Interview with Liu Ye: OpenClaw Is Merely the “Hands and Feet”—We Must Evolve from “Digital Employees” to “Digital Organizations,” and from “Recruiting Soldiers” to “Deploying Formations”
TechFlow Selected TechFlow Selected
Interview with Liu Ye: OpenClaw Is Merely the “Hands and Feet”—We Must Evolve from “Digital Employees” to “Digital Organizations,” and from “Recruiting Soldiers” to “Deploying Formations”
When digital employees become ubiquitous, the decisive factors in AI startups’ success lie in “orchestration” and “aesthetics.”
Navigating Web3 tides with focused insightsDialogue | Zhang Peng
As everyone rushes to develop “digital employees” and “Agent tools,” endlessly competing within narrow use cases, where does the true moat for AI startups actually lie?
Recently, Zhang Peng, founder & president of GeekPark, held a forward-looking discussion with Liu Ye, founder of VisionFlow, following the emergence of OpenClaw. Born in 1979, Liu Ye is among China’s first-generation programmers—having witnessed the full evolution from low-level hardware to software, and from enterprise integration (ToB) to online education (industrial internet). After months of seclusion—and exhaustive conversations with leading AI researchers globally and top-tier domestic founders—he arrived at a sobering conclusion: treating AI as a “digital employee” to replace individual tasks reflects engineers’ over-simplification of real-world business.
In this dialogue, Liu Ye introduced highly insightful concepts and frameworks—including “progressive exposure” and the “high-dimension/low-dimension task matrix.” A future possibility gradually crystallized: AI’s next step won’t be an explosion of generic tool agents—but rather the construction of “digital organizations” equipped with coordination, reporting, and reflection mechanisms. When corporate culture becomes obsolete and low-dimensional work vanishes entirely, tomorrow’s CEO may no longer be the “Chief Executive Officer,” but instead a “Producer”—a role defined by exceptional aesthetic judgment.
This is a speculative exploration into organizational forms, commercial moats, and the ecological niches of next-generation entrepreneurs in the AI era—intended to spark deeper discussions among founders.
Below is GeekPark’s edited transcript of the conversation:
01 The “Ten-Thousand-Agents War” Has Already Begun
There’s So Much You *Can* Do—But What You *Should* Do Is What Matters Most
Zhang Peng: From Zuoye Box to your current deep engagement with the changes triggered by OpenClaw—what personal transformation have you undergone?
Liu Ye: I’m among China’s first-generation programmers—I started coding as a child. I’ve lived through the eras of BASIC, DOS, Windows, and today’s Mac; I witnessed the rise of China’s three major web portals. I worked on enterprise IT systems, aiming to build “China’s IBM.” Later, I pivoted to Zuoye Box, diving deeply into online education. Online education is an exceptionally profound industry—it represents the highest form—and arguably the “last train”—of industrial internet. This experience taught me that the core of industrial internet isn’t technology, but the industry itself—the business. Its pattern follows a clear progression: first information matching, then standardized products, followed by supply-chain integration, and finally complex, non-standardized services. Margins increase—and difficulty intensifies—as you move down this chain.
So when the AI wave hit, my first move was to spend nearly six months doing nothing else—tasking HR to arrange conversations with *everyone* possible. I spoke with chief scientists from top startup “unicorns,” core algorithm engineers and researchers from foundational model giants, and emerging AI founders—exhaustively. In total, I accumulated nearly one thousand hours of high-density dialogue. We reached such depth that when someone uttered the first half of a sentence, I could finish the second half. Consensus across the board had already converged.
After completing this round, the conclusion was strikingly uniform: everyone was building the same thing—“digital employees.” It reminded me of a strategic misjudgment once made by a prominent figure about cloud computing: “Alibaba’s cloud? Isn’t it just a fancy network drive?” Interpreting new phenomena through old frameworks only reveals their shallowest layer.
