AI Growth Hacking Weekly — EP#59: ByteDance Growth Retrospective, YC AI-Native Organization Overhaul, Small-Town Mom's AI BL Dramas, Ant Group's Three-Generation CTO Dialogue, Naval's Latest Blog Post
The true AI-native organization needs “trust by default” and “egalitarian.”
▪️Editor’s Note
1/ I recently posted this tweet: “I practiced this for half a year: not listening to any AI entrepreneur podcasts that talk big, regardless of how impressive the guest’s resume or how famous the podcast brand. What did I miss? Nothing. I missed nothing.” This garnered widespread resonance.
After reading my tweet, netizen Neo commented that he also often listens to podcasts during flights, has a similar “pitfall avoidance” system, and has become quite experienced at selecting programs. He shared with everyone:
First, select podcasts not hosted by VCs looking to pivot into investment work. Of course, podcasts still require solid content skills; if you can’t get through an episode, the guest might not be a good fit—if this happens twice, you can rule them out entirely.
Next, directly exclude all programs featuring non-unicorn CEOs who take yuan. Those who haven’t hit enough pitfalls basically can’t share anything useful, only visions. Not that visions are without value, just that they don’t resonate with listeners.
Podcasts under 30 minutes can be directly excluded—they’re mostly superficial AI news commentary.
For guests with technical backgrounds, most can provide good industry insights. But if they can’t express themselves using the clear logic of “problem occurs -> how to analyze -> how to solve,” if they start by dumping jargon, you can close the episode.
2/ My new course How I Use AI to Develop Virtual Content Products: Experience and Retrospective is still in early bird pre-sale. If you’re interested, please support it—I only teach practical gems from firsthand experience. (If the link doesn’t open, try copying and pasting this address into your browser: https://zerodaybook.mikecrm.com/LZtJIOB)
OK, here is the main content of this issue, enjoy:
▪️CASE
Total Suppression, No Gaps: How ByteDance Does Growth
via Luan Fan Shu
This podcast systematically discusses how ByteDance approaches growth, with a data-driven science at its core.
It uses long-cycle LTV prediction models to force-test customer acquisition bids, trading space with long-term thinking guided by “beginning with the end in mind.” The goal is to push products to a 15% penetration critical point, triggering a natural growth flywheel.
The growth team serves as a core platform, standardizing advertising, attribution, and red envelope referrals into SDKs that empower all business lines.
Global expansion adopts a “high-culture attacking low-culture” strategy: aggressively recruiting top creators paired with high-intensity user acquisition, pushing retention and DAU beyond competitors within six months.
Risk control (compliance, content review, anti-fraud) is the lifeline that determines business survival in high-budget growth environments.
BTW, my recommended best podcast speed-listening tool is Podwise.
How YC Is Undergoing AI-Native Organizational Transformation: Agent Capabilities Must Be Open to Everyone
via Founder Park
The YC team has always been at the AI frontier. How they internally use AI to transform the organization is itself worth studying.
This podcast is a conversation between Pete and YC CEO Gary Tan, discussing how YC built a company-wide Agent system over the past year-plus. A registry of 350+ tools, visible to all employees. Agent conversations accessible to everyone. Nightly automated learning and improvement. In some capabilities, it has already surpassed what any individual partner could achieve.
Pete’s judgment: A truly AI-native organization needs “Trust by default” and “Egalitarian”—completely opposite to traditional command-and-control logic.
2 Sentences: Using AI to Recreate a TikTok Viral Video with 20M+ Views
via The Second Curve Growth
The hardest part of making TikTok content isn’t editing—it’s selecting topics. The dumbest yet most effective method is recreating viral hits.
The author deconstructed a 24.6 million-view children’s foam bath video to extract the complete formula:
Step 1: Find benchmarks through four channels—For You Page, TikTok Creative Center, competitor accounts, and third-party tools. The standard is views 5x higher than the account average.
Step 2: First have AI deconstruct the video formula across 6 dimensions (opening hook, product appearance, value anchoring, closing CTA, overall rhythm, one-sentence formula), rather than writing the script directly.
Step 3: Send the deconstructed structure to a multimodal model to generate a 100% precise recreation blueprint, generating video segments no longer than 15 seconds each.
Core insight: clear deconstruction is more important than direct generation—only when you can clearly explain why something went viral will the script stay on track.
Post-80s Small-Town Mom, Making 20K yuan Monthly with AI BL Dramas
via Entertainment Capital
In China, a group of veteran fujoshi—2D Cosers, moms—are using AI to batch-produce BL short dramas, exploding simultaneously on Bilibili and TikTok. Charging for tips + member payments, monthly income peaks at 60K yuan.
For the same creators, traditional comic dramas earned “5 yuan per 10,000 views,” while AI BL dramas have multiplied this figure tens of thousands of times. Lower barriers allow ordinary people without professional backgrounds to post daily plot updates and complete IP serialization operations.
Extended Reading:
Miao Ang: From AI Tools to AI Comics and Dramas, How to Build a New Business?: Jelly Design founder Miao Ang shared at Dedao’s New Business School how their team crossed from design into AI, then killed it in the AI comic drama space.
▪️OPINION
Ant Group’s Three-Generation CTO Closed-Door Dialogue: Cycle Navigation, Technical Decisions, and AI-Native Organization
via Huo Taiwen
Ant Group’s three generations of CTOs (Lu Su, Miao Miao, Liang Ge) appeared together for the first time. Core topics: How to maintain organizational sharpness in technical decisions during cycle navigation, and that the essence of AI-Native organizations is not tool upgrades, but architectural reconstruction.
It’s quite long, with relatively scattered content (some answers are quite official). If you’re interested, find a solid block of time and read it slowly.
