Feedback from a former Atlas user: Chat vs. Work and browser integration

Start by reproducing the Chat and Work flows in the Codex App, using the reported Feedback ID and the Home Assistant navigation scenario. Measure the reported navigation and screen-analysis delays, then define completion as a seamless workflow with lower latency and browser responsiveness comparable to Atlas.

What version of the Codex App are you using (From “About Codex” dialog)?

26.715.70719

What subscription do you have?

ChatGPT Plus

What platform is your computer?

Darwin 25.5.0 arm64 arm

What issue are you seeing?

I've been testing the new desktop app after using Atlas extensively for projects like Home Assistant administration.

The new browser is promising, and I confirmed that Work mode can eventually see and interact with my Home Assistant web interface. However, the overall workflow currently feels like a step backward in usability.

Observations:

• Switching between Chat and Work feels artificial. As a user, I don't want to decide which mode I'm in—I want to describe the task. If I say "Help me edit a spreadsheet" or "Open Home Assistant and fix this automation," the app should automatically transition into the appropriate execution environment when needed.

• In my testing, asking Work to open Home Assistant, navigate to Settings → Automations, and locate a specific automation took 3 minutes and 18 seconds (excluding my login time). Doing the same manually in Atlas took only a few seconds because I already had bookmarks and knew where I was going.

• Once Work reached the page, it was able to understand what it was seeing, which is impressive. However, even asking a simple question like "Can you see this screen?" took about 35 seconds to answer. For interactive troubleshooting, that level of latency makes the experience feel cumbersome and this was not my experience in Atlas as it was much faster.

Another concern is efficiency. If each browser interaction requires significant "thinking" time, that likely increases computational cost for OpenAI while also consuming more electricity and computing resources than a more streamlined approach would. Faster execution would improve the user experience while potentially reducing operating costs and environmental impact.

The capability is clearly there, and it's exciting. I think the biggest opportunity is to make it feel more seamless:

• One assistant, not separate Chat and Work personalities.
• Automatic transition into browser/computer tools when a task requires them.
• Much lower latency for navigation and screen analysis.
• Browser performance that matches or exceeds the responsiveness Atlas provided.

I think this could become an outstanding workflow once the interaction speed and user experience catch up with the underlying capabilities.

What steps can reproduce the bug?

Feedback ID: no-active-thread-019f8608-3875-7431-b22a-56a89f642c4b

What is the expected behavior?

N/A

Additional information

N/A