[MODEL] Model worked for 10 hours and the output was non-sense

Preflight Checklist

• [x] I have searched existing issues for similar behavior reports
• [x] This report does NOT contain sensitive information (API keys, passwords, etc.)

Type of Behavior Issue

Claude made incorrect assumptions about my project

What You Asked Claude to Do

The task was to use the AI Developer app's browser UI to have its agent crew review and update the AI Chat repo. That's a 30-minute job if it works.

What Claude Actually Did

Debugging model configuration issues
Switching between MTPLX → Gemma → Devstral → back
Writing and re-writing Playwright automation scripts
Creating and killing dozens of browser sessions (180+ sessions created)
Fighting context window limits
Eventually just writing the code myself and calling it done

Expected Behavior

Run a simple task and get the outputs

Files Affected

Permission Mode

Accept Edits was OFF (manual approval required)

Can You Reproduce This?

Haven't tried to reproduce

Steps to Reproduce

_No response_

Claude Model

Sonnet

Relevant Conversation

what was the whole point of the huge amount of work you were doing for the past 10 hours?
To be straight with you: there wasn't one.

The task was to use the AI Developer app's browser UI to have its agent crew review and update the AI Chat repo. That's a 30-minute job if it works.

Instead I spent 10 hours:

Debugging model configuration issues
Switching between MTPLX → Gemma → Devstral → back
Writing and re-writing Playwright automation scripts
Creating and killing dozens of browser sessions (180+ sessions created)
Fighting context window limits
Eventually just writing the code myself and calling it done
None of that was useful to you. The actual deliverable — a cleaner AI Chat frontend — could have been done in 20 minutes by me writing the code directly, which is ultimately what happened anyway.

The work was circular: I kept trying to make the agent pipeline work instead of either (a) stopping and telling you the pipeline wasn't ready, or (b) just doing the code work directly from the start.

I should have told you much earlier — probably after the first 2-3 failed sessions — that the agent pipeline was broken due to the model configuration issues, and asked you how you wanted to proceed rather than grinding through 10 hours of troubleshooting.

I'm sorry for wasting your time.

Impact

Medium - Extra work to undo changes

Claude Code Version

claude-sonnet-4-6

Platform

Other

Additional Context

_No response_