An AI Founder’s Vanity, Pretension, and Peak of Folly
via 42 Zhang Jing
A podcast that has “blood-washed” the AI scene over the past two weeks. Friends who listened all described it as extremely candid and real.
The guest, Meng Qi, is an AI software entrepreneur with a ByteDance background and an impressive team lineup. After a year of entrepreneurship, they’ve pivoted multiple times.
In the podcast, she retrospected: Why, as AI coding capabilities improve exponentially, are there still extremely few products that actually succeed?
Two structural dilemmas cause this: First, vertical Agents are forced to become agencies, making money through services rather than products; second, doing B2B in China is extremely difficult, while overseas there’s a ceiling for Chinese founders.
She reflected on her early mistake of choosing the wrong Sourcing scenario, misjudging the criticality of data quality. Core judgment: In the future, all ideas are fully exposed, every track has opportunities, but only companies that truly care about experience and can deliver promised value will succeed. Embodied AI and AI hardware are currently hot, making software entrepreneurs even more squeezed.
AI Hardware Going Global: Is It a Loss Unless Gross Margin Exceeds 80%?
via Feifan Research
The previous item mentioned hardware. Actually, the core logic of AI hardware going global isn’t selling hardware—it’s using hardware for customer acquisition and subscription renewal.
Chao Guang and Duan Ran represent two different paths—Chao Guang stands on Silicon Valley streets with a camera letting users try the product, while Duan Ran chooses a lightweight online launch event—but both ultimately move toward the same model: hardware gross margin must exceed 80% to cover hidden costs like tariffs, shipping, and high return rates, then rely on 9.9 USD to over ten USD monthly SaaS subscriptions for real profitability.
Pricing below 150 USD goes viral on TikTok/Twitter; above 150 USD requires offline trial experiences to build trust.
Quick Reads:
The AI Industrial Revolution: In this latest blog post, Naval discusses the manufacturing revolution in the AI era with three frontier founders—Guillermo Rauch (Vercel), Blake Scholl (supersonic aircraft), and Max Hodak (brain-computer interface)—revealing that “building factories oneself rather than assembling finished parts” is becoming the core variable of a new wave of innovation. AI cloud infrastructure, supersonic commercial aircraft, and brain-computer interfaces—three seemingly unrelated fields—are governed by the same logic: vertically integrating manufacturing capabilities without relying on traditional supply chains.
A Girl Lets AI “Supervise” Her Therapist: AI enters psychological counseling rooms—cheap and never judgmental—but therapists think AI will pander to users and even give them “poison.” One person’s spiritual crisis won’t be solved just because a certain tool appears.
AI Investors Flock to Zhongguancun: Meituan co-founder Wang Huiwen reviewed his investment portfolio and discovered a roughly 2.5 square kilometer “golden box” formed around Tsinghua University, Peking University, Xueyuan Road, and Dazhongsi. Projects inside the box significantly outperform those outside. Almost all top companies in the mobile internet and AI eras were born here.
Tutorials & Resources for Individuals:
Overnight, ChatGPT Became the Second Claude: OpenAI announced integrating Codex into ChatGPT, fully rivaling Anthropic’s Claude.
Gemini Omni: Clone Yourself with AI in Under 15 Minutes: Claire Vo uses Google Flow and Gemini Omni to create her own AI avatar and generate a complete promotional video in 15 minutes.
To Save That 120 USD, I Built Mac Cleanup Software as an Open Source Skill: Digital life creator Kazke discovered his MacBook had nearly 100GB of Bilibili cache to clean. Remembering when he couldn’t afford CleanMyMac’s 120 USD one-time price, he spent some Tokens building a “Cleanup Junk Skill” and open-sourced it. Core design: generate interactive HTML reports with green lights (pure cache, one-click cleanup), yellow lights (requires human judgment like video cache), and red lights (system files, never touch). Supports both Mac and Windows.
I Built a Fitness Personal Trainer Agent with QClaw: From Exercise Records to Diet Advice: Still from Kazke, trying to build a personal fitness Agent with QClaw. Unlike one-shot Q&A Skills, Agents can save body data, diet records, exercise history, and long-term memory through independent workspaces.
Coze 3.0 Absurd Update: Pull Codex and Claude Code into One Project?: Hua Shu tried Coze 3.0 and discovered cloud Agents and local Agents are brought into the same project space. The core update is that coze-bridge automatically connects local Codex and Claude Code to cloud projects.
Kimi Work: “Cyber Zen, 887 Articles Published”: The author tested Kimi Work scraping the WeChat public account “Cyber Zen” for 887 articles’ dates, titles, and read counts, generating an interactive table.
After Making 700+ Courses, We Finally Solved a Problem: Based on Hundun Academy’s 700+ courses of business cognition system, they launched a Q&A Agent.
Full AI Video Editing, 100 Viral Videos Daily—HyperFrames, Remotion, Git Getting Started Guide: Introduces the three layers of the AI editing toolchain: HyperFrames for quick verification, Remotion for quality polish, and Git for backup. AI editing has moved from “concept” to “mass production”—the key is choosing the right toolchain and establishing backup mechanisms.
How I Use AI Agents to “Stop Getting Distracted”—Tracking Slack Channels, Emails, Following New Information: Esor Huang shares the “inbox organization” method on Computer Wanwu: let AI do the first round of filtering, output a very short pending list, then operate targetedly. Email processing is fixed at 3 time blocks; ChatGPT connects to Gmail and Google Calendar to filter 3-5 worth-dealing-with emails; community monitoring uses Codex for Chrome to scan X and Facebook; RSS readers are likewise pre-filtered by Codex.