable lines.
When a research interview contains unfamiliar terminology, users can select a word, phrase, or full sentence in the transcript and click explain. The AI explains it using the current video’s context, without requiring the user to copy the passage into ChatGPT or Google with no surrounding context.
Selected quotations can also be cleaned with one click to remove filler words, slips of the tongue, and inaccuracies, then combined with personal reflections into notes. All 3 of Long Cut’s functions came from Zara’s own long-standing pain points when watching interviews, rather than from listing features first and searching for demand afterward.

14. Content and Products Are Repeatedly Calibrated Prediction Systems for Human Behavior

Zara gives the same definition to both types of work: they are “prediction systems for human behavior.” Content predicts who will watch, like, repost, or not care; products predict whether a feature will be used and what feedback it will generate.
You cannot find the answer before publishing by sitting at a computer and thinking harder. The data only arrives after publication. Every “publish–feedback–update your understanding–publish again” cycle is like adding a training sample to a model. The more cycles you run, the more reliable your content sense and product sense become. There is no shortcut to being born with user intuition.
Koji summarizes people’s remaining value as taste, distribution, and agency. Zara also cites Naval’s view that code and media are leverage that “does not require anyone else’s permission”: both can compound over time, while large models further lower the barrier to using them.

15. The Key to Cold-Starting Action Is Shrinking the Change to an Immediately Doable 5%

Zara believes many people assume they need permission from parents, teachers, bosses, a degree, a major tech company, or a CS label before they can act. Today they can download an AI coding tool, build a product, record a video, and publish it directly. “Ask yourself: who is stopping you? No one is stopping you.”
She also acknowledges that knowing and doing are different. Before deciding to post regularly on Xiaohongshu, she hesitated for about a year, bought lights and a tripod, and let the equipment gather dust without recording anything. The real starting point was simply giving up polish, picking up her phone, and answering in plain language the questions people around her asked most often.
Her cold-start method is to think back to “the 3 questions people ask you most often.” Experiences that feel obvious to you may be exactly what outsiders find most interesting. She began with questions such as “Why did you leave VC for an internet company?” and “How does a liberal-arts graduate enter the internet industry?”
Koji offers another 5% change: put a friend with whom conversation feels natural behind the camera to ask questions, or publicly commit to a regular posting schedule. At a crossroads, he repeatedly told people that he hosted a weekly podcast. External recognition then forced the team to deliver, and eventually they were publishing twice a week.

16. Learning in Public Connects Input, Output, and Positive Feedback into a Sustainable System

Zara recommends sharing from the posture of a student rather than a teacher: “I’m learning with you; I’m just sharing my notes.” This lowers the psychological barrier of feeling insufficiently expert and avoids the condescending tone of speaking from above, making the content more useful and companionable.
Continuous public output pulls input along with it. When she has nothing to post, she watches a YouTube interview; explaining the ideas in her own language then forces her to digest them. As the outside world sees her as someone who studies hard, she becomes more willing to study every day. Identity and behavior reinforce each other.
As a student, exams provided feedback. In the workplace, part-time learning often has no answer to “what happens after I learn this?” Zara’s stronger formulation is: “If you don’t learn in public, it’s as if you never learned.” Content, a demo, or a small project at least leaves visible output behind.
What she seeks is growth with ease. Posting content, trying models, and vibe coding feel more like entertainment than additional labor. Scrolling short videos may leave her empty, while doing something she likes replenishes her energy. The sustainable state is what Naval describes as: “find what feels like play to you but feels like work to others.”

17. Ease Is Not the Result of Lying Flat; It Comes from Positive Feedback Generated by Action

Zara was anxious while learning AI last year: her background felt mismatched, model capabilities were limited, vibe coding was full of bugs, and many ideas could not be brought to life. The obvious breakthrough this year came because she kept trying until she finally received the feedback that she really could make things work.
She stresses that neither positioning nor ease arrives like a pie falling from the sky. If you post a few videos that nobody watches, try a few products that fail, and then quit, you will never reach the point where feedback appears. You have to persist until feedback arrives; only then can the work change from a burden into an activity you enjoy.
Koji cites Paul Graham’s advice on great work: look for the thing that feels effortless to you but difficult to others. They also note that this area need not be AI or content. Everyone has to “get your hands dirty” to discover their own form of easy growth.

