Junior entrants are 70-80 percentage points less likely to become senior VCs than those who enter at mid-level: The single most powerful predictor of making partner is the seniority at which you enter the industry. Middle-level entrants (VP/Principal) have dramatically better odds, and lateral moves across firms count. The implication: if your goal is partnership, skip the analyst track and build operational experience first.

Prior startup experience is the strongest positive signal, while gender remains the only significant negative factor: VCs who previously worked at VC-backed startups, whether as founders, C-suite, or rank-and-file, are significantly more likely to reach partner. Female VCs are significantly less likely to be promoted, a gap that persists even in recent years, making gender the only economically significant negative predictor across all individual characteristics studied.

MBA holders are more likely to make partner but less likely to back successful deals: Holding an MBA increases promotion probability but actually lowers the likelihood of making successful investments, all else equal. Advanced non-MBA degrees (MS/PhD) help with both. Top-school MBAs (Stanford, HBS) negate the negative deal performance effect, creating a clear puzzle: the "direct path" through an MBA program is actually the harder route to real investment success.

A 1.5% improvement in pool quality raises portfolio success rate to ~9%, while a 10% improvement in picking skill only gets you to ~7%: Grounded in Bayesian statistics and base rate analysis, the article demonstrates that pool quality acts as a mathematical ceiling on returns. Even the most skilled investor is destined to pick false positives when the candidate pool is weak, because noise overwhelms signal.

The Thiel Fellowship produced a 13.79% unicorn hit rate among fellows, nearly an order of magnitude better than Y Combinator and other elite institutions: The article uses this as the clearest case study for origination-driven alpha. By finding future founders before they start companies, through grants and community building, 1517 Fund (an outgrowth of the fellowship) created a structurally superior deal pool that no amount of picking skill can replicate.

Research-driven origination at firms like Compound (tracking academic citation alerts, monitoring sleeping beauties in scientific literature) enabled early positions in voice models (Deepgram, 2016), autonomous driving (Wayve, 2017), and world models (Runway, 2018): Rather than following great founders toward interesting markets, research-oriented firms identify breakthrough domains first and meet the founders there, creating timing advantages that are invisible to network-dependent investors.

Public SaaS multiples have collapsed to 4.1x NTM revenue (down 80% from the 2021 peak of 22x), with implied perpetuity growth dropping from 4.7% to 1.1% in just three months: Horizontal SaaS is down 35% over the past 12 months while vertical SaaS and infrastructure are roughly flat. The market is pricing in a world where AI permanently compresses growth rates, moats, and addressable markets for traditional software.

Private AI companies trade at 61x ARR at Series B/C while public high-growth software sits at 9.7x, a 528% premium, but AI-native companies generate 10x more revenue per employee: Cursor produces $6.1M ARR per FTE vs. Salesforce at $0.54M. The gap explains the valuation divergence: these are structurally different businesses. Growth-adjusted, private AI multiples are actually at a steep discount to public markets (0.05x vs. 0.37x).

54% of CIOs are actively pursuing vendor consolidation, and 83% say they are open to replacing their CRM with an AI-native vendor: 45% of AI budgets come directly from existing software line items, not new budget. AI spending is largely zero-sum for the current software stack. Meanwhile, software engineer job postings have recovered to near-baseline (97 on an indexed basis), suggesting AI is expanding the demand for software, not replacing developers.

Over 50% of venture deal value now flows into AI-oriented investments, with early AI fund vintages tracking closest to SaaS-era returns, not the dotcom bubble: Hamilton Lane compared AI-era venture IRRs by fund age to prior cycles and found the pattern most closely resembles the SaaS era, which delivered strong sustained returns, rather than the dotcom period which produced poor overall performance despite a few big winners.

AI-native startups are reaching $30M revenue faster than any prior venture cycle, and Hamilton Lane assigns roughly 60% probability to "no valuation bubble": The key differentiator vs. prior hype cycles: capital is being deployed by companies with massive balance sheets and cash flows (not leveraged startups), and real-world infrastructure constraints (data centers, energy) are throttling the kind of unchecked expansion that typically inflates bubbles.

Public market concentration risk is extreme, with the Magnificent 7 effectively determining portfolio performance for the past 6+ years: Hamilton Lane argues bluntly that a "diversified" public equity index is an illusion. Seven AI-related stocks drive returns, and this dynamic will persist for at least another two to three years. Private markets, by contrast, offer genuine diversification into AI across earlier stages and broader verticals.

Funds optimize for a single objective (maximize carry with fewest people, shortest time); firms add a second: building compounding competitive advantage: Most funds are run by an alpha decision maker focused on the next marginal deal, spending almost no time thinking about moats. Compensation structures reward investment returns split among small teams, not institutional durability. Firms, by contrast, are run by entrepreneurs who think constantly about competitive advantage.

a16z raised over $15B and captured 18% of all U.S. venture dollars allocated in 2025, structured across six distinct strategies (American Dynamism, Apps, Bio, Infrastructure, Growth, and other venture): Haber frames a16z as a product company for entrepreneurs rather than a fund to manage. The decentralized GP structure, where domain experts like Alex Rampell, Martin Casado, and Chris Dixon lead their own strategies, is positioned as a model that traditional fund structures cannot replicate.

Enduring financial institutions like Apollo (permanent capital structures) and Goldman Sachs (embedded distribution through wealth management) compound advantages in ways that most VC firms never attempt: Haber explicitly draws on Goldman's 160-year history of entrepreneurial partner-led expansion to argue that the best venture firms of the next decade will look more like diversified financial institutions than traditional GP/LP vehicles.

USV has deployed named agents across core VC functions: Sally (meeting scribe), Ellie (email monitor), Arthur (deal analysis), Nancy (news monitor), Felix (finance data), Connor (calendar), and Leo (legal counsel): Each agent is onboarded like an employee with a job title, access to internal tools, and explicit responsibilities. The firm treats agent naming and anthropomorphization as a deliberate adoption strategy, not a gimmick.

The core data model is built around "mentions," structured records created by background agents that process meeting transcripts, emails, and calendar invites to track every time a company or person is referenced internally: This creates a continuously updating internal knowledge graph that replaces manually maintained Notion pages. The system connects to Granola transcripts, Google Drive documents, historical blog posts, and tweets, giving agents the same context a human team member would accumulate over years.

USV frames the shift as moving from "Build something people want" to "Build something you want," drawing an analogy to MIT's Building 20 where researchers freely modified their own workspace: Each Friday, Arthur (the analyst agent) runs a self-improvement reflection loop, analyzing changes in deal log status and team meeting mentions to refine its model of "USV Taste" in companies. The firm views encoding tacit institutional knowledge into agent systems as the key unlock.