Today, everyone thinks building a digital employee—say, a “digital sales rep” or “digital customer service agent” powered by Claude—is trivial. Where’s the technical barrier? Where’s the moat? When burning billions of tokens per day has become routine, this resembles manufacturing—not a scalable leap forward. So I ask every founder the same question: Why *you*? Why are you the one? Are you younger? Smarter? Better at pulling all-nighters? Competing along a single dimension—that’s just the difference between “10.69 seconds” and “10.70 seconds.”
Zhang Peng: Yes—today there’s so much you *can* do, but what you *should* do is what truly matters. Any thoughts on that?
02 The Decade of Industrial Internet Will Repeat Itself—Again
Liu Ye: AI is profoundly different—but I believe its underlying patterns will echo those of industrial internet. Early stage: tools. Mid-stage: business processes. Late stage: consulting. When technology is immature, engineers inevitably lead the charge—they’re trained to over-abstract reality. Think of Baidu’s “box computing” concept: everything fits in a box. But the latter half of mobile internet was about content and services—not boxes.
Engineers’ mental models of organizations often oversimplify business complexity. Consider China’s first-generation internet portals—the ones who ultimately succeeded were Tencent and Alibaba. Neither was closest to the bleeding edge of tech—but both were deeply embedded in industry. Today, technology itself is becoming less decisive.
Zhang Peng: Liberal-arts graduates must be thrilled—coding skills seem optional now. But long-term, what *does* AI demand of people? What’s changed?
Liu Ye: Within China’s talent structure, I’ve observed a key issue. China’s first-generation programmers *were* product managers—because the PM role didn’t formally exist yet. It only became widely recognized around 2010, after Jobs launched the iPhone 4 and Zhang Xiaolong articulated his product philosophy—sparking the “everyone is a product manager” trend. Before that, programmers handled PM duties themselves. Programmers came first; PMs came later—so the first-generation programmers *were* PMs. They learned to code not for jobs, but out of passion—driven by love. It was precisely these unclassifiable, unconventional thinkers who excelled.
But the second generation—especially over the past decade of industrial internet—turned programmers into “code farmers,” while PMs became architects. Code farmers were trained *not* to think about business. Now AI has eliminated the “code” part—if they don’t evolve, they’ll literally be left with only the “farmer.” These young people are brilliant—but their understanding of industry is blank. Thus, today’s “Ten-Thousand-Agents War” is fundamentally just tool-layer saturation.
Look at late-stage industrial internet: companies like Alibaba and Meituan routinely hired top-tier consultants (MBB) for commercial analysis—and used consultants to guide PMs through business-process design. Why? Because internet PMs lack systemic thinking by default. Feishu was built this way. ByteDance, though purely internet-native, also heavily leveraged consulting firms to build internal processes. In the AI era, this pattern won’t weaken—it will intensify.
03 A Company’s Problems Are Never About Employees—They’re About Organizations
Zhang Peng: So you believe obsessing over “digital employees” as isolated units is largely meaningless.
Liu Ye: That’s my central thesis: Digital employees aren’t the endgame—digital *organizations* are. If digital employees proliferate to the point where even recruitment roles vanish—and everyone can access top-tier digital employees—then what? Does that guarantee every company will thrive? No. All company problems are *strategic* and *organizational*—never merely personnel issues.
Today’s Agents still *do work for humans*, not *make decisions for them*. Internally, we’ve modified OpenClaw to create something called MetaOrg—a kernel capable of generating entire *agent teams*. We solve any task not by dispatching a single agent, but by assembling an *organization*: one with coordination structures, reporting lines, missions, goals, and execution protocols.
Zhang Peng: But could one person someday *be* a department—or even an entire company?
Liu Ye: Excellent question. Let’s zoom in on micro-tasks—like using AI to produce a short video or draft a document. That requires multi-turn dialogue: you say something, it replies, you give feedback—that’s tool-agent usage. It’s clever, yes—but still just a tool.
