DSTL - Audio in. Alpha out.

dstl

TERMINAL

LIBRARY

David Haber on Building at the Intersections: From Bond Street to Andreessen Horowitz

Join

David Haber on Building at the Intersections: From Bond Street to Andreessen Horowitz

Mar 20, 2026

The High Flyers Podcast with Vidit Agarwal

1:14:40

1K Views

THESIS

AI-native software companies that lead with workflow automation and embed deeply within traditional industries represent the dominant investment opportunity, as they access labor spend rather than IT budgets while building defensible data flywheels through outcome-based learning loops.

ASSET CLASS

SECULAR

CONVICTION

HIGH

TIME HORIZON

5 to 10 years

PREMISE

Traditional industries remain operationally inefficient despite decades of software adoption

Financial institutions, law firms, healthcare organizations, and other traditional enterprises continue to run operations with expensive human labor living in Excel, performing manual workflows rather than using software as a modeling tool. These industries historically represented small or difficult markets for software because the products could not actually perform the work—they could only assist humans. The cost structures of these businesses remain dominated by labor spend rather than technology spend, creating massive inefficiency that has persisted for decades.

MECHANISM

AI enables software to perform work, unlocking labor budgets as addressable market

The fundamental paradigm shift with AI is that software can now actually do the work, not merely assist with it. This transforms the addressable market from narrow IT budgets to the much larger labor spend across every function within traditional enterprises. Companies selling AI-native products that own end-to-end workflows—from intake through outcomes—can embed themselves as systems of record while accumulating proprietary data assets that improve product performance over time. The combination of bottom-up adoption from individual employees using AI tools and top-down board-level urgency around AI creates unprecedented enterprise sales velocity and shorter sales cycles.

OUTCOME

Vertical AI companies with workflow ownership and data network effects will capture disproportionate value

Companies that lead with software rather than point solutions, deeply embed within customer workflows, become the system of record, and build network effects or proprietary data assets will establish durable competitive advantages. These businesses can demonstrate not just cost reduction but meaningful revenue growth and better outcomes for customers. Industries that were never interesting for software—plaintiff law, home services, healthcare—are becoming the fastest-growing investment opportunities because AI unlocks access to labor economics.

NECESSARY CONDITION

Regulatory frameworks must remain permissive to innovation (avoiding the 'European' model) and open source development must remain unencumbered by downstream liability.

What's most unique about kind of this AI product cycle is that the software can actually do the work. And so whether that's in areas like plaintiff law in a company called Eve that I sit on the board of or in the home services space or in healthcare or in financial services, these were hard markets or small markets in some cases to sell software into. You know, they're some of the faster growing companies we have in our portfolio.

77:45

RISK

Steel Man Counter-Thesis

The thesis that AI-native fintech represents a massive opportunity because software can now do the work may be precisely inverted: the incumbents who control existing customer relationships, regulatory licenses, and balance sheets are better positioned to deploy AI than startups are positioned to acquire customers and regulatory approval. The speaker admits customer acquisition costs have become prohibitive for fintech, that the best fintech outcomes came from infrastructure plays like Plaid rather than direct financial product companies, and that New York lacks the scaling talent to build generational companies. If AI primarily creates efficiency gains rather than new distribution channels, the rational winner is the Goldman Sachs that employs AI to cut costs, not the startup that must spend those savings on customer acquisition.

RISK 01

Customer Acquisition Cost Death Spiral in AI-Native Fintech

THESIS

The speaker explicitly acknowledges that acquiring customers for financial products has become much more difficult and much more expensive, with Meta and Google extracting significant taxes on keywords and user acquisition. This creates a fundamental tension with the AI-native fintech thesis: while AI can reduce operational costs, the customer acquisition economics may deteriorate faster than AI can create savings, leading to unprofitable unit economics regardless of technological efficiency gains.

DEFENSE

The speaker articulates a defensive strategy of investing in companies that lead with software and have potential for network effects, rather than pure financial product companies. The thesis is that software-led approaches with embedded distribution can bypass the CAC problem that plagued product-led fintech companies like Chime, Brex, and Robinhood.

