Claude Code is unusable for complex engineering tasks with the Feb updates
anthropic claude
| Source: HN | Original article
Claude Code’s February rollout has back‑slid into a state where the tool can no longer be trusted with anything beyond trivial scripts. Users on Anthropic’s Max x5 plan report that the new v2.1.53–v2.1.59 builds, released on Feb 25‑26, trigger rapid consumption of usage quotas, frequent “auto‑memory” bloat, and outright freezes when the model attempts complex engineering steps. A GitHub issue thread opened four days ago describes the regression as “cannot be trusted to perform complex engineering,” echoing complaints that the system behaves like a stripped‑down version of its January incarnation.
The problem matters because Claude Code was positioned as a full‑stack coding assistant capable of reading any language, mapping component interactions, and iteratively refining solutions. Its promise attracted enterprises looking to automate large‑scale refactoring, security audits, and multi‑service deployments. The sudden loss of reliability undermines those use cases, forces teams back to manual code reviews, and erodes confidence in Anthropic’s roadmap. Moreover, the accelerated hit of usage limits—an 8 % session consumed in roughly 18 minutes according to community monitoring—means higher costs for customers who already pay premium rates for the Max plan.
Anthropic has acknowledged the issue in a public statement, labeling the fix as “top priority.” The changelog released alongside the updates notes patches for 100 % CPU loops and deadlocks caused by permission prompts and bulk skill‑file changes, but no timeline has been given. As we reported on April 6, 2026, Claude Code’s auto‑mode and permission‑trap quirks were already under scrutiny; this latest setback deepens the concern.
What to watch next: a formal patch release, likely before the end of the month, and any revision of the usage‑limit algorithm that could restore the model’s cost‑effectiveness. Equally important will be Anthropic’s communication on whether the “auto‑memory” feature will be rolled back or re‑engineered, and how the company plans to regain developer trust after this regression.
Sources
Back to AIPULSEN