The modern AI stack has three layers.
Models. OpenAI, Anthropic, Google. Billions in capex, rapidly commoditising. Indie devs can't compete here.
Harnesses. Claude Code, Cursor, Copilot, custom chat UIs. The runtime that calls models, manages conversation, executes tools. Commoditising fast. Indie devs probably can't compete here either.
Context. Domain knowledge, structured data, business rules, accumulated expertise. The stuff that turns a generic AI into a domain AI expert. This is where indie devs build.
Context is the new SaaS layer.
Models get cheaper every quarter, harnesses get replaced every year, but a well-built context layer compounds.
Store briefs, pricing playbooks, supplier notes, audit history. Each entry makes the next agent smarter, and none of it is locked to a specific runtime.
So pick a vertical (retail, legal, healthcare, music production) and build the context infrastructure that makes any harness an expert in that domain.
Files the user owns. Indexed for search. Versioned with git. Exposed through MCP, so it works with Claude Code today and whatever wins tomorrow.
Don't build the AI. Build what makes the AI worth using.
Holy shit, this is BRILLIANT: This dude breaks down why everyone thinks the whole WHCD event was fake... everything about trump is FAKE.
Best video you'll see today.