Excess Returns
When Agents Replace Labor | $11 Billion Tech Manager on What Investors Miss About AI
Most Important Insight
The transition from AI as a productivity tool to autonomous agents will decouple software value from human headcount, rendering the traditional per-seat SaaS licensing model obsolete.
Most Original Insight
AI agents will trigger a massive deflationary shock in white-collar labor costs that will fundamentally alter the long-term neutral interest rate by 2027.
Key Points
- AI is shifting from 'copilots' that assist humans to 'agents' that autonomously execute complex multi-step workflows across different software interfaces.
- The 'Inference Era' has arrived, where the recurring cost of running models at scale will dwarf the initial capital expenditure required for training.
- SaaS companies relying on seat-based pricing face an existential threat as their customers' headcounts shrink due to agentic automation.
- Proprietary, non-public data sets are the only sustainable moats as foundational models become increasingly commoditized through open-source competition.
- Physical constraints, specifically electrical grid capacity and data center power density, are now the primary bottlenecks for AI scaling through 2028.
- Vertical AI applications tailored for specific industries like legal or healthcare will capture more value than horizontal, general-purpose LLMs.
- The velocity of software development is increasing so rapidly that the half-life of code is dropping, favoring companies with the fastest iteration cycles.
- Enterprise spend is shifting from 'software as a tool' to 'software as a service' in the literal sense, where companies pay for outcomes rather than access.
Investment Implications
| Asset / Sector / Instrument | Action | Source | Notes |
|---|---|---|---|
| Electrical Utilities & Grid Infrastructure | BUY | explicit | Power availability is the 'hard ceiling' for AI growth, making energy providers the ultimate gatekeepers of the compute cycle. |
| Vertical AI Startups (Outcome-based pricing) | BUY | explicit | Companies that charge per 'task completed' by an agent are positioned to capture the value currently going to human labor. |
| Proprietary Data Owners (e.g., RELX, Thomson Reuters) | BUY | explicit | High-quality, specialized data is the essential fuel for agents to perform professional-grade tasks without hallucination. |
| NVIDIA (NVDA) | HOLD | implicit | While dominant in training, the shift to inference may invite competition from custom ASICs and specialized edge-computing chips. |
| Legacy SaaS (e.g., CRM, NOW) | SELL | implicit | Exposure to seat-count contraction as agents replace the human users these platforms were designed to support. |
Hang on a sec…
- The claim that seat-based pricing will 'collapse' within 24 months ignores the friction of enterprise procurement and the multi-year nature of existing software contracts.
- The speaker suggests agents will 'replace' entire departments, yet fails to address the significant legal and regulatory liability hurdles of removing 'humans-in-the-loop' for professional services.
- The assertion that power is the 'only' bottleneck overlooks the potential for algorithmic breakthroughs in small language models (SLMs) that could drastically reduce energy requirements.