← Home
The Changelog · April 24, 2026 · 1h 36m 52s

Exploring with agents

Amelia Wattenberger, a designer with experience at GitHub Next and now at Augment Code, explores the impact of AI coding agents on software development. She argues that while agents make prototyping dramatically faster, the hardest part—the last 30% of a project—becomes even more challenging. The conversation covers the developer identity crisis this shift causes, the redesign of tools for an agent-first world, and the evolution of dev tool interfaces from autocomplete to chat to CLI and back to UI. Amelia explains why Intent uses the workspace as the core primitive rather than a chat thread and delves into the tradeoffs of per-agent versus per-task worktree strategies.

This summary was generated from show notes and public descriptions, not from a full transcript review. Details may contain inaccuracies.

Novel

Amelia argues that while AI agents accelerate the early stages of building, the final 30%—polishing, integrating, shipping—becomes significantly harder.

Highlights

Developer identity crisis in the agent era
Developers are grappling with a fundamental shift in what it means to write software when AI agents are increasingly doing the actual typing and implementation.
The arc from autocomplete to chat to CLI back to UI
Amelia traces the evolution of developer tooling interfaces: from simple autocomplete, to conversational chat, to command-line interfaces, and now returning to graphical UIs in the agent era.
Workspace as the core primitive, not the chat thread
Intent, the tool Amelia designs, treats the workspace—the full project context—as the fundamental unit of interaction, rather than a linear chat thread.
One-worktree-per-agent vs. one-worktree-per-task tradeoffs
In agent workflows that modify code, there is a choice: give each agent a dedicated worktree (isolated branch) or create a new worktree per task to keep changes scoped.
Was this useful?