AI Agents That 'Just Write Code' Force a UI Rethink

Key Info

The post highlights a growing tension in AI coding agents: as agents become capable of simply writing code, users vary widely in how much they care about oversight and permissions. The author notes that opencode2's codemode was built to show useful UI and still prompt for permissions even when the model returns a block of code, but their AI (Astra) now writes raw Python to do everything, which means the UI needs to be reworked.

Highlights

• User attitudes toward agent visibility and permissions exist on a spectrum, and the author suspects more users will stop caring about these safeguards entirely.
• opencode2's codemode was designed to keep a decent UI and permission prompts even when the model hands back a hunk of code.
• Because Astra solves tasks by writing raw Python, the team must rethink the UI to match that behavior.