Hi,
Most conversations about AI still sound like 2023 - Faster models, bigger data centers and better chatbots.
That was phase one.
What I think we’re entering now is phase two — and it may reward very different companies.
Not AI that helps you think. AI that does the work. That shift sounds subtle.
It isn’t.
From Assistance to Execution
The first wave of AI improved productivity at the margin. The software assisted.
In the agentic era, the software executes. Imagine systems that:
• Identify prospects
• Send outreach
• Follow up automatically
• Adjust pricing dynamically
• Reorder inventory
• Resolve support tickets
• Allocate marketing budgets
Without waiting for you. That’s not a feature upgrade. That’s a structural shift in how labor is allocated and when labor allocation changes, capital flows change.
Why This Changes the Economics
The last 15 years rewarded SaaS companies that scaled with headcount. More employees → more seats → more subscriptions → higher ARR.
But what happens if AI reduces required headcount? If a company needs 30% fewer employees to produce the same output, seat-based revenue models come under pressure.
In the SaaS era, software empowered labor. In the agentic era, software may replace portions of labor.
Very different margin story.
The winners may not be the companies selling more licenses. They may be the ones replacing them.
I think about this in two layers.
Layer 1: Agentic Software
Digital systems that move from “copilot” to “operator.”
Monetization may shift from:
• Per seat
To:
• Per task
• Per workflow
• Per outcome
If pricing aligns with completed work, not employee count, revenue becomes tied to output — not staffing levels. That’s a structural difference.
Layer 2: Physical & Embodied AI
Where software meets the real world:
• Warehouse automation
• Autonomous logistics
• AI-directed manufacturing
• Robotics in construction and defense
Here, margins expand through:
• Labor compression
• Throughput increases
• Fewer operational bottlenecks
It’s slower, more capital intensive but potentially more durable.
Physical productivity gains tend to compound quietly for years.
Where Profits Often Accrue First
Cloud captured value before SaaS and Semiconductors captured value before apps.
In AI, the stack roughly looks like:
1. Chips & compute
2. Model providers
3. Enterprise integrations
4. Full workflow automation
Applications often look exciting early. Infrastructure often compounds first. That doesn’t mean blindly chasing hardware.
It means understanding which layer captures pricing power.
The Filter I’m Using
Instead of asking, “Does this company use AI?”
I’m asking:
• If headcount shrinks, does this company benefit or suffer?
• Does it monetize seats or outcomes?
• Does AI widen its moat or compress it?
• Does automation strengthen pricing power — or commoditize the product?
• If labor costs fall 20%, who captures the savings?
If you can’t answer those, you’re speculating — not investing.
Markets rarely price structural shifts correctly at first. They extrapolate the last cycle forward. That’s where mispricing lives.
Portfolio Reality
This is not a 20% portfolio allocation position now. It’s a thematic tilt. I believe there would be better entry opportunities into agentic AI and robotics companies
Your base still compounds through:
• Broad market ETFs
• Tax-advantaged accounts
• Systematic contributions
Themes amplify. Structure protects.
The Real Question
The biggest mistake in new cycles is assuming the last winners automatically dominate the next one.
Sometimes they do. Often, the layer where profits accumulate shifts underneath.
The first AI cycle rewarded AI infrastructure and tools that made humans faster. This one may reward systems that replace tasks entirely.
Different layer. Different economics. Potentially different winners.
But the edge isn’t predicting headlines.
It’s asking:
Where does margin expand?
Where does labor compress?
Where does pricing power strengthen?
That’s the layer that compounds.
And compounding is the only story that matters.
— Ben