Maybe one of the only moats in 2026 is the context layer. AI improvements mean: ✅ UI/UX might simplify and consolidate. Instead of a lot of …

Maybe one of the only moats in 2026 is the context layer. AI improvements mean: ✅ UI/UX might simplify and consolidate. Instead of a lot of fancy buttons/knobs, you need simple, clean interfaces where agents can do an e2e task, and you can see the outputs. ✅ Agent abstractions are solidifying, and there’s no need to constantly reinvent the harness layer. Though there is still value in deterministic code. ✅ Users are programming increasingly in English instead of code. What’s not clear: ❓ What the tool layer looks like. AI can vibe-code software extremely quickly, but it’s inefficient to code everything from scratch - clearly it still imports libraries and uses MCP tools. ❓ Related to the above, whether you need a lot of targeted tools and subagents, or agents just need a few tools (sandbox, web search, skills files), and can do everything else. ❓ Whether SaaS companies can monetize with e2e agents What is clear though is that every agent does need context, and some ways to read and operate over that context (in 2023 it was naive RAG, in 2026 it’s file sandboxes). This includes everything from systems of record, to web context, to document context (us). I talk about this and other hot takes in this @VentureBeat podcast. I am sure 50% of my takes will be wrong within the next year, but I do think that my core assumptions about the importance of the context layer will continue to hold. Don't let Anthropic own your stack. @jerryjliu0 on why modular architecture is the only real hedge in the agent era. https://www.youtube.com/watch?v=HbXvX-KtkSs

Source: Jerry Liu (X) | 2026-05-09