Tons of new AI engineering models/products shipped and it feels like the rankings have fully reset again.
My current thoughts:
- The new Cursor is a lot better than claude code for planning, coding, reviewing, running shell commands, getting into a flow with the UX
- Claude code is sometimes better for things like exploration, using sub-agents
- Opus 4.5 is significantly better than everything else, but it’s slow. Best for initial planning and review, then use Composer to crush edits and execution
- Exa MCP auto-connects your agent to whatever technical docs/research capabilities you need
- Markdown is more important than code