TOPIC • 6 CLAIMS
AI Disruption
AI as an operating reality for businesses, distinct from AI as a trade.
HOWARD MARKS/
Direct lending carries far heavier software exposure than the public credit markets do, which turns a sector-specific AI shock into a portfolio-level one.
- software is 4-5% of high yield bonds
- 10-15% of broadly syndicated loans
- 20-30% of direct lending
HOWARD MARKS/
Adoption reached a scale in two years that most technologies need a decade for.
- roughly 400 million individual users
- 75-80% of companies, as of February 2026
HOWARD MARKS/
He tracks capability by year to argue the rate of change itself is the thing being mispriced.
- unable to do basic arithmetic in 2022
- passing the bar exam in 2023
- writing working software in 2024
- completing tasks autonomously by 2026
HOWARD MARKS/
He cites recursive self-improvement as having arrived in a shipped product.
- GPT-5.3-Codex described by OpenAI as instrumental in creating itself
HOWARD MARKS/
He points to physical-world deployment as evidence this is not confined to demos.
- Waymo handles roughly one-fifth of San Francisco taxi trips
HOWARD MARKS/
He frames the economic transfer in terms of labor value migrating to compute.
- $150-250 billion in annual labor value migrating to AI compute, in software alone
Claims are restated; figures are the authors' own. Each links to the document it came from, which in turn links to the original in full.