So “person” vs. “department” isn’t about headcount. When we describe a senior job description, it usually says: First, execute diverse tasks using varied tools. Senior roles go further: understand intent, proactively plan paths, execute, deliver, report regularly, reflect on outcomes, and dynamically adjust strategy based on deviations. That’s high-order capability.
Zhang Peng: A competent department would need to operate at “Level 4 autonomous driving.”
Liu Ye: Exactly. Give it a skill—it handles complex tasks. Give it a *skill system*—it tackles complex integrated tasks. Orchestrate many agents—and you achieve even more: filming a short drama, for example. I often tell my team: when using MetaOrg, don’t see yourself as a manager—see yourself as the *chairman*. Push its boundaries.
For future young founders: previously, families might gift ¥500,000 to launch a startup. Tomorrow, they might grant a token budget for experimentation. How many tokens you’re willing to spend determines how sophisticated a role it can fulfill. Higher-level roles require longer reasoning chains—and more iterative trial, error, and reflection.
Zhang Peng: Returning to your earlier point—if an agent group can be decomposed into finer units—akin to role-and-capability decomposition—then when assembled into a team facing a core mission, the quality of each individual agent determines success or failure. This circles back to last-gen business-organization logic: talent density—i.e., superior talent quality—enables core missions to succeed and outperform competitors.
The crux is: if AI becomes universally capable—and anyone can tap the best AI—then beyond value creation via more efficient delivery of niche services, does another dimension emerge? Must we return to “talent density”—but now redefined as the atomic-level granularity of your agents’ and bots’ capabilities? Higher granularity = higher talent density = better results, efficiency, and even innovation on complex tasks. Is this reasoning sound?
Liu Ye: I agree. Large enterprises have an internal function—often called OD (Organizational Development). To assess whether an organization can win, standard practice is to map and benchmark *all* rival talent against your own—evaluating person-to-role and capability-to-role alignment to predict outcome. Thus, most battles are won by *organizational capability*, not business strategy. Alibaba is the textbook case: its relentless focus on organizational development enabled its “second spring.” Founding teams age—but organizations endure. Fundamentally, if you and I were rivals both using AI, and I built a powerful AI organization—with strong AI Organizational Development (AIOD) capacity—how would I construct it? I’d deconstruct every competitor’s agent skill system, analyze their skill code, then engineer superior skills in my own stack—even filling functional gaps they missed. For instance, if I have a strategy division, I’d begin with observation and analysis.
Huawei uses the “Five Views, Three Determinations” methodology. I joke with friends: adopting this alone lets us beat 99% of competitors. “Five Views”: industry trends, market customers, competitors, internal capabilities, strategic opportunities. “Three Determinations”: control points, targets, strategies. This framework filters out most rivals—because most play chess randomly, relying on fast thinking, while masters default to deep reasoning. Their first instinct is to think like a commander.
Zhang Peng: So “Five Views, Three Determinations” essentially means rejecting knee-jerk reactions—and institutionalizing long-chain reasoning.
Liu Ye: Masters combine deep research with structured thinking: first survey global best practices and data, then synthesize, reason deeply, and act decisively—striking with one lethal move.
Thus, I believe the sole core of future competition is *modeling traditional industry operations*: abstracting them into systemic capabilities that enable intelligent agent orchestration. This is the next-generation OD capability—and will evolve into AIOD: the only sustainable competitive advantage.
Alibaba’s edge lies in organization-building. Once established, its organizational strength ensures competitiveness across *any* business or rival. Jack Ma famously said: “The goal of battle isn’t conquering territory—it’s growing the organization.” Alibaba judges whether a battle is worth fighting *by its impact on organizational growth*. That’s elite-level thinking. Ma himself functions as a super information hub—flying 200 times yearly to gather insights, then refining organizational development. He’s a true chairman—not just a CEO.
