The Changelog: Software Development, Open Source Exploring with agents (Interview)
36 snips
Apr 24, 2026 
Amelia Wattenberger, designer and data-viz veteran (ex-GitHub Next, now designing Intent at Augment Code). She argues agents make prototyping faster but finishing the last 20–30% much harder. Conversation covers the arc from autocomplete to chat to CLI back to UI. She explains why Intent uses workspaces as the core primitive, tradeoffs of one-worktree-per-task versus per-agent, and new agent orchestration patterns.
AI Snips
Chapters
Transcript
Episode notes
Copilot Initially Dismissed Then Surprised Users
- Early Copilot reactions dismissed it as better autocomplete, but users discovered qualitatively different effects after day-one use.
- Amelia recalls internal and external skepticism flipping to surprise when people realized the tool's practical differences.
Agents Shift Effort From Prototyping To Polishing
- Agents compress early-stage work so a prompt can produce ~70% of a feature very quickly.
- Amelia observed that this squeezes the initial work into a small fraction of time while the final 30% of polish becomes disproportionately hard and time consuming.
Personalize Docs By Indexing Repos And User Context
- Use RAG and personalized indexing to make docs interactive and tailored to a developer's background.
- At GitHub Next Amelia built doc-pilot ideas that index repos and personalize walkthroughs based on a user's GitHub history.