18. The Long Game of Content Is Building the Resources Future Businesses Need in Advance

Zara is explicit that her Xiaohongshu account generates no monetization and that her side projects were not built for profit. She has a full-time job and does not plan to become a full-time creator. She distinguishes the short game from the long game: the former is taking ads immediately after gaining followers; the latter is using distribution to meet founders, investors, talent, customers, and future collaborators.
She treats every new follower as marketing cost saved in the future. You cannot wait until launching a product to think about distribution, or start a company only to discover that you do not know any investors, cannot find test users, and cannot think of anyone to invite onto the team. By then, building the network is already “too late.”
In her mind, free sharing is like karma or merit. People she has helped sometimes bring gifts to their first offline meeting because they already feel they have benefited. She compares influence to money deposited in a bank: “one day you can withdraw it.” The longer it compounds, the more it grows, so there is no need to cash it out early.

19. The AI Products She Recommends All Deliver Low-Friction Positive Feedback

For everyday conversation, Zara recommends Claude. She likes its personality, which differs from other AI: it does not simply flatter the user or assume the user is always right. On personal questions, it feels “like a gentle breeze,” and it also works well for writing or as a therapist-like conversation partner.
For learning, she recommends NotebookLM’s remix concept: convert source material into text, podcasts, video overviews, or PPTs to match different ways of absorbing information. She also uploads her résumé and conversation transcripts, asks 2 AIs to praise her for 20 minutes, or turns messy thoughts into a PPT. She extends the slogan into “understand anything including yourself.”
Beginners in vibe coding can start with Replit Design, which covers everything from idea to generation, deployment, and domain in almost one flow. There is no need to touch code, and the resulting pages are reasonably attractive. She suggests first turning a résumé into a personal website and getting positive feedback from a small result with no database or backend.
Another less mainstream tool is Faces, suited to lightweight websites such as résumés, portfolios, and resource lists, with visual design and interactive motion effects. She has used it to organize AI podcasts, videos, and newsletters. The priority remains helping beginners quickly make something “worth taking a screenshot and sharing.”
Zara also suggests using the prompt “have AI ask me one question at a time” to write 2026 goals now, instead of staring at a blank document and trying to think everything through alone.

20. Having AI Ask Questions Gets You to “Your Own Answer” Faster Than Further Prompt Optimization

Zara reverses the default human–AI relationship. Instead of endlessly asking questions herself, she tells Claude or ChatGPT: “Ask me one question at a time and help me think this through.” The AI follows up on the audience, pain point, and competitive difference; she only has to speak naturally for 10-plus minutes and then ask it to organize the result into a document.
Because all the raw material comes from her own speech, the finished draft is “completely mine.” It only needs light editing and does not read like templated AI writing. She has even completed an entire document while stuck in traffic. She also recommends writing 2026 goals by answering one question at a time.
Another high-value prompt is: “What do you need from me to do this better?” Many models start working directly despite insufficient context, so the result is predictably poor. Asking AI to list the information it needs makes the assistant actively discover what it “doesn’t know it doesn’t know.”
The same interaction works for personal confusion. Ask AI to behave like a therapist and ask one question at a time; users often articulate the solution themselves while answering. Zara’s observation is that people sometimes do not need someone else to give them the answer. They only need to be guided accurately toward the answer they already know.