That’s the highest organizational form we know: spanning generations and industries, sustaining success—and rebounding from decline. Typically, appointing the wrong CEO within ten years dooms a company. So learning from history—and applying higher-dimensional perspective to today’s challenges—makes optimizing existing models far more efficient than rebuilding from scratch.
Anyone can now easily assemble an agent; employee onboarding is trivial; open-source communities amplify transparency—there are few secrets left at the tool layer. Tool-layer competition can never outpace open source. So what *is* the core competence open source *cannot* replicate?
04 The Physics of AI Organizations: Why “Progressive Exposure” Is Key
Zhang Peng: Last era, organizational discourse emphasized culture, values, KPIs, etc. Transitioning from legacy management to AI-agent organizations—which elements can we discard entirely, and which must be retained but transformed?
Liu Ye: Anthropic introduced “skills” partly due to the “progressive exposure” principle in AI coding: feeding AI massive, chaotic inputs causes context corruption and attention collapse. Progressive exposure preserves AI’s focus and output quality. Manual progressive exposure equals full human dialogue—inefficient. Hence, skills’ core value is *hierarchically decomposing complex tasks* to enable progressive exposure.
This mirrors corporate governance: boards focus on strategy; CEOs on tactics and executive management; employees handle simple tasks. A 300-person meeting would be impossible. The essence of organization is *layered information processing*—like database normalization boosting efficiency via compression. Complex problems *must* be decomposed and exposed progressively—not dumped as monolithic context. That’s the timeless logic of traditional enterprise organization, constrained by finite compute at any given time.
Zhang Peng: Models waste huge compute regenerating everything from scratch—inefficient.
Liu Ye: Impossible otherwise. Layered, progressive exposure is fundamental—resources must be called *only when needed*, dictated by AI’s inherent capability limits. Another reason Anthropic launched skills: complex tasks now exceed basic physics—skills break them into low-dimensional subtasks. Task differentiation hinges not on difficulty, but *complexity dimensionality*: low-dimensional hard (e.g., coding), high-dimensional hard (e.g., strategic planning).
Horizon Robotics’ Yu Kai proposed a classic quadrant model: all professions fall into four categories by “competition intensity” and “dimensionality”—High-Dim/High-Competition, Low-Dim/Low-Competition, Low-Dim/High-Competition, High-Dim/Low-Competition. Sales and engineering are Low-Dim/High-Competition; PMs and CEOs are High-Dim/High-Competition; scientists are High-Dim/Low-Competition—where perhaps only one person globally researches a topic. High-Dim/High-Competition tasks (e.g., premium short dramas, great novels) remain beyond AI; Low-Dim/High-Competition tasks (e.g., code optimization) AI handles well. Higher-dimension tasks have *scarcer data sources*—yet demand *more training data*. That’s why text models emerged first, image/video models later—and short-video models struggle: a supply-demand mismatch resolved only by skills-based decomposition, akin to splitting an irreplaceable CEO role into three base positions.
Zhang Peng: Low-Dim/High-Competition tasks will likely be fully replaced by AI.
Liu Ye: Fully replaced—and replacement is already underway.
Zhang Peng: Indeed. So all Low-Dim/High-Competition work should be rapidly automated via AI—decomposed into skills, then executed via agent orchestration—potentially without human involvement.
Liu Ye: My initial vision: IBM and Accenture—the world’s two largest consultancies—sell *processes*, not tools. Their core business is distilling industry best practices, aligning them with digitalization, and implementing them for clients purchasing risk frameworks or IP. Our current focus is building *skills clusters*: partnering with domain experts to extract, align, and standardize their capabilities. Like Zuoye Box: collaborating with Beijing No. 4 High School, Renmin University High School, Gaokao exam designers, and Xueersi teachers to codify question design, explanation, and grading methods—then partnering with Baidu’s algorithm engineers to build the system. It’s all about aligning best practices. Core organizational capability? Assembling cross-domain teams—deeply versed in industry, engineering, *and* able to engage top vertical experts—while mastering business development, hiring, and management. That’s the DNA of next-gen AI SaaS companies.
