Set the standard for your agents with Rubrics, Keep track of usage across a workspace

This week helps you set standards for agent work, maintain wiki knowledge, and review usage and delivery.

Set the standard for your agents with Rubrics

Reconstructed demo of rubric criteria and a judged run, using fictional example content.

Rubrics help you define what good work looks like. Create a rubric from an existing agent, based on the behavior and standards you expect from your agent's output.

Maintain your workspace wiki Beta

Illustrated Autowiki demo, using fictional example content.

Autowiki is a new way for your agents to build and reference key information about your team and your work. Agents automatically write to the wiki to preserve useful context beyond a thread. Your agent can consult that reference instead of reconstructing the history each time. Workspace owners can start an Autowiki for a given workspace, and then set their agent to dream on a daily schedule. Learn more.

Keep track of your team's use

Reconstructed demo of workspace Usage, with fictional agents and costs.

You can now review usage for an individual workspace. Choose an available date range to inspect recorded totals, a daily chart and agent-level costs.

For scheduled work, open standard schedule or Live run history and look for Not sent beside a completed run. Hover or tap the warning to read the available delivery issue. Work completion and delivery are separate: the absence of a warning does not prove a result arrived.

A clearer view of what your agent is working on

Reconstructed demo of agent Work, Library and About, using fictional example content.

Finding an agent's work got a lot more straightforward. Open an agent to find its work, then switch between Work, Library and About. New indicators help you identify running or waiting work without opening several conversations.

Additional improvements

Models

GPT-6.1 Terra. You can choose GPT-6.1 Terra in the model selector for compatible conversations and agents.

Agents

Schedule coworkers. Review the proposed coworker grant on eligible schedule cards before saving. Not every card provides a complete named roster.
Drafts. Eligible personal drafts distinguish Create character options from Save this version, making alternatives and the final save easier to recognize.

Teams

Membership. Removal confirmations can show an affected-agent count when a teammate’s personal connected accounts support Team agents, before you confirm removal.

Threads

Plan and Execute. Normal new requests start in Execute where the chooser is available; choosing Plan for one request does not carry forward.
Fork cleanup. Conversation deletion can show optional related forks and required dependencies separately. Changed lists or publication conflicts can require another review.
Archive recovery. Choose Undo after a successful menu archive in supported web thread lists and grids that provide restoration.
Team filters. Web Team Threads remembers supported sort, source, Archived and Starred preferences in the same browser, without cross-device synchronization.

Account

Profile photos. Upload, replace or reset your personal profile photo. Your Marketplace publisher photo remains a separate choice.

Integrations

Ask-first approvals. Eligible interactive MCP writes can continue after supported approval, while existing account permissions and action limits still apply.
Improved Slack delivery. Later turns avoid repeatedly sending to the same destination after a permanent refusal. This does not reroute results elsewhere.
Telegram replies. Supported text replies can recover when the original reply target disappears. Unreachable chats receive clearer handling within configured delivery.

Browser

Logins. Clear your own conversation-linked sign-ins from Browser logins even when that conversation is locked, without reopening its messages.

Media

Image handling. Detected unreadable images are omitted from supported model requests rather than treated as usable input. Supply a readable replacement when needed.