Anthropic admits engineering missteps behind Claude Code decline as users revolt

Anthropic Confirms Technical Errors Behind Claude Code Performance Decline

Anthroplic, the AI lab valued at $380 billion, has publicly acknowledged that engineering missteps were responsible for a noticeable performance decline in Claude Code that frustrated developers and sparked user backlash throughout March and April 4

4

. The company published a detailed post-mortem identifying three specific technical errors in its deployment pipeline that degraded the code generation tool's capabilities 5

5

.

The first issue emerged on March 4th when Anthropic reduced Claude Code's default reasoning effort from "high" to "medium" in response to user complaints about large delays that made the application appear frozen 2

2

. While this change improved response times, it significantly limited Claude Code's capabilities. After receiving feedback from developers who preferred the option to manually adjust effort levels when needed, Anthropic reverted to the high default on April 7th 2

2

.

The second bug appeared on March 26th affecting Sonnet 4.6 and Claude Opus 4.6 2

2

. A system designed to clear older thinking after an hour of chat inactivity malfunctioned, instead wiping prior context after every turn. This caused Claude Code to forget previous interactions and repeat phrases frequently, undermining its effectiveness for developers working on complex coding tasks. The company fixed this issue on April 10th 2

2

.

System Prompt Changes Hurt Coding Quality

The third problem began on April 16th when Anthropic added a system prompt instruction aimed at reducing verbosity 2

2

. The prompt limited text between tool calls to 25 words or fewer and final responses to 100 words unless more detail was required. This change, combined with other prompting tweaks, negatively impacted coding quality and was reverted on April 20th 2

2

. As of April 20th, all three issues have been resolved, though the damage to user trust may take longer to repair.

Testing Removal from Claude Pro Plan Amid Capacity Constraints

While addressing these bugs, Anthropic simultaneously sparked controversy by testing removal of Claude Code from its Claude Pro plan, which costs $20 per month 1

1

. The company's pricing page was updated to show an explicit "X" rather than a checkmark for Claude Code access on the Pro tier, though this change was inconsistently applied across Anthropic's documentation 1

1

.

Amol Avasare, Anthropic's head of growth, clarified on social media that the change represented an A/B test affecting approximately 2 percent of new prosumer signups, with existing Pro and Max subscribers remaining unaffected 1

1

. However, the lack of clear communication about subscription plan changes alarmed developers and raised questions about how Anthropic manages customer relations, particularly for enterprise customers who require predictability 1

1

.

AI Pricing Models Face Sustainability Crisis

Avasare explained that when Claude Max launched a year ago, it didn't include Claude Code, Cowork didn't exist, and agents that run for hours weren't available 1

1

. "Max was designed for heavy chat usage, that's it," he stated, noting that engagement per subscriber has increased dramatically while current plans weren't built for this level of usage 1

1

. The fundamental problem is that Anthropic's subscription plans charge far less than the book value of tokens consumed, sometimes by a factor of ten or more 1

1

.

This situation mirrors broader challenges facing AI pricing models across the industry. GitHub recently suspended signups for its flat-rate Copilot plans and removed Claude Opus access from its $10-per-month Pro plan 3

3

. These moves suggest current pricing structures are becoming unsustainable as AI capabilities advance, potentially forcing significant restructuring across providers 3

3

.

Computing Resources Strain Under Surging Demand

Widespread speculation suggests Anthropic may be facing capacity constraints as usage of its products has soared in recent months 4

4

. "Demand for Claude has grown at an unprecedented rate, and our infrastructure has been stretched to meet it, particularly at peak hours," Anthropic stated, acknowledging that compute is a constraint across the entire industry 4

4

.

The company has introduced usage limits to encourage use during off-peak hours and is exploring further limitations 1

1

. Anthropic announced it is scaling computing resources through expanded partnerships with Amazon and Google, which will bring significant new capacity online in coming months 4

4

. The company's annualized recurring revenue run rate has reached $30 billion, more than triple its figure at the end of last year, driven largely by developer loyalty to tools like Claude Code 4

4

.

Transparency Questions and Competitive Implications

The feeling among some users that Anthropic had been gaslighting them about the performance issues potentially undercuts the company's attempts to market itself as more transparent than rival OpenAI 4

4

. Initially, Anthropic executives suggested performance issues resulted from changes made to improve latency and respond to user feedback about token use, pointing to its public changelog 4

4

.

Sam Altman, OpenAI's chief executive and Anthropic's competitor, trolled Avasare with an "ok boomer" comment, hinting that OpenAI would welcome developers angered by Claude subscription plan changes 3

3

. The controversy comes at a critical time, with both companies reportedly preparing for initial public stock offerings later this year 4

4

. Developers will be watching closely to see whether Anthropic follows through on Avasare's promise that if changes affect existing subscribers, they'll receive "plenty of notice before anything changes" directly from the company rather than discovering updates through screenshots on social media 1

1

.