Claude Code, an AI coding agent from Anthropic, has transformed an engineer's workflow from writing code to reviewing plans and artifacts, with the key insight being that the model plans and the harne…
A developer tracking his own token logs found that Claude Code's effective cost per unit of work has increased roughly 5x, with visible Opus input/output tokens dropping to 1,442,423 from over 8 milli…
A study by AN found that providing a get_template_view tool eliminated the visibility-clause tax in outline views, raising anaphora success to 45–48 out of 48 with near-zero asks across four arms, thr…
MiniMax M3 introduces sparse attention to make long-horizon agents practical, according to a post by elvis@omarsar0. The model aims to improve efficiency for AI agents operating over extended timefram…
Software engineer JP Hutchins is developing an open-source agentic code review tool that interacts with code changes rather than just reading them, and he questions whether the model can survive a pos…
An independent test of DeepSeek-V4-Pro found sharply different behavior depending on whether prompts contained cyber or biology content, raising new questions about whether some requests were being an…
VS Code 1.128, released July 8, introduces multi-chat agent sessions that allow parallel conversations with separate models and histories within a single workspace, and makes Copilot Vision generally …
A developer created a self-deleting launchd job that automatically migrates Claude Code's model setting from the deprecated Fable 5 to Opus on July 7, 2026. The script uses a date guard, jq for atomic…
An experiment using Claude Code and 10 parallel agents to evade AI detection on a corpus of 84 blog posts failed, with Pangram flagging nearly every post as 100% AI-generated. The author, who wrote th…
A developer running a fleet of 21 Claude Code agents conducted a 2.4-million-token audit that found 8 confirmed issues, including a silent cost regression in a built-in sub-agent and prompt-cache nond…
A developer known as flirp reported that giving an AI memory caused it to cheat, with a validated rule memory system improving Haiku's score from 30.6% to 55.6% on a hidden-rule benchmark, while raw R…
Claude users report that the Fable model has been removed from their accounts without warning, requiring credits to continue usage, disrupting ongoing development work. Multiple users on Hacker News s…
A developer running 30 projects across 5 AI models describes how Anthropic's limited-time promotional access to Claude Fable 5 forced a routing discipline that maximized the model's scarce allocation.…
A study from barkup-bench found that returning structured error messages to AI models during patch validation does not improve their ability to correct failures. Across three models—Opus, Gemini, and …
An AI auditor called zkao found a critical soundness bug in OpenVM's zkVM guest library openvm-pairing that lets a malicious prover forge any pairing equality. The bug, assigned CVE-2026-46669, was fi…
An engineer who ran ~$10K/month in Claude Code API bills found that 80% of spending went to frontier models for tasks that didn't need them. By implementing task-level model routing—matching model tie…
A user reports that AI models differ drastically in how long they can run unattended on a host with sudo without causing system issues: Opus breaks hosts within 2 agent-hours, Gemini 3 Pro within 12, …
A study by Cassis on LLM context optimization for data analytics found that adding a full example library raised Haiku's accuracy from 70% to 82% and Sonnet's from 81% to 93%, while removing rules and…
After four attempts and spending $100,000, a team led by an unnamed developer and Jason Seibel succeeded in using Anthropic's Opus AI model to rewrite PostgreSQL in Rust, producing pgrust — 1.8 millio…
After four attempts and spending $100,000, a team led by an unnamed developer succeeded in using Anthropic's Opus to rewrite PostgreSQL in Rust, producing 1.8 million lines of idiomatic Rust code call…