April 15, 2026
In this edition
Memory & Long-Conversation Behavior
- Improved long-term context for Nova: new context compaction and sliding-window summarization reduce token usage, prevent "Agentic Amnesia," and help the assistant retain relevant information across long conversations.
Agent Controls & Safety
- Explicit approval flow for tool capabilities and background dispatches: users must confirm actions that modify data, giving clearer control over autonomous operations.
- New AI & Automation workspace setting to opt in to autonomous AI data mutations, with a prominent visual safety warning to communicate associated risks.
Agent Chat & Model Selection
- Model selection in Agent Chat: choose the assistant's intelligence level with a dedicated selector (e.g., Haiku, Sonnet, Opus) and see the active choice clearly indicated to control response style and behavior.
- Improved handling of long messages to prevent truncation and preserve conversation continuity.
Tools, Feedback & Visibility
- Faster, more discoverable workspace tooling: a capability-discovery tool surfaces available actions and a feedback-search tool makes past workspace feedback easier to find.
- Background tool activity is now visible in the chat timeline via fallback visualization chips, making ongoing processes and results clearer to users.
- Optimized client-side fetching in chat for snappier interactions and reduced perceived latency.
Integrations & Reliability
- More reliable Jira synchronization and reduced tracker errors; older installations will be prompted for any required permissions to maintain healthy issue tracking.
Stability & Error Handling
- Improved error handling across chat integrations to reduce silent failures and ensure more reliable behavior during operations such as thread deletion.