[Experimental UI] Requests take 15–50 min + Tasks panel accumulates 100+ stuck "Running" entries that reappear after close
Environment
- Client: Claude Code web version (claude.ai/code)
- UI mode: New experimental interface (switched after it was introduced; have not reverted to old UI)
- Date: April 2026, since experimental UI became available
---
Problem 1 — Requests processing for 15–50 minutes
After switching to the experimental interface, individual requests that I would not consider complex for an AI of this capability began taking 15 to 50 minutes to complete.
These are standard software-engineering tasks: reading files, making targeted edits, running shell commands, committing — not fundamentally different from what the same session was doing before the UI switch.
Expected: similar latency to pre-experimental-UI sessions (seconds to a few minutes for comparable tasks).
Actual: 15–50 min (different, incomprehensible delays and suspensions are noticeable)
I have not been able to rule out whether this is:
- A server-side regression introduced alongside the experimental UI rollout, or
- Something about the experimental UI's request/stream handling that adds overhead.
---
Problem 2 — Tasks panel accumulates 100+ stuck "Running" entries
In the Tasks panel of the experimental UI, background tasks accumulate as "Running" and never clear themselves.
Current state: I have 100+ entries showing as Running.
Behavior observed:
- Closing the browser tab / session does not remove them — they reappear when the session is reopened.
- There is no "clear all" or "cancel all" action visible in the UI.
- It is unclear whether these stuck entries consume server-side resources, affect request prioritization, or contribute to the slow response times described in Problem 1.
Steps to reproduce:
- Use the experimental Claude Code web UI for a session with background Bash commands (
run_in_background: true). - Close and reopen the session.
- Observe Tasks panel — previous "Running" entries are still listed.
- Repeat across sessions → count grows unboundedly.
---
Questions / requests
- Are stuck Running tasks consuming server resources or affecting throughput? If yes, this could partially explain the 15–50 min slowdowns.
- Is there a way to bulk-clear the Tasks panel? If so, it is not discoverable in the current UI.
- Is the 15–50 min latency a known regression in the experimental UI? If it is a server-side issue unrelated to the UI, knowing that would help narrow down the root cause.
- Plan: add a "Cancel all / Clear completed" action to the Tasks panel to prevent unbounded accumulation.
---
Additional context
I did not test whether these problems reproduce in the old (non-experimental) UI — I switched to the experimental UI and stayed there.
This issue has 1 comment on GitHub. Read the full discussion on GitHub ↗