Zhang Peng: Extending this: future organizational forms should be reverse-engineered from business needs. Organization is fundamentally an *orchestration architecture*—an “operating system” placing human productivity units into optimal structures. Replace humans with infinitely scalable AI—and you get self-reinforcing expansion. Legacy culture may now translate to *goals and context*, shedding slogans, “three-strike meetings,” icebreakers.
Liu Ye: Culture manages *intent*, not business intent. Last era: strategy began with vision → vision shaped values → organization served strategy → business validated all. Culture was merely governance—neither directly strategic nor necessarily the founder’s personal preference.
Zhang Peng: Human execution of strategy involved massive friction—does AI eliminate that?
Liu Ye: Yes—culture is irrelevant in AI era. Culture is the belief layer of human organizations—but AI has no flesh and blood, no cultural need. AI’s core requirement is compute.
Zhang Peng: So AI needs *goals and principles*. One document suffices—goals and principles instantly sync and faithfully execute across all productivity units, eliminating deviation. A huge friction point vanishes.
Liu Ye: Correct. Legacy org: Strategy → Culture → Talent → Execution. AI org: Goal → Principle → Skills → Orchestration. The management chain halves.
05 The Final Moat: Aesthetics and Orchestration
Zhang Peng: What’s the enterprise’s new moat? Talent quality becomes Skill Set—if I possess aesthetic judgment, I can source the world’s best skills. Then “orchestration” becomes the next layer—yes?
Liu Ye: Like Shenzhen’s Huaqiangbei selling all electronic components—yet not everyone builds Apple. Steve Jobs’ biography defines aesthetics clearly: having seen enough great things worldwide to discern quality. Without exposure to great products, processes, or organizations—you cannot create greatness.
Zhang Peng: Exposure precedes aesthetics.
Liu Ye: Exposure plus innate talent—that’s all.
Zhang Peng: Aesthetics manifests two ways: deliberate design/orchestration, and recognizing/embracing emergent excellence amid chaos—non-conflicting approaches.
Liu Ye: Absolutely non-conflicting. Apple both builds in-house *and* acquires third parties—the core is aesthetic judgment: no reinvention unless necessary; build only when essential.
Zhang Peng: So is orchestration about setting modules first, then confirming paths post-emergence—or predefining all paths for design-driven orchestration?
Liu Ye: Emergence is non-controllable—you set seed rules and principles first. That’s where aesthetics live. Great engineers build usable OpenClaw in 500 or 5,000 lines; weak ones write 50,000 lines with inferior results. Seed rules remain human-defined.
Zhang Peng: So waiting for emergence in chaos is inefficient—orchestration remains critical. Ultimately, must this originate solely from the founder—or more aptly, a “producer”?
Liu Ye: “Producer” fits perfectly. Even with emergence and scale effects, you need data labeling, cleaning, and continuous algorithm alignment to prevent chaotic sprawl.
Orchestrators depend on business complexity—complex tasks defy solo completion. Filming a short drama or crafting prompts faces real hurdles. “One-person companies” are overhyped—reality resists infinite simplification. While one person operates a computer, mastering *all* high-dimensional skills is near-impossible. Superhuman polymaths like Elon Musk or Li Feifei—commanding arbitrary roles across domains—are exceedingly rare.
Zhang Peng: If we can tap global top-tier agents and skill systems—say, a stellar screenwriter—could theoretical access to these resources yield a globally renowned, profitable film? A writer brings the core spark (great script) but can’t execute all steps. Is this “core spark + global resources” closed loop viable?
Liu Ye: It’s fundamentally a *data problem*: does high-dimensional information exist in storable form? Training CEO skills lacks sufficient data: Ren Zhengfei’s 10,000-word essays or Jack Ma’s oral histories can’t fully capture their high-dimensional cognition. Even aggregating global earnings reports and CEO speeches won’t train a CEO-capable model—because core CEO competence is *tacit knowledge*, impossible to fully textualize.
