Essays - Benedict Evans · Benedict Evans
articleinsightAI, networks and Mechanical Turks
23 November 2025Strategy
Source excerpt
About Essays - Benedict EvansHow far do LLMs give us a step change in how good a search and recommendation system can be? Do they let you build one without needing a vast user base of your own?
This is a limited feed-provided excerpt, not the full original work.
Apollo layer
What a founder can learn
Founder takeaway
Don’t assume LLMs eliminate the scale advantage in search or recommendations. Test whether model intelligence can overcome cold-start constraints—or whether relevance still depends on proprietary user behavior and feedback loops.
Why it matters
The answer determines whether you can enter with a model-led product advantage or must also build a defensible system for collecting data, learning from interactions, and compounding relevance.
Put it to work
For your product, list the inputs needed to produce a strong first recommendation and those needed to improve it over time. Which can an LLM supply, and which require usage data or network scale?