Limitations and FAQ

What Architect doesn't do yet, and answers to common questions.

Known limitations

  • No scheduled or background work yet. Architect only works while you’re in a conversation with it. It doesn’t yet scan agents on a schedule, send reports, or open proposals on its own. Scheduled automations are coming soon. See Proactive proposals for the current path from Spotlight to a proposal.
  • It can’t review merge proposals. Architect can open a proposal and suggest reviewers, but approving, commenting, merging, and closing are done by people.
  • Some resources are read-only to Architect. It can’t manage MCP servers, secrets, integrations, channels, alerts, or phone number assignments. See What Architect can’t do.
  • Dashboards aren’t saved. A dashboard built in the chat keeps its data only in the browser tab where it was built.
  • Chats are private. You can’t share an Architect chat with a teammate. Share the result instead, such as a merge proposal or a triage ticket comment.

FAQ

Spotlight observes and recommends: it monitors your agent’s conversations and surfaces summaries, suggested investigations, alerts, and recommendations. Architect builds and fixes: it investigates a question or a Spotlight finding, changes the agent, tests the change, and prepares it for your review. Spotlight cards hand off to Architect with Analyze with Architect and Investigate with Architect.

Not through the agent’s configuration. Architect stages configuration changes in your draft, and only you can publish a draft. In the default Approval required mode, it also asks before actions that affect live traffic or shared resources, such as merging a branch, changing the traffic split, or updating a tool. In Auto-approve mode it doesn’t ask before those actions, so use that mode only for work you are confident about. See Permissions, approvals, and drafts.

Open the agent and go to Version Control > Proposals. Architect also posts a link to the proposal in the conversation. The proposal lists you as its author, because Architect acts as you.

Architect acts as you, so you are the proposal’s author, and authors can’t review their own proposals. A teammate with editor access has to approve it. Workspace admins can merge without an approval.

Yes. There’s no separate publish step after a merge. The merged version serves live traffic immediately for whatever share of traffic the target branch receives. See What happens on merge.

Yes. It works on the agent you have open by default, but it can read and edit any agent you have access to. Use the workspace Architect page (Agents > Architect) for questions that span agents.

No. Architect runs as you, with your permissions. If you can’t edit an agent or merge into a protected branch, neither can Architect.

Architect writes tests and simulations and runs them in the conversation as part of building a change, before it asks you to publish. To be certain a fix is verified, ask Architect to show the new tests failing without the change and passing with it. Results also appear on the proposal’s Test runs tab.

Yes. In an open chat with an empty composer, select Talk to agent to start a voice conversation. You can also dictate messages in any composer by holding Cmd/ Ctrl+..

No, but they call the same APIs. Architect runs in the dashboard, stages changes in drafts, and uses its own approval modes. The MCP server lets external assistants such as Claude and Cursor manage your agents, applies changes directly, and relies on your client’s tool approvals. See Claude, Cursor, and other AI assistants.

Architect is free during the alpha, which runs through October 2026. Usage doesn’t count against your workspace’s agent minutes or LLM cost. Check the pricing page for terms after the alpha ends.

Rate a response with the thumbs up or down under it, use Feedback in the sidebar chat header to rate the session, or type /feedback in the sidebar chat.