Here's a summary of all user-facing changes on Superconductor from August 23-29:
This Week's Changes
Refreshed agent model choices
Pi now includes GLM 5.3 Flash with Max reasoning effort, giving teams another GLM 5.3 option when comparing agent quality, speed, and cost.
Also, Superconductor's agent profile lists now put newer, stronger model choices front and center. Profile pickers and automatic defaults emphasize current Pi, OpenCode, Grok Build, and Cursor options. Older models remain available for existing work but no longer appear in the agent selector.
QA Check improvements
For projects with the QA Checks Beta enabled, you can now choose which agent profile runs QA checks. Leave the setting on Best available when you want Superconductor to choose automatically, or pick a specific runnable profile from Project Settings when a project needs QA checks from a particular agent.
QA Checks also now prioritize fewer, higher-signal test recommendations. Reports should be less likely to treat "add tests" as a default requirement when the change does not need extra coverage.
Recent Activity has a ticket shortcut
Recent implementation cards in your activity sidebar now include View ticket in the overflow menu, so you can jump from a specific implementation back to the parent ticket and compare other implementations faster.
Update to workspace role naming
The workspace role previously named Developer is now Collaborator across the product. Permissions for that role did not change.
Email assistant addresses are easier to identify [Beta]
For teams using Email Assistant [Beta], project signal email addresses now include a project-name slug, so it is easier to tell which address belongs to which project when you are juggling multiple projects.
Bug Fixes
Slack-launched agent failures no longer look stuck
When a coding agent launched from Slack fails, Superconductor now replies in the originating Slack thread with the error and clears the working status. Previously, some failed runs left the thread showing that Superconductor was still working indefinitely.
Typing stays responsive in long conversations
We fixed typing lag that could show up when composing messages in very long conversation threads.
ChatGPT usage windows display correctly
In response to OpenAI's recent ChatGPT Plus limit update, ChatGPT plan usage stats now show the windows returned by OpenAI with accurate labels, including Plus-style 5-hour and 7-day windows and Pro-style 7-day windows.
Unsupported ChatGPT model errors are clearer
If a selected Codex model is not available on your ChatGPT plan, Superconductor now shows a clear plan-support message instead of a raw provider error.
Implementation headers wrap on narrower layouts
Implementation page headers now wrap their title and action buttons on narrower layouts instead of letting long titles or disabled preview controls collide.
Multi-profile selections fit in the composer
When multiple agent profiles are selected, the composer now uses the compact profile label instead of trying to fit several long model names next to the Create Ticket button.
Live steering messages stay in the right order
When you send follow-up instructions to a running Claude Code agent, the live work log now stays below the newest instruction the agent has seen. Previously, the work log could appear above a follow-up message even though the agent had already incorporated it.
Have feedback or ideas for things you'd like to see on Superconductor? Let us know!