2026年3月9日的报道显示，风险资本机构在大规模布局人工智能之后，开始测试AI是否能取代其自身交易流程中的部分工作。创业公司Infinity Artificial Intelligence Institute获得了$100,000种子轮投资，其尽调由ADIN（Autonomous Deal Investing Network，AI投资决策网络）完成；ADIN于2025年上线，包含约12个拥有不同投资主张的AI“投资者代理人”（如Tech Oracle、Unit Master、Monopoly Maker），可在约1小时内输出商业模式与创始团队分析、尽调问题、合规风险、可寻址市场规模估计和估值建议，相比传统分析师的数日或数周流程更快。Aaron Wright称，目前风险投资的承销逻辑成功率很低：10倍以上“本垒打”般的回报仅约1%，而四分之三的交易甚至无法回收资本成本。他认为AI可推动向量化选投，形成类似Moneyball的转向，提高命中率。

与此相对，文中也呈现出VC对“投资是艺术”的反驳。Marc Andreessen强调，资本分配不仅是机械评分，而是对正确项目、正确时点、正确团队和“直觉口味”的判断。Keval Desai亦表示早期投资像在幼儿园“挑Michael Jordan”，Brian Nichols认为AI可辅助撰写备忘录、找项目或筛选创始人，但难以替代以个人关系与背书为核心的网络信誉。ADIN在尽职调查方面已显示优势，可集中揭示常被忽视的问题，包括出口管制与跨境数据传输风险，并可规模化识别长期尾部合规风险。然而最终决策仍由人工伙伴承担，尤其是创始人会面和是否放款；ADIN亦依赖创投猎头网络，并对其高达50%的carry收益分成（一般归属GP）给予异常高激励。

与此同时，AI也降低了创立公司的门槛，形成对传统创投经济模式的潜在生存威胁。Wright提到ADIN甚至会否定已获逾$20 million融资的项目，提示部分“热度”可能是非理性高估。更广泛的效应是：过去可能需要$2 million种子轮的SaaS软件公司，现在可能在六位数以下投入和少量AI增强团队下就能启动。过去软件类独角兽平均募集约$370 million，而Midjourney据称凭约100名员工并超过$300 million年收入达到独角兽地位。依赖高额资金和高频投资的VC因此面临可避免投资需求增加的新局面，行业可能回归更窄的职能，即连接科学突破与商业化；而在计算、数据中心与人才成本极高的基础模型领域，仍可能继续吸收大规模VC资金。

The article says that after venture capitalists poured record sums into AI, they are now testing whether AI can replace parts of their own investing workflow. In March 2026, Infinity Artificial Intelligence Institute received $100,000 in seed funding after investors debated it as both risky and promising. The startup was reviewed through ADIN (Autonomous Deal Investing Network), launched in 2025, which uses about a dozen AI investor agents (for example, Tech Oracle, Unit Master, and Monopoly Maker) with different theses. ADIN returns diligence questions, risk flags, TAM estimates, valuation guidance, and an allocation recommendation in about one hour instead of the days or weeks traditional analysts take. Aaron Wright says current VC underwriting has a low-success profile: “home-run” outcomes of 10x or more occur about 1 percent of the time, while three-fourths of deals do not even recover the cost of capital. He argues AI could move venture into a Moneyball-style, more quantitative era with higher hit rates.

Other voices stress that venture remains partly artistic. Marc Andreessen says selecting investments is not pure science; it depends on timing, team, judgment, and even taste. Keval Desai compares early-stage investing to “picking Michael Jordan in kindergarten.” Brian Nichols says AI can support memos, sourcing, and founder scoring but cannot replace network trust and personal endorsement. ADIN, however, already improves diligence by surfacing overlooked issues such as export controls and cross-border data transfer risks, and by flagging long-tail compliance threats at scale. Human partners still handle the last mile—founder meetings and final check decisions. ADIN also relies on scouts, whose incentives are unusually high because they can earn up to 50 percent of carried interest usually reserved for general partners.

A second trend is that AI may reduce demand for venture money itself. Wright says ADIN can reject even startups already above $20 million in funding, raising the question of whether enthusiasm is sometimes hype. More broadly, software startups that once needed around a $2 million seed round may now be built for under six figures by small AI-assisted teams. As a benchmark, SaaS unicorns have averaged about $370 million raised, while Midjourney is reported to have about 100 employees and over $300 million in annual revenue. If startup formation becomes cheaper, fewer deals require mega checks, forcing the industry toward a leaner role as a bridge from scientific advances to commercialization, while compute- and data-center-intensive foundational AI firms may still need very large VC rounds.