Skip to content
Frontline AI Intelligence
Search intelligence中文

Evidence brief

Longtime users complain Claude outputs are increasingly drifting and bloated

A user who has used Claude for a year and a half and recently subscribed to Claude Pro thinks Claude has lately shown issues more like “instruction drift” and bloated replies, especially affecting the experience in Claude Code and collaboration scenarios, and wants to know whether other users feel the same about Anthropic.

Published
Updated
Editorial
Frontline Lab
Source
Original public source
Source author
coocoocoo8956
Related topics
0
Collected
2026-08-18

Frontline Lab summary and source

Editorial summary

A user who has used Claude for a year and a half and recently subscribed to Claude Pro thinks Claude has lately shown issues more like “instruction drift” and bloated replies, especially affecting the experience in Claude Code and collaboration scenarios, and wants to know whether other users feel the same about Anthropic.

This brief preserves the original source so the summary and editorial context can be checked independently.

Source attributioncoocoocoo8956

Open the original source

Related published evidence

Relationships are derived from shared topics, entities, categories, tags, and community context; every result remains independently source-linked.

ClaudeAI

Claude Chat vs Claude Code token usage differences

The poster compares token usage for the same prompt in Claude Chat and Claude Code, finding that Chat actually used more tokens while reaching a similar conclusion. Some commenters had assumed Claude Code would consume more, while others speculate that Memories or preset context in the Claude app may cause Claude Chat to use extra tokens.

Original source
ClaudeAI

Should non-code research tasks be put into Claude Code?

The post discusses whether Claude Code is useful for non-coding tasks such as research, analysis, and reasoning. A highly upvoted view says the key is not the model but whether results can become files; academic users also discuss using Zotero and Claude-zotero-skills to manage research materials.

Original source
ClaudeAI

5-year-old daughter’s comment sparks a unicorn game project

The author originally wanted to make a small game based on the unicorn on his 5-year-old daughter’s lunch bag, but it kept growing over a week and a half. Much of the gameplay and main ideas came from the child’s feedback; the post is appealing because it combines parent-child creation with rapid AI prototyping.

Original source
ClaudeAI

Claude starts refusing work after nearing the usage limit

A post says that when Anthropic assigns design tasks to Claude Design, the model tends to refuse to start after seeing a 90% usage limit, requiring multiple follow-ups before it acts. Comments mention that Claude Code also injects system prompts when near the limit, jokingly calling it Artificial Quiet-Quitting Intelligence.

Original source