The AI-Native Firm as a New Species
作为新物种的 AI 原生公司:一个研究纲领
First draft — June 2026 · maymedoc@gmail.com
Abstract
This paper proposes and argues a thesis: the AI-native firm is not a traditional firm improved by AI, but a new species requiring a new analytical category. We do not assert "new species" as a premise; we derive it as a conclusion from two completed theoretical results. Traditional theory of the firm carries four implicit axioms: (1) value grows monotonically with accumulation; (2) assets depreciate and liabilities are repaid to zero, the direction of the books set by time; (3) firm value equals the present value of cash flow; (4) marginal cost is positive and hard to attribute per unit. We show: Paper I's saturation-decoupling theorem rewrites Axiom 1; Paper II's reverse balance-sheet dynamics, value-leads-cash-flow, and marginal-cost collapse rewrite Axioms 2, 3, 4. Crucially, these four rewritings are mutually interlocking and irreducible to a parameter limit of the old framework. We therefore argue that a new analytical object is required, honestly positioned as a research programme.
Keywords: theory of the firm; AI-native firm; new species; paradigm; research programme; saturation decoupling; valuation. JEL: D21, L20, O33, G31.
AI disclosure. Generative AI tools were used to assist with drafting, editing, and literature organization. All propositions, arguments, and the core ideas are the author's own; the author has reviewed and verified all content and takes full responsibility for it.
1.Introduction: An Evaded Question
Most discussion of AI and firms stops at "AI is a powerful new tool." This frame carries an implicit judgment: the essence of the firm is unchanged; only the efficiency of its tools changes. This paper challenges that judgment. Our thesis: when a firm's capability base is AI-driven, accumulable almost without bound, and its value, assets, and cost structure thereby deviate systematically from the traditional firm, treating it as "an old firm with good tools" will systematically misunderstand it—it is a new species.
"New species" is a weighty term, easily abused as rhetoric. Our methodological discipline is: do not assert new species as a premise; derive it as a conclusion. We have proved several hard results in two prior papers; the task here is to show that, taken together, they force a conclusion stronger than "better tools."
This is a synthesis paper plus research programme, not a closed axiomatic theory. Its original contribution is to show that Papers I and II jointly rewrite four axioms, and that the rewriting is irreducible to the old framework. It does not claim that every firm using AI is a new species (only those for which all four axioms are rewritten). Borrowed and flagged: theory of the firm (Coase 1937; Williamson 1985), Kuhn's (1962) paradigm, and the species criterion from biology (as analogy).
2.Four Implicit Axioms of Traditional Theory of the Firm
A1. Value grows monotonically with accumulation (more capital/capacity → more value). ↳ supports growth theory, economies of scale.
A2. Assets depreciate, liabilities repaid to zero; direction of the books set by time. ↳ supports accounting depreciation, financial statements.
A3. Firm value \(=\) present value of future cash flow (value and cash flow synchronous). ↳ supports DCF valuation.
A4. Marginal cost is positive and hard to attribute precisely per unit. ↳ supports pricing, scale constraints.
Together these four axioms define the analytical object "what a firm is." Below we show, one by one, that they no longer hold for the AI-native firm—not slightly amended, but rewritten.
3.The Systematic Rewriting of the Four Axioms
Axiom 1 → value decoupled from accumulation (Paper I)
Paper I's saturation-decoupling theorem: when \(K\) is unboundedly accumulable but value enters through a bounded map \(x(K)\le\bar{x}\), firm value is finite for every discount rate, its finiteness independent of the speed of accumulation. This overturns A1: accumulation is not value; value has a structural ceiling. "More compute/data → higher value" no longer holds.
Axiom 2 → reverse balance-sheet dynamics (Paper II)
AI-native assets rise then saturate rather than depreciate monotonically; liabilities can convert into assets rather than being repaid to zero; the direction of the books is set not by time but by meta-capability \(A_5\). This overturns A2: time no longer unilaterally governs the books.
Axiom 3 → value leads cash flow (Paper II)
Anchored on net-worth accumulation, value systematically leads cash flow, the lead set by the Amdahl bottleneck. This overturns A3: value and cash flow are no longer synchronous; DCF, anchored on the lagging cash flow, systematically underestimates.
Axiom 4 → marginal-cost collapse (Paper II)
The AI-native firm's variable marginal cost tends to zero and is precisely accountable per token. This overturns A4: marginal cost is no longer a scale constraint, nor hard to attribute.
A single rewriting is only a "variant"; four mutually interlocking rewritings, in the same direction, are what constitute a new species.
