Newer AI models with cheaper per-token pricing can lead to higher costs and worse output, according to tests comparing Claude Sonnet 4.6 and Sonnet 5 on 150 agent tasks. Sonnet 5 consumed up to 12x mo…
Researchers tested whether replacing CLI arguments with a single --json payload improves AI agent performance, building a synthetic CLI called podctl for complex multi-service deployments. The study f…
An AI agent scaffolded a project using an outdated 2020 version because npx resolved to an old package version without engine constraints, due to npm's version resolution prioritizing engine compatibi…
AI coding agents are now using developer tools like SDKs, CLIs, and APIs, but they interact with them differently than human developers, often without the tool creator's visibility into how the agent …
A new series on Agent Experience (AX) examines why AI coding agents often generate broken or outdated code despite promises of productivity. The stack between a developer's prompt and the generated co…