
Field Notes: The Multimodal Gap in AI Coding Agents
Some AI coding agents can generate images on demand. Others can't. This seemingly minor capability gap reveals a deeper strategic divide in how these tools are evolving.
Cursor, Windsurf, Copilot, and AI-powered development.
Powered by Claude Opus 4.5—understands meaning, not just keywords. Try “how do I configure Claude Code?”
No posts published in the last 14 days.

Some AI coding agents can generate images on demand. Others can't. This seemingly minor capability gap reveals a deeper strategic divide in how these tools are evolving.

Google released Gemini 3 Flash today—Pro-level performance at Flash-level speed for a quarter of the cost. As I dictate this with four Opus 4.5 agents running in the background, I can't stop thinking about what this means for vibe coding.

Cursor's multi-agent started as best-of-N: the same prompt raced N times, and you pick a winner. In 2026 Cursor shipped real orchestration. What changed, and which one you should reach for.
Get practical AI insights delivered to your inbox or schedule a consultation to discuss your AI strategy.