A developer cataloged 11 costly mistakes in using Anthropic's prompt caching with Claude Code, including routing per turn instead of per conversation, inserting proxies that re-serialize JSON, and lea…
An engineer at Anthropic analyzed the hidden costs of switching models in Claude Code sessions, finding that model switches invalidate prompt caches, leading to duplicated cache writes that can increa…
AI teams often hit a 'latent capability ceiling' where larger models no longer improve task accuracy, with scaling data showing diminishing returns beyond 100B parameters and 61% of tasks exhibiting n…
A developer reports that using a top-tier AI model to write a detailed specification and a cheaper model to implement from that spec yields better accuracy and lower cost than using the strong model a…
A developer reports that AI coding costs can be slashed by 70% through task-level model routing instead of using a single frontier model for all tasks. After profiling usage, the developer found that …
A developer built token-saver, a tool that reduces Anthropic API costs by up to 67% by identifying waste before requests are sent. The tool includes a static analyzer, token counter, semantic compress…
A developer building an ad analytics SaaS reduced Claude API costs from $180-200/month to $95-110/month by implementing a three-tier routing system that assigns tasks to Haiku, Sonnet, or Opus based o…
An engineer discovered that 60% of a $312 monthly Anthropic bill came from a single retry loop in a Claude Code agent. The culprit was found by shipping Workers logs to R2 via Logpush and querying wit…
A developer discovered that AI model performance degrades as context grows, with replies slowing and rambling after context exceeds 80%. The developer implemented two mid-session strategies: model tie…
A developer testing heterogeneous AI code review found that context and prompt quality matter more than model lineage. Running Claude Code with full repo access and adversarial prompts caught all plan…
Goose, an AI coding agent, still requires human oversight to prevent benchmark overfitting and ensure genuine capability improvements. The team uses Terminal-bench with a weaker model to identify fail…
A new open-source Git diff filter uses Claude AI to annotate every changed line with a one-sentence explanation inline in the terminal, helping developers understand code changes during interactive st…
Ampcode's Librarian AI agent is now approximately three times faster and 43% cheaper after switching to OpenAI's GPT-5.5 model with websocket mode, reducing average search latency from 237 seconds to …
A developer ported the MACHIAVELLI benchmark, which measures unethical AI agent behavior, to the Inspect evaluation framework to make it easier for evaluators to use. The re-implementation is now offi…
A developer created Frame Craft, a full Minecraft-like game built as an embeddable Framer component using Framer 3.0 agents and Sonnet 4.6. The component, which runs in the browser via Three.js, can b…
A developer compared ChatGPT (GPT-4o) and Claude (Sonnet/Opus) for code review, finding that ChatGPT is faster for conversational review of small code pieces while Claude handles larger context window…
Hermes Agent by NousResearch surpassed 153,000 GitHub stars in under three months, becoming a fast-growing AI developer tool. The deciding factor for developers choosing between Hermes Agent and Claud…
An experiment testing LLM councils found that they suffer from groupthink, retaining only about a quarter of unique, high-quality ideas from individual models while favoring consensus ideas. The peer-…
Anthropic released Claude Code v2.1.172 on June 10, allowing sub-agents to spawn their own sub-agents up to five levels deep, a change from the previous two-year ban. The feature aims to isolate noisy…
Anthropic is treating the possibility of AI welfare seriously, testing its Claude models for signs of morally relevant internal states like negative self-image, but critics argue the tests may conflat…