Ben Thompson

Updated 2026-04-08

American tech analyst. Founder and author of Stratechery, one of the most influential paid newsletters in technology strategy and business analysis. Formerly worked at Microsoft and Apple, and has written independently from Taiwan since 2013.

The Stratechery Lens

Thompson analyzes technology companies primarily through strategy rather than product implementation. His best-known framework is Aggregation Theory: companies that control user access capture disproportionate value because suppliers must compete for access to that user base.

Another core concept is modular versus integrated. Profits tend to pool in integrated parts of a value chain, not in modularized, commoditized ones.

Agents Over Bubbles (March 2026)

In an influential essay, Thompson argues that the current AI wave is not just a bubble. His logic moves through three paradigms:

  1. ChatGPT in November 2022 made LLMs visible but still passive and error-prone
  2. Reasoning models in September 2024 made them more reliable and less dependent on manual guidance
  3. Agents in late 2025 turned them into systems that can work autonomously for hours

Why He Thinks This Is Not a Bubble

  • Agents multiply a single motivated user through compute-heavy workflows
  • Enterprise customers pay for productivity, and agents deliver it in a way that reduces coordination cost
  • Hyperscaler capex is not speculative in his reading, but a response to demand that already exceeds supply

Why Anthropic Is Not Being Commoditized

Thompson’s strongest AI claim is that agents require more than a model. They require a harness, the software layer that steers the model, verifies results, and connects it to tools. The differentiator is the integration of model plus harness, not the model alone.

Connections

  • Anthropic - his analysis supports the profitability case for Anthropic
  • Claude Code - key example of productive harness integration
  • OpenClaw - a good illustration of harness differentiation outside enterprise software
  • Vibe Coding - the paradigm becomes much stronger once agents can act instead of only respond

Sources