Please Do Not A/B Test My Workflow

backnotprop.com
March 14, 2026
3 min read
56/100
Summary
Anthropic is conducting A/B tests on Claude Code, which has impacted user workflows. The author expresses frustration over the degradation of their workflow due to these tests.
Key Takeaways
- Anthropic is conducting A/B tests on Claude Code that have negatively impacted user workflows, leading to frustration among users.
- Users of Claude Code, who pay $200/month, seek greater transparency and configurability in the tool's functionality.
- Changes to core features, such as plan mode, have been implemented without user notice, affecting the quality of outputs.
- The engineer responsible for the A/B test acknowledged that limiting plan lengths was intended to reduce rate-limit hits, but early results indicated minimal impact.
Coverage & discussions
Hacker News Β· backnotprop.com
Community discussions
Historical count: 5 Β· 5 saved in sample Β· Historical sample
Analysis ready
Community Sentiment
Positives
- LLMs can significantly enhance productivity by quickly generating boilerplate content, which can guide users in their thought processes more efficiently than traditional methods.
- The ability of newer LLM models to perform effectively with less direction indicates advancements in AI capabilities, potentially improving user experience.
Concerns
- LLMs currently lack reliability and replicability in results, raising concerns about their suitability for professional workflows, especially when subjected to A/B testing.
- The risk of LLMs producing misleading or irrelevant information can derail users, emphasizing the need for a solid understanding of the subject matter to utilize them effectively.
- Open source tools may struggle to compete with proprietary models like Claude Code due to their inability to conduct A/B testing, which limits data-driven design improvements.
Related Articles

Anthropic tries to hide Claude's AI actions. Devs hate it
Feb 16, 2026
![[PSA] Anthropic's Method to Losing Goodwill in a Few Easy Steps](https://pub-90f0ac00d93c47daac3e1d2cdd28d496.r2.dev/articles/bc0d5e0d91ce2945c6d54237cac8fe90.webp)
Anthropic's Method to Losing Goodwill in a Few Easy Steps
Jul 6, 2026

How I use Claude Code: Separation of planning and execution
Feb 22, 2026

I Cancelled Claude: Token Issues, Declining Quality, and Poor Support
Apr 24, 2026

Claude Code users hitting usage limits 'way faster than expected'
Mar 31, 2026