4.Why This Is a "New Species" and Not a "Variant": Irreducibility
The AI-native firm cannot be understood as "a traditional firm with marginal cost going to zero." Taking \(mc\to 0\) and accumulation speed \(g\to\infty\) in the traditional framework cannot yield the two structural features: (i) the finite ceiling on value (Paper I: value enters through a bounded map, independent of \(g\); a traditional firm, even with \(mc\to 0\), still has value diverging with accumulation); (ii) the lead of value over cash flow (Paper II: arising from the Amdahl bottleneck, an organizational structure, not a cost parameter).
(i) In \(V=\int e^{-\rho t}\Pi(K_t)\,dt\) with \(\Pi\) increasing in \(K\), taking \(mc\to 0\) only raises the margin, not the unbounded dependence of \(\Pi\) on \(K\); value still diverges. The finite ceiling comes from the bounded map \(x\le\bar{x}\), a structural assumption, not the limit of any cost parameter. (ii) The lag arises from the "execution accelerable, judgment bandwidth not" Amdahl structure—an organizational feature with no corresponding parameter in the traditional framework. ∎
In biology, a single trait mutation does not make a new species; only when multiple key loci change jointly, and the result cannot interbreed with the original, is it a new species. The proposition gives the economic version of "cannot interbreed": the AI-native firm cannot be obtained by a parameter limit of the traditional firm. The claim has explicit falsification conditions—if it can be shown derivable from an old-framework parameter limit, the proposition is overturned.
5.Analytical Consequences of the New Species
To grant the new species is to grant that old tools will systematically err. Three direct consequences:
- Valuation. DCF is not "imprecise"; it is anchored on the wrong object—on cash flow already decoupled from value. The new species needs valuation anchored on net-worth accumulation (the FACE direction of Paper II).
- Accounting and governance. Traditional statements record only monetary items, while the new species' core assets (judgment capital, meta-capability) and liabilities (bandwidth debt, dependency debt) are off the books. Governing such firms needs a statement recording non-monetary net-worth dynamics.
- Boundary and scale. Coase explained firm boundaries by transaction costs; when marginal cost collapses and coordination can be borne by AI, the determinant shifts from transaction cost to judgment bandwidth—a hundred-person firm doing the work of ten thousand, the bottleneck not in cost but in judgment.
6.Open Problems as a Research Programme
This is a programme, not a closed theory. Its value is half in what is argued and half in the problems it honestly delineates as unsolved:
- Operationalizing the species criterion: how is "axiom rewritten" measured in data? Identifiable indicators are needed.
- Measurability of the non-monetary balance sheet: how to measure and audit the \(A/C\) accounts (Paper II proposes a framework; measurement is unsolved).
- Reconstructing boundary theory: judgment bandwidth as the new determinant of firm boundaries needs a Coasean formal model.
- Dynamics and equilibrium: when new and old species coexist, how do market equilibrium, factor prices, and competition evolve?
- Empirics: the testable predictions of Papers I and II need panel-data examination; the species criterion needs cross-sectional validation.
Claims like "new species" carry high academic risk and tend toward vagueness. This paper lowers that risk by reducing the big claim to the logical consequence of two falsifiable technical results, and giving an irreducibility proposition with explicit falsification conditions. Even so, this remains only the start of a programme—an invitation to the field, and especially to collaborators.
7.Conclusion
The AI-native firm is a new species, not an improved traditional firm. The four implicit axioms—value grows with accumulation, books governed by time, value synchronous with cash flow, marginal cost positive and hard to attribute—are systematically rewritten by Paper I's saturation decoupling and Paper II's reverse dynamics, value-leads-cash-flow, and marginal-cost collapse; and this rewriting is irreducible to a parameter limit of the old framework. The four joint rewritings plus irreducibility constitute a strict criterion for "new species." The consequences—for valuation, accounting, governance, and boundary theory—are real, and require reconstruction. We position this as a research programme and invite further collaborative research.
References
Coase, R. H. (1937). The nature of the firm. Economica, 4(16), 386–405.
Kuhn, T. S. (1962). The Structure of Scientific Revolutions. University of Chicago Press.
Williamson, O. E. (1985). The Economic Institutions of Capitalism. Free Press.
Mei, J. (2026a). Saturation Decoupling: Finite Firm Value under Unbounded Capital Accumulation. Working paper (Paper I).
Mei, J. (2026b). Saturation Room and Firm Value: A Valuation Framework for AI-Native Firms. Working paper (Paper II).
