Everyone is asking which AI company to invest in. That is the wrong question.
(Bear with me, this may take a minute.)
AI is not the investment thesis. Disruption is. The question is which industries AI is about to upend and when, whether you back the companies replacing incumbents or transforming them, and whether the company you are backing has an unfair advantage in doing so. Historical disruptive platform shifts sort themselves into three layers.
Infrastructure is the physical foundation: chips, data centers, networking, energy. Capital-intensive, winner-take-most, defined by scale and defensibility. The barriers to entry are so high that most never get the chance to play..
Enablement is the software layer that bridges infrastructure and application. Workflows, processes, middleware, the coordination tissue that turns raw technology into finished product. Historically defined by stickiness and integration depth.
Application is where the end customer lives. Consumer or enterprise, it doesn't matter. Behavior changes here. Value concentrates here.
Look at the companies with the most durable, large-scale outcomes across any technology wave: Google, Meta, Amazon, Apple, and a pattern emerges. They started with a dominant application and methodically integrated down: building or acquiring enablement tools, then owning infrastructure. Vertical integration executed from a position of distribution strength.
The inverse is equally true. The great infrastructure companies of previous waves, Sun Microsystems, Cisco, the telcos, saw their margins compress and their luster fade as the layers above them scaled and commoditized what they built. Infrastructure can produce enormous near-term outcomes, NVIDIA is proof of that, but history suggests it rarely holds as a standalone position. The anchor is application. Everything else follows, or gets squeezed.
The early venture opportunity in AI infrastructure is mostly behind us. The LLMs, the data centers, the foundational models, those bets are largely made and it has mostly become a game for the giants. It’s hard to write a meaningful venture check into that layer today.
That closes one door. The other two are wide open.
AI is doing something different to the enablement layer. Not the slow commoditization that took down previous infrastructure giants, but a faster compression driven by the models themselves.
In every previous wave, enablement was a clean, standalone category built to last. Salesforce spent decades helping companies sell better and never needed to become a seller. That worked because the lane was defensible: bridging infrastructure and application required human coordination, and enablement companies built durable businesses on top of it.
AI is eliminating that human layer. The decisions, the logic, the business rules that once required human-built connective tissue are now handled by the models. The coordination layer that justified a generation of SaaS businesses is compressing fast.
Enablement companies without a path to something more defensible are on a clock.
The enablers that will matter are the ones that use their position to go deeper, not wider. They get inside the atomic unit of work in specific verticals. Not software that sits alongside the workflow, but software that creates the work product. The natural direction is toward co-production: working side by side with the customer to produce the output itself, not just the tools to produce it.
Three ingredients earn that right: data, trust, and integration.
You need to hold the fundamental building blocks of how work gets created. You need the relationship. And you need to be woven into the work so completely that others can’t see the entry point, let alone catch up. Not because you are difficult to remove, but because you have something truly proprietary and you are part of that business’s core equation.
This is where two distinct investment theses emerge.
De novo:
A new company doing something better, faster, cheaper, made possible by AI, that replaces what existed before. These are application plays, the ones that bypass or displace incumbents entirely. The winners tend to come last in a technology cycle, built on the infrastructure and enablement already in place. We are early in that wave and it’s exciting.
Our portfolio reflects this: Decagon displacing the traditional call center with AI-native customer support; Arya replacing the administrative layer in healthcare, handling scheduling, onboarding, and compliance with modular AI agents; Harmattan building autonomous defense systems that simply did not exist before AI made them possible. Different industries, same thesis.
Co-producer:
A company that works alongside incumbents, not as a vendor, not as a tool, but as a strategic partner with skin in the output. This is not transformation consulting. This is not SaaS. It is shared production, earned through depth of data, accumulated trust, and irreplaceable integration.
Our investment in Laurel illustrates the second thesis. They’ve spent nearly a decade automating time capture and billing intelligence for enterprise legal and accounting firms. But the origin is not the opportunity. The real asset is what sits underneath: the atomic unit of data on how work gets created across knowledge work more broadly, the fundamental building blocks of how professionals think, prioritize, execute, and deliver. That understanding, combined with trust earned over time and deep ecosystem integration, gives Laurel the right to move from tool to co-producer. Not to become the law firm or the consulting firm or investment bank, but to work alongside them, producing the output. An indispensable partner in the work output itself.
SaaS investors, almost by definition, think like enablers. They built careers helping enterprises operate better, serving the companies serving the consumer, never the end customer directly. That instinct served them well.
It may also be exactly the wrong lens for this moment. Not because SaaS investors lack instinct, but because their instinct was built entirely one layer removed from the end customer. They have never had to ask what a company owns, only what it enables. That is a different question, requiring a different muscle: different customer relationships, different metrics, different moats, different risk profiles.
I developed this customer-centric mindset through investments in Uber, Facebook, DraftKings, LinkedIn, Zappos. Application bets, every one. Direct outcomes, end customers, businesses that replaced or bypassed what existed before. Because of AI, I find myself applying that same lens to the enablement layer, not asking what a company enables but what it could co-produce, and ultimately own.
That is the shift.
The application era is opening. The companies that will define this cycle, the way Amazon, Netflix, and Uber defined the last, are beginning to emerge. They will move fast, dancing among the giants. And unlike previous cycles, they will do so in territory where the largest technology companies in history are already operating. That is not a reason to stay out. It is a reason to be precise about where you get in. I’ll unpack this in a future post.
We are not primarily investing in AI. We are investing in disruption, and AI has taken the door off its hinges.
That means two types of bets: de novo builders doing it better, faster, cheaper, replacing what came before. And co-producers working alongside incumbents, not just enabling transformation but sharing in the output.
Every industry is in play. The decision founders and investors face is the same: do you try to replace the incumbents or help them transform?
I back both.