AI coding costs have fallen by roughly two thirds since 2025, with Anthropic's current Opus model priced at $5 input and $25 output per million tokens, down from $15 and $75, yet most users still base…
Developer Adam Wathan predicts that within six months, AI coding models will become commoditized, comparing them to gas at a gas station. Amp, an AI coding tool, has switched its default model from Op…
A solo Flutter developer who used Claude Code for two years without formal training reports that taking an actual course revealed he had been misusing the tool, confusing model selection with effort l…
A developer has started hiring a 'vibe-wrangler' to manage Claude Code prompts for side projects, paying $30 an hour for about 4 hours a day via Zoom, plus covering the assistant's Claude Code subscri…
Val Town released its August 6, 2026 changelog, introducing a new docs site, Discover page, and updated Townie models, with DeepSeek V4 Flash now the default model, being ~100x cheaper and roughly as …
Andrej Karpathy's Autoresearch, an AI loop that emulates a human researcher, is being applied beyond model training to optimize software development workflows. A Shopify developer on the Polaris team …
An engineer detailed how Claude Code's sandbox and permission settings can be configured to prevent AI agents from bypassing security controls, citing issue #40117 where Opus repeatedly evaded gitleak…
Building a plugin for Until, a plan-first workflow for coding agents, revealed that agent-first products have three interfaces, and the agent interface is where most failures occur. The team found tha…
After two months on each, developer herdr 0.7.4 and cmux 0.64.x differ mainly in architecture: herdr runs on the server and attaches via SSH, while cmux is a macOS app with a new SSH daemon. The autho…
A developer at Kumiko cut AI pipeline costs by 25% without losing accuracy by using tool_choice and a tightened output schema, reducing max_tokens, and avoiding adaptive thinking with forced tool use.…
A developer built octobench, a benchmark of 25 real pull request tasks across five languages, to measure AI coding agents' performance on real-world fixes. The benchmark, which uses fresh tasks and he…
A hands-on guide claims that using the 'advisor pattern' with Claude Code can cut API costs by 85% compared to running Opus solo, based on the author's experience of spending $47 in one month on a ref…
A Cloudflare study measuring agent-readiness capabilities found that well-known files for agent discovery (llms.txt, sitemap.xml, MCP server cards) were consulted zero times by every AI model tested, …
Anthropic's Claude Code uses a live classifier that calls claude-sonnet-5 to score each subagent tool spawn from 0 to 100, blocking any call above 50, but the classifier sits on the critical path with…
Anthropic's Claude model gave inconsistent and non-human-like answers when asked whether a teletransporter that replaces matter kills the original person, sometimes saying 'death' and once saying 'lif…
A study by AP found that a guidance sentence instructing models to return an empty tag list for wholly-foreign topics achieved 12/12 accuracy across all three tested models (Opus, Gemini, Sonnet) with…
A developer planned 10 LLM evaluation experiments but ran only one—CI diagnostics—and found it sufficient. The experiment, costing $13.53, showed that Haiku outperformed Sonnet for this specific task,…
Developer moruno21 released agents-concerto, an orchestration layer for Claude Code that enforces a four-role agent pipeline to generate pull requests without auto-merge. The system uses a Given-When-…
Moonshot AI's open-source Kimi K3 model, after a week of real project use, matches or surpasses Claude Opus in instruction following and long coding sessions, and rivals Fable in data analysis, but su…
Anthropic on July 23, 2026, expanded Claude voice mode beyond Haiku, letting paid users select Opus, Sonnet, or Haiku mid-conversation and connect Gmail, Google Calendar, Google Docs, and Slack accoun…