Zhang Peng: So CEO core competencies remain un-vectorizable—constraining the “one-person company” ideal. Even with perfect dimensional advantages and global resources, you still lack the orchestrator—the essence is orchestration capability. Having the best “components” demands elite orchestration.
Liu Ye: Same for PMs—tacit knowledge defies full textualization. That’s why AI companions and AI-generated content feel “flat”: insufficient high-dimensional tacit data. When data is scarce, prioritize skills; when abundant, build models. Robots stall due to insufficient data.
Zhang Peng: Thus, future company competition hinges not on accessing top models—initial AI resources appear equal, compute ties to capital and business-closure capability—and differences ultimately converge on the “producer”: their orchestration ability and the novelty/significance of their goals. These define core competitiveness.
Liu Ye: A former McKinsey partner told me: McKinsey’s core business is extracting best practices, modeling them, and helping clients implement them. Consulting Chinese automakers? They learn Toyota’s methods from Japanese colleagues—essentially copying and deploying best practices.
Mimeng’s short-drama case is instructive. A Chinese-literature graduate, her core team comprised Peking/Tsinghua math and CS talent—dedicated to reverse-engineering viral short-video logic, achieving exceptional hit rates. This is essentially *social engineering modeling* for an industry—overfitting risks exist, but the modeling direction is right.
IBM, Accenture, McKinsey all do this: First-gen McKinsey modeled best practices into partners; IBM digitized them into processes—both sell “management and process.”
Zhang Peng: Core is extracting best practices, then rigorously validating deployment—that’s the future business-organization battleground. Only deep decomposition enables efficient orchestration. So your next focus advances along this path?
Liu Ye: For three years, we focused on AI-ToC, rebuilding the entire teaching/research system via MetaOrg—not just “AI for efficiency.” We built a full Agentic teaching/research organization—running virtual teams: language-learning researchers tracking second-language acquisition theory; vertical corpus collectors harvesting authentic expressions; dialogue evaluators establishing multi-dimensional oral-proficiency standards; dialogue designers translating pedagogy into natural human-AI interaction; exercise-container designers solving practice-format/content alignment; data analysts mining real user-behavior signals for learning efficacy. Each team has its skills, workflows, and evaluation metrics. Today, ~80% of textbook tagging, monitoring, user insight, and product iteration is AI-executed.
Our evolution: from “AI as feature” → “AI as organizational capability.” English teachers occupy mid-complexity roles—we abstracted them via MetaOrg to generate other roles; integrating latest skill architectures may unlock even higher roles.
We’ve completed end-to-end AI tutor construction—including orchestration abstraction and engineering implementation. Next, MetaTutor will evolve into MetaOrganization—its smallest unit is *role*, not employee—centered on inter-role collaboration and management. Our priority now: engaging top-tier CEOs across industries—CEOs are the core “producers.”
Zhang Peng: So you’re delivering something closer to a scalable department?
Liu Ye: Target is “company”—large companies comprise multiple small companies, whose smallest unit is *role*. Strategic choices span industries, but product iteration starts with roles—if roles fail, even stellar managers can’t forge efficient organizations.
Zhang Peng: Building a department well begins with decomposing its capabilities and roles—then decomposing role-specific skills—and pursuing SOTA-level skills.
Liu Ye: One core method: co-create with top-tier client enterprises. Skills must be validated by top clients—as subordinates’ proposals require managerial review—no self-indulgence. Short-drama modeling must earn top-industry validation—else it’s not truly elite. Everything demands evaluation.
Midjourney produces superb images because its team blends photographers and engineers—possessing elite visual aesthetics. LV’s Stable Diffusion model vastly outperforms generic ones because LV holds the world’s finest image aesthetics and data. Evaluation capability is paramount. To build an AI company, emulate IBM or Huawei: IBM serves top automakers, masters automotive best practices, and sells them outward; Huawei spent ¥4 billion acquiring IPD processes—for internal use *and* external licensing. That’s the moat.
