In Parts 1-3, I argued that the most durable AI companies will:
That naturally raises the next question: Where, specifically, is the opportunity today?
After years of investing across regulated operations and critical systems, I think about the landscape in two broad categories:
Both are underpenetrated. Both are consequence-heavy.
1. Soft Assets: Regulated Workflows that Run Institutions
These are systems that encode rules, compliance, oversight, and accountability.
They include:
What makes these compelling:
This isn’t about making employees 10% more productive. It’s about preventing regulatory penalties, compliance failures, revenue leakage, and operational blocking.
In these systems, AI doesn’t replace people. It structures how people make decisions under scrutiny. When embedded deeply, these systems become very hard to displace.
2. Hard Assets: Physical Infrastructure that Cannot Fail
This is where AI meets the real world.
Examples include:
What makes these attractive:
These environments are messy:
The AI that succeeds here is not flashy. It's reliable, integrated, explainable, and operationally aware. In these systems, AI becomes part of the control layer. And control layers are durable.
Acceleration Changes the Window
There is a common narrative circulating: accelerating AI means faster disruption everywhere.
Which is true...at the surface layer.
Differentiation that took two years to build can now be reproduced in six months as models improve. Feature velocity compresses. UX advantages fade. Model access equalizes.
But for systems embedded in soft and hard assets, the dynamic inverts. Faster capability allows them to compound more quickly:
You’re racing for trust and operational dependency. This model compounds through deployment, scrutiny, and time under pressure.
Penetration across both soft and hard asset categories remains low. The companies that embed now—before the surface layer fully commoditizes—will be structurally harder to displace when it does.
Acceleration doesn’t eliminate the window. It shortens the time you have to claim it.
Where I'm Seeing the Most Asymmetry
Across both soft and hard assets, three themes stand out:
1. Backlog Compression
Massive institutional backlogs exist in healthcare records, permitting, government case processing, and compliance documentation. AI that can reduce multi-year backlogs without increasing risk creates immediate ROI and structural dependency.
2. Risk Transfer
When AI reduces regulatory, safety, or compliance risk, it changes insurance profiles, audit outcomes, and institutional behavior.
3. Human-AI Collaboration, Not Autonomy
I hear a lot about how AI will replace jobs, and unemployment levels are about to increase dramatically. There may be pockets of the economy where AI will replace human workers, but some of the largest opportunities won't be fully autonomous systems.
In mission-critical operations, AI and humans will need to collaborate.
These are supervised intelligence platforms that surface risk, triage complexity, escalate exceptions, and improve decision quality.
In regulated environments, AI that augments judgment is far more scalable than AI that attempts to replace it.
What I'm Less Excited About
Not all AI opportunity fits this thesis.
I’m less excited by:
AI will continue to improve productivity broadly, but productivity tools are not infrastructure. And infrastructure is where I see durability compounding.
The Opportunity Window
We’re still early. AI capability has crossed a reliability threshold. Regulatory complexity is increasing. Infrastructure modernization cycles are accelerating. Penetration across soft and hard asset categories remains low.
That combination—improving capability + rising consequence + low penetration—creates asymmetry. The companies that will win this cycle will embed early, earn trust, and become operationally indispensable before the surface layer commoditizes.
This isn’t just about chasing hype cycles. It’s about identifying systems that customers cannot afford to operate without five, ten years from now—and building toward that outcome today.
Frequently Asked Questions
1. What makes AI defensible in regulated industries?
AI is defensible in regulated industries when it is embedded into mission-critical workflows that require compliance, auditability, and oversight. Durability comes from reducing regulatory risk and becoming operationally indispensable. Once AI systems are trusted inside compliance-heavy environments, switching costs compound over time.
2. Why is AI in infrastructure different from horizontal AI tools?
AI in infrastructure and physical operations must prioritize reliability, integration, and explainability over surface-level features. In utilities, transportation, and facilities systems, downtime has real economic and safety consequences. The most durable platforms become part of the operational control layer.
3. Will AI replace workers in mission-critical environments?
In regulated and high-stakes environments, AI is more likely to augment human judgment than replace it. The most scalable systems surface risk, triage complexity, and escalate exceptions under supervision. Human-AI collaboration is typically more durable than full automation in compliance-driven and consequence-heavy operations.