RISK 02

Commoditization of AI Capabilities Undermining Competitive Moats

THESIS

The speaker claims that moats still matter and are largely the same in the AI era, citing workflow ownership, system of record status, and network effects. However, foundation model capabilities are rapidly commoditizing and becoming accessible to all players. If the underlying AI capabilities that enable agentic work can be replicated by incumbents with existing distribution advantages, the software can actually do the work thesis benefits established financial institutions more than startups.

DEFENSE

The speaker never directly addresses what happens when Goldman Sachs, JP Morgan, or other incumbents with massive existing customer relationships deploy the same AI capabilities. The Eve plaintiff law example relies on proprietary outcome data, but most fintech verticals lack such defensible data assets. The assumption that startups can embed faster than incumbents can adapt is stated but not defended.

RISK 03

Talent Depth Constraint in New York Ecosystem

THESIS

The speaker explicitly identifies a structural weakness in the New York startup ecosystem: the lack of depth at the middle management level including VPs of marketing, VPs of product, and COOs who have seen companies scale from 100 to 1000 or 1000 to 5000 employees. This creates execution risk for portfolio companies that may have strong founders but cannot recruit the operational talent necessary to scale, particularly as Andre Horowits doubles down on New York-based investments.

DEFENSE

The speaker acknowledges this gap will close as the ecosystem matures and produces more successes. Additionally, the operating platform thesis is implicitly positioned as a partial mitigation, where firm resources can substitute for some of the scaled operational expertise that individual hires might otherwise provide.

ASYMMETRIC SKEW

Downside: Investments in AI-native fintech face compounding risks from CAC inflation, incumbent AI adoption, and talent scarcity that could result in widespread failure of the category thesis. Upside: Selective winners that achieve true workflow ownership and proprietary data flywheel effects in specific verticals could generate outsized returns, but this requires correctly identifying the narrow categories where startup advantages outweigh incumbent distribution. The skew favors concentrated wins in vertical-specific plays over broad category success.

ALPHA

NOISE

The Consensus

The market broadly understands that AI is transforming enterprise software and financial services, that software-led businesses with network effects are valuable, and that venture capital involves evaluating business models and competitive moats in established ways.

The market assumes that AI companies will win through better models, faster iteration, or cost reduction. Traditional moats (network effects, data flywheels, system of record status) apply, but the focus is on technology differentiation.

SIGNAL

The Variant

The speaker believes that AI represents a fundamentally different paradigm shift because software can now actually do work, not just enable it. This makes previously unattractive markets (plaintiff law, home services, traditional industries) suddenly valuable because you are accessing labor spend, not just IT budgets. He also believes that living between fields of expertise and resisting categorization is itself a compounding competitive advantage.

The speaker argues moats still matter and are largely the same as before. The critical insight is that companies must own the end-to-end workflow and deeply embed themselves to generate proprietary data that foundation models cannot train on. The example of Eve demonstrates this: plaintiff law case outcomes are not public data, so the company builds a unique data asset that informs better intake, time to resolution, and ultimately outcomes. This creates defensibility that pure technology plays lack.

SOURCE OF THE EDGE

First Principles Reasoning combined with Operator Experience. The speaker's edge comes from having been a founder who made mistakes (Bond Street leading with the wrong wedge product), observing Goldman Sachs operations from the inside, and now synthesizing these experiences to identify where AI can access labor budgets in industries that never bought software before.

CONVICTION DETECTED

• Moats still matter and they're largely the same • It's the fastest growing business we've ever seen • You can just imagine okay wow that is whatever 50% of the job of an investment banking analyst • Showing up to litigation in this market without Eve is like being unprepared for battle - swords against guns • The software can actually do the work • I always try to be the hungriest

HEDGE DETECTED

• I don't know that I have any superpowers • Maybe I'm projecting forward • Time will tell • I don't know that I have a perfect recipe to be honest • I don't know that it's endemic to like the founders necessarily • It's hard to describe