Codex App: improve the Subagents panel with live activity and per-agent configuration

Open 💬 2 comments Opened Jul 31, 2026 by andrewylies
💡 Likely answer: A maintainer (github-actions[bot], contributor) responded on this thread — see the highlighted reply below.

What variant of Codex are you using?

App

What feature would you like to see?

Summary

The existing Subagents panel already shows which subagents were created, their status, and a brief description of their assigned work.

As GPT models make more active use of multiple subagents, I would like this panel to provide a clearer, more timely overview of how the delegated work is progressing.

Current limitation

When multiple subagents are active, the existing information is not always enough to understand the parallel workflow at a glance.

It is difficult to quickly compare:

  • the current focus or latest meaningful activity of each agent
  • how each delegated task is progressing
  • which model and reasoning configuration each agent is using

Proposed improvements

Enhance the existing Subagents panel with:

  • a concise, live activity summary for each active agent
  • the current stage of work, such as exploring, reviewing, implementing, or validating
  • the latest meaningful progress update and elapsed time
  • each agent’s role, resolved model, and reasoning effort
  • an expandable detail view containing the assigned task, visible activity history, progress updates, and final result

The goal is to make parallel execution easier to understand, rather than to require users to manually manage or control individual subagents.

This does not need to expose private chain-of-thought reasoning. Concise summaries based on user-visible operational activity would be sufficient.

Why this matters

As Codex delegates more work to subagents, the quality of the overall experience increasingly depends on making that delegation understandable.

Improved visibility would help users:

  • understand how Codex divided the work
  • follow multiple workstreams without opening every agent individually
  • see how the overall task is progressing
  • verify that the intended model and reasoning configuration was applied
  • better understand and trust long-running multi-agent workflows

Additional information

Related issues

  • #32283 covers per-agent model and reasoning-effort visibility, but not the broader live activity overview.
  • #22099 and #12047 explore related monitoring concepts for the CLI/TUI, while this request targets the Codex App.

View original on GitHub ↗

2 Comments

github-actions[bot] contributor · 28 days ago

Potential duplicates detected. Please review them and close your issue if it is a duplicate.

  • #35915

Powered by Codex Action

andrewylies · 28 days ago

https://github.com/openai/codex/issues/35915 focuses on exposing detailed command and tool activity for an active task in the Desktop app, comparable to the mobile remote view.

This issue specifically focuses on the existing Subagents panel: providing an at-a-glance view across multiple parallel subagents, including each agent’s latest meaningful activity, role, resolved model, and reasoning effort.

The two requests could share some underlying activity data, but the intended UI and multi-agent use case are distinct.