21. The Builders Worth Following Show Both Product Intuition and Go Direct

At the product level, Zara is most inspired by Granola founder Chris Pedregal. User interviews are not a matter of doing whatever the user says; they train a product manager’s intuition across countless conversations, after which the manager still has to make the judgment. At the application layer, there is also no need to fight the model layer over problems such as context windows that will ultimately be solved below.
Google’s Josh Woodward showed her what an entrepreneurial state can look like inside a large company. He asks Google Labs product managers to spend less time staring at dashboards early on and more time watching whether an aha moment appears in users’ eyes. Before a project reaches Google’s billion-user scale, lived experience may matter more than immature data.
At the PR and GTM level, she recommends Lulu Cheng Meservey’s “Go Direct”: CEOs should speak directly to audiences rather than outsourcing expression to media or PR copy. A founder going on a podcast is not about being extroverted; it is a business strategy. “I’m an introvert” cannot replace a clear choice to hire, acquire customers, and build trust.
Lulu’s positioning model is the intersection of 3 circles: what you want to say, what the target audience wants to hear, and what benefits the business. For founders, “who is more important than how many.” Long podcasts can explain minority viewpoints in depth and then be cut down for distribution. She also points out that publicly supporting an underrated product can activate the audience’s impulse to help an underdog spread. But the “hidden gem” advantage only lasts until the brand is no longer niche.

22. Technical Boundaries Change Daily; Stable Needs Must Be Re-Matched Through Continuous Use

On balancing AI’s capability frontier with user needs, Zara’s breakdown is that human pain points and needs are relatively stable over time, while technical capability changes every day. New opportunities usually do not come from discovering a need nobody has ever had, but from the fact that “something technology couldn’t do before can now be done.”
User interviews alone are therefore insufficient. You also need to track the frontier through Twitter, newsletters, and researcher interviews, and you must personally use new models and APIs. If you can stay ahead, you not only know the current boundary but can infer from research directions what may be unlocked over the next few months, making it easier to connect stable pain points with new capabilities.
She also says this is the era when “everyone is a product manager.” In the past, learning product might have required assembling an R&D team or even raising money. Now you can open Cursor or Google AI Studio and use natural language to build at least a prototype and test a hypothesis quickly.
Zara says “AI taught me product” not because the model poured knowledge into her, but because it lowered the barrier to practice. If you want to learn product, build more, publish more, and watch more feedback. The same applies to content. Capability is formed through repetitions of practice; tutorials cannot substitute for that sense.

23. Agency Determines Whether AI Becomes a Golden Age or the Worst of Times

In the latter part of the episode, the host cites Andrej Karpathy’s judgment that “agency is greater than intelligence,” translating agency into a founder-like spirit of “my fate is in my own hands”: initiating, advancing, and completing projects rather than letting the environment and old labels decide for you.
The host’s action hack is to give a goal a vivid project name, then make a logo, build a website, and publish it publicly. He uses a hypothetical “2026 Purple Year” vegetarian challenge as an example. Ritual turns a vague wish into an entity that demands accountability and creates external feedback for action.
Zara’s final distinction is clear. For high-agency people, AI is the best of times because it lowers the barriers to learning and creation. For others, it may replace old capabilities and make their existing positions unnecessary. Ordinary people should at least learn how to work with AI and ask, “What value can I still create together with AI?” The goal is to let it empower rather than replace them.
The personal biography in the bonus segment explains why she resists a single label. She is the mother of a 1-year-old, a former reporter at The Information, from Northeast China, a birder, a Japanese learner, someone who does not drink coffee or tea, and someone who has enjoyed self-expression since childhood. She has lived in China, the United States, and Singapore, studied psychology at Harvard, and has spent nearly 10 years in China’s venture-capital ecosystem. The Information was the publication where she interned in her junior year: subscription-only, ad-free, and focused mainly on exclusive news.
She moved step by step from journalism to VC to marketing and eventually toward product, finally calling herself a builder. The host summarizes it as “life is an infinite game.” Zara’s response is that what she is ultimately building is her own life.

Fuse CEO Alan Chang: The Revolut Playbook of Speed & Ownership, Why Founders Aren’t Ambitious Enough Building & Scaling the AI Safety Research Community, with Ryan Kidd of MATS