Zhang Peng: Fundamentally: decompose best practices into skills → achieve SOTA skills → upgrade to SOTA roles/departments → orchestrate into SOTA business. That’s the clear path to business excellence. One final question: how keep skills current? Like biological evolution, today’s SOTA may be obsolete tomorrow—how adapt?
Liu Ye: Core logic mirrors human/biological evolution: sense → plan → act → reflect. Maintain high talent density and cross-domain agility—connect one end to tech frontiers (researchers), the other to business models—and co-create with top clients in real scenarios for continuous evaluation and optimization. That’s the only way.
Zhang Peng: Thus, top-company best-practice systems could catapult mid-tier firms—but such systems likely require resources and capital inaccessible to SMEs and young founders. Consulting has evolved from services to toolified products. Are new-gen opportunities confined to the skill layer? How disrupt skill-layer innovation—and avoid “noble-cycle” entrenchment?
Liu Ye: In last-gen SaaS, Salesforce, Palantir, Notion, Slack proved young founders retain opportunity—avoiding domains lacking personal advantage, focusing on universal skills, finding their niche. Notion exemplifies this: avoiding specific business processes, it abstracted text-note functionality into a universal tool. The world will become countless agents collaborating—new founders must find their niche first, then leverage strengths, anchor to future trends, and avoid becoming time’s enemy. Past decades: first-gen internet founders were overseas returnees (cognitive edge); second-gen were programmers (tool boom); third-gen industrial-internet founders were serial entrepreneurs. Patterns are clear—new founders must see the mid-game and their strengths.
Zhang Peng: So you view localized skill innovation/optimization as limited—thus, new-gen’s biggest opportunity lies in *goal innovation*: identifying emergent era-defining goals, combining elite skills, and evolving continuously—to build new systems atop new goals and break through.
Liu Ye: Skill competition is subtle: today’s hot skills will be displaced if someone aligns with even more elite human experts to build superior ones. Back to the moat: early entrants rarely win—they often become “nutrient soil” for higher-dimensional rivals.
Zhang Peng: Fear being just a “loader”—merely laying groundwork for higher-dimensional rivals. Efficiency optimization on existing goals is meaningless—efficiency advantages inevitably flatten. So breakthroughs demand *fundamental goal divergence*.
Liu Ye: Correct—without evolving into core power, you only nourish higher-dimensional rivals. Business is simple: know your customer, serve them well, make them indispensable. Any young founder unclear on “who is the customer” cannot optimize.
Zhang Peng: Also watch *incremental markets*—competing in saturated ones is brutal. If your business succeeds, it lifts peers to the same advanced level—these incumbents possess both wealth and cognition, making it near-impossible for youth to compete in存量.
Liu Ye: Last-gen SaaS success—Notion, Slack—centered on *goal differentiation*.
Early SaaS days: Chinese funds favored scientists. Later, they realized scientists excel at collaboration—not entrepreneurship. Scientists inhabit high-dimension/low-competition domains, clashing with commerce’s high-dimension/high-competition logic. Higher the domain dimension, harder the pivot—and core mindsets differ entirely. Early in any field: technical competition (low-dim/high-comp, immature tech). Post-maturity: commercial competition (high-dim/high-comp, led by industry veterans, PMs, operators). Example: early iPhone App Store charts featured programmer-built apps; years later, industrial internet rose—and all programmer-led apps vanished from charts.
If AI follows mobile-internet logic, Silicon Valley’s core force remains experienced practitioners—just as China’s industrial internet was driven by serial entrepreneurs. Young founders’ opportunity remains *differentiated goals*.
Join TechFlow official community to stay tuned
Telegram:https://t.me/TechFlowDaily
X (Twitter):https://x.com/TechFlowPost
X (Twitter) EN:https://x.com/BlockFlow_News
极客公园