作为新物种的 AI-Native 企业
The AI-Native Firm as a New Species: A Research Programme
初稿 — 2026 年 6 月 · maymedoc@gmail.com
摘要
本文提出并论证一个命题:AI-native 企业不是被 AI 改良的传统企业,而是一个需要新分析范畴的新物种。我们不把"新物种"当作前提来断言;我们将其作为结论,从两项已完成的理论结果中推导出来。传统企业理论携带四条隐含公理:(1) 价值随累积单调增长;(2) 资产折旧、负债偿还至零,账簿方向由时间单向设定;(3) 企业价值等于现金流现值(二者同步);(4) 边际成本为正且难以逐单位精确归属。我们表明:Paper I 的饱和解耦定理改写公理 1;Paper II 的反向资产负债表动态、价值领先现金流与边际成本坍缩分别改写公理 2、3、4。关键在于,这四项改写相互锁合且不可约为旧框架的参数极限。我们因此论证需要一个新的分析对象,并诚实地将本文定位为一个研究纲领。
关键词:企业理论;AI-native 企业;新物种;范式;研究纲领;饱和解耦;估值。 JEL:D21, M21, O33, G31.
AI 披露。 生成式 AI 工具被用于辅助撰写、编辑与文献整理。所有命题、论证与核心思想均为作者本人;作者已审阅并核实全部内容,并对其负责。
1.引言:一个被回避的问题
关于 AI 与企业的多数讨论止步于"AI 是一个强大的新工具"。这一框架携带一个隐含判断:企业的本质不变,改变的只是其工具的效率。本文挑战该判断。我们的命题:当一个企业的能力基底由 AI 驱动、几乎可无界累积,其价值、资产与成本结构因而系统性地偏离传统企业时,将它当作"拥有好工具的旧企业"会系统性地误解它——它是一个新物种。
"新物种"是一个分量很重的词,易被滥用为修辞。我们的方法论纪律是:不把新物种当作前提来断言,而是将其作为结论来推导。我们已在两篇先前论文中证明了若干硬结果;此处的任务是表明,它们合在一起,迫使出一个比"更好的工具"更强的结论。
本文是综合论文加研究纲领,而非封闭的公理化理论。其原创贡献在于表明 Paper I 与 II 的结果共同改写了四条公理,且该改写不可约于旧框架。它不主张每个使用 AI 的企业都是新物种(只有四条公理全部被改写者才是)。借用并标注:企业理论(Coase 1937;Williamson 1985)、Kuhn (1962) 的范式概念,以及来自生物学的物种判据(作为类比)。
2.传统企业理论的四条隐含公理
A1. 价值随累积单调增长(更多资本/产能 → 更多价值)。↳ 支撑增长理论、规模经济。
A2. 资产折旧、负债偿还至零;账簿方向由时间设定。↳ 支撑会计折旧、财务报表。
A3. 企业价值 \(=\) 未来现金流现值(价值与现金流同步)。↳ 支撑 DCF 估值。
A4. 边际成本为正且难以逐单位精确归属。↳ 支撑定价、规模约束。
这四条公理共同定义了"企业是什么"这一分析对象。下面我们逐一表明,对 AI-native 企业它们不再成立——不是被略加修订,而是被改写。
3.四条公理的系统性改写
公理 1 → 价值与累积解耦(Paper I)
Paper I 的饱和解耦定理:当能力基底 \(K\) 可无界累积,但价值经由一个有界映射 \(x(K)\le\bar{x}\) 进入时,企业价值对每一贴现率都有限,其有限性独立于累积的速度。这直接推翻 A1:累积不是价值;价值有一个结构性天花板。"更多算力/数据 → 更高价值"不再成立。
公理 2 → 反向资产负债表动态(Paper II)
AI-native 资产先升后饱和而非单调折旧;负债可转化为资产而非偿还至零;账簿方向不由时间、而由元能力 \(A_5\) 设定。这推翻 A2:时间不再单方面支配账簿。
公理 3 → 价值领先现金流(Paper II)
以净值累积为锚,价值系统性地领先现金流,领先量由 Amdahl 瓶颈设定。这推翻 A3:价值与现金流不再同步;DCF 锚定于滞后的现金流,系统性低估。
公理 4 → 边际成本坍缩(Paper II)
AI-native 企业的可变边际成本趋于零,且可按 token 精确归属。这推翻 A4:边际成本不再是规模约束,也不再难以归属。
单一改写仅是一个"变体";四项相互锁合、方向一致的改写,才构成一个新物种。
4.为何这是"新物种"而非"变体":不可约性
设传统框架由任一有限的标量参数集合 \(\Theta\) 参数化(\(mc\)、\(g\)、\(A\)、\(\rho\)、\(\delta\) 等),企业价值为 \(V=\int_0^\infty e^{-\rho t}\Pi(K_t;\Theta)\,dt\)。则没有任何参数极限 \(\Theta\to\Theta^\star\)(包括 \(mc\to 0\)、\(g\to\infty\))能产出由 Paper I–II 所刻画的 AI-native 企业。障碍不在有限天花板本身——一个传统模型可用一个 ad hoc 有界生产函数模仿它——而在于:(i) 一个结构性且与增长无关的价值天花板;与 (ii) 一个由组织结构自由度(Amdahl 串行份额 \(1-p\),无法加速的判断带宽)生成的价值–现金流领先——它在 \(\Theta\) 中没有坐标——的联合存在。
