Most AI companies are renting their position. These 4 prompts tell you if yours is one of them.

Original article: Read on Nate's Substack

Published: March 19, 2026 | Processed: March 20, 2026

Summary

Main Thesis

Perplexity Computer — launched February 25, 2026 — is arguably the best agentic product shipped that month. It is also a cautionary tale. Perplexity's core reasoning runs on Anthropic's Claude, its deep research on Google's Gemini, its speed layer on Grok. Despite brilliant execution, Perplexity's structural position is fragile: it is renting its intelligence from the very companies building competing products. This is not just Perplexity's problem — it is the condition of almost every AI company that isn't Anthropic, Google, OpenAI, or Meta.

Key Data Points & Findings

Perplexity Computer: $200/month, routes work across 19 frontier models, persists for months, delivers finished artifacts asynchronously. Genuinely worth it for heavy research and ops workflows.
Perplexity's 2026 revenue target: $656 million, with major enterprise customers already running their search API in production.
The middleware trap: Most AI product companies depend on foundation model providers who are simultaneously competing with them. The hyperscalers' $690 billion infrastructure bet makes this asymmetry worse every month.
Anthropic's counter-move: The same week Perplexity launched Computer, Anthropic shipped the enterprise expansion of Claude Cowork — with deep connectors, private plugin marketplaces, and seamless context passing. One model. They own it.
Four positions that survive the middleware trap:
1. Data Moat — proprietary datasets the hyperscalers can't replicate
2. Distribution Lock-in — deep workflow integration that's expensive to rip out
3. Vertical Integration — owning the full stack in a specific domain
4. Infrastructure — being below the model layer, not above it

Practical Takeaways

Run a five-step diagnostic on your company, portfolio company, or client to determine if you're building a durable position or renting one from a competitor.
Ask: If Anthropic, Google, or OpenAI decided to compete directly with our product next quarter, what's left? If the honest answer is "not much", you're in the middleware trap.
Good execution on the wrong layer of the stack will not save you. Perplexity is proof that product excellence and structural fragility can coexist.
The four prompts in the article help you diagnose: (1) your dependency map, (2) your switching costs, (3) your moat surface area, and (4) where your value actually lives.

Frameworks

The Middleware Trap: Occupying the layer between foundation models and end users without owning either side
Structural position diagnostic: 5-step test to determine if your AI position is owned or rented
The Four Durable Positions: Data, Distribution, Vertical Integration, Infrastructure