第 1 步(天花板本身可约——承认):一个传统模型可设定有界生产函数使 \(\sup_K\Pi<\infty\),有限天花板出现;故 (i) 单独不区分物种。第 2 步(领先不可约):\(1-p\) 是一个组织结构参数,标示价值交付中受不可加速判断所限速的份额;\(\Theta\) 中 \(mc,g,A,\rho,\delta\) 无一控制变现中的串行瓶颈,故 \(\Theta\to\Theta^\star\) 的像无一呈现结构性领先。第 3 步(联合不可约):正是那个为价值封顶 (i) 的有界映射 \(x\le\bar{x}\),被带宽受限的链以滞后 (ii) 来变现;天花板与领先共享一个机制。用独立参数复现二者需两个自由旋钮,而物种只有一个耦合机制,故归约要么遗漏 (ii),要么破坏耦合。∎
在生物学中,单一性状突变不构成新物种;只有当多个关键基因座共同改变、且结果无法与原种杂交繁殖时,才是新物种。命题给出"无法杂交繁殖"的经济学版本:AI-native 企业无法由传统企业的任何参数极限达到——并非因为有限天花板异乎寻常,而是因为价值–现金流领先需要一个组织结构坐标(\(1-p\)),而传统参数空间缺乏它,且该领先通过单一机制与天花板耦合。这一带明确证伪条件的主张:举出一个有限参数的传统模型,其极限能从单一机制联合复现增长无关的天花板与带宽生成的领先,则本命题被推翻。
5.新物种的分析后果
承认新物种,不是换标签,而是承认旧工具将系统性出错。三个直接后果:
- 估值。 DCF 不是"不精确";它锚定于错误的对象——锚定于已与价值解耦的现金流。新物种需要锚定于净值累积的估值(Paper II 的 FACE 方向)。
- 会计与治理。 传统财务报表只记录货币性资产与负债,而新物种的核心资产(判断资本、元能力)与核心负债(带宽债、依赖债)都在账外。治理此类企业需要一张记录非货币净值动态的报表。
- 边界与规模。 Coase 以交易成本解释企业边界;当边际成本坍缩、协调可由 AI 承担时,我们猜想企业边界的决定因素从交易成本转向判断带宽——一个百人企业做着万人的工作,瓶颈不在成本而在判断。我们将此标为一个待形式化的猜想。
6.作为研究纲领的开放问题
这是一个纲领,而非封闭理论。其价值一半在于已论证者,一半在于它诚实划定为未解的问题:
- 物种判据的可操作化:"公理被改写"如何在数据中度量?需要可识别的指标。
- 非货币资产负债表的可度量性:如何度量与审计 \(A/C\) 账户(Paper II 提出框架;度量尚未解决)。
- 重构边界理论:判断带宽作为企业边界的新决定因素,需要一个 Coase 式的形式化模型。
- 动态与均衡:当新旧物种在一个经济中共存时,市场均衡、要素价格与竞争如何演化?
- 实证:Paper I 与 II 的可检验预测需要面板数据检验;物种判据需要横截面验证。
"新物种"一类主张带有高学术风险,且易流于含糊。本文以一种不同的方式降低该风险:将这一宏大主张化约为两个可证伪技术结果的逻辑推论,并给出一个带明确证伪条件的不可约性命题。即便如此,这仍只是一个纲领的开端——一份献给该领域(尤其是合作者)的邀请。
7.结论
我们论证了 AI-native 企业是一个新物种,而非被改良的传统企业。传统企业理论的四条隐含公理(价值随累积增长、账簿由时间支配、价值与现金流同步、边际成本为正且难以归属)被 Paper I 的饱和解耦与 Paper II 的反向动态、价值领先现金流、边际成本坍缩系统性地改写;且该改写不可约为旧框架的参数极限。四项联合改写加上不可约性,构成"新物种"的严格判据。承认这一点的后果是真实的:估值、会计、治理与边界理论都需要相应的重构。我们将本文定位为一个研究纲领,并由此邀请进一步的合作研究。
参考文献
Coase, R. H. (1937). The nature of the firm. Economica, 4(16), 386–405.
Kuhn, T. S. (1962). The Structure of Scientific Revolutions. University of Chicago Press.
Williamson, O. E. (1985). The Economic Institutions of Capitalism. Free Press.
Mei, J. (2026a). Saturation Decoupling: Finite Firm Value under Unbounded Capital Accumulation. Working paper (Paper I).
Mei, J. (2026b). Saturation Room and Firm Value: A Valuation Framework for AI-Native Firms. Working paper (Paper II).