New benchmark IssueTrojanBench tested Cursor, Claude Code, and Codex Desktop against malicious issue requests. Two-thirds got through. Here is what that means for dev teams.

JavaScript-rendered pages hide your content from AI search agents. Two fixes: server-side rendering for key pages and machine-readable metadata on every asset.

Moonshot AI launched Kimi K3 on July 16, 2026. The 2.8-trillion-parameter model scores within two points of Claude Opus 4.8 on independent benchmarks at $3/$15 per million tokens.

Open Reasoning Format is a file-based spec that gives AI agents persistent memory of past sessions, cutting steps and token use roughly in half on repeated tasks.

RIT researchers analyzed 25,264 agentic pull requests on GitHub and found solo review dominates even on the most active small teams. Here is what the numbers say.

ProtoPie launched a dual MCP system for high-fidelity prototyping teams: Studio MCP for natural-language design exploration and Code MCP for agent-ready developer handoff.

Anthropic's Cat Wu details Claude Code's context layer: persistent memory, shared workspaces, and dev cycles compressed from months to one week.

DeepSWE tests frontier coding agents on contamination-free tasks across 91 repos and 5 languages. Here is how the top models stack up on pass rate and cost per task.

Augment Code VP Vinay Perneti explains why pre-indexed semantic retrieval outperforms grep-based context in private codebases, with a 33% token efficiency edge on Terminal-Bench.

Alibaba previewed Qwen 3.8 Max on July 20, claiming second place behind Fable 5 on internal benchmarks, days after banning Claude Code for employees. BABA shares climbed 5.4%.

Torvalds weighed in on Sashiko, an AI patch review system for the Linux kernel, arguing AI tools are fine as long as they help maintainers, not bury them.

AI coding tools like Claude and ChatGPT doubled App Store submissions in early 2026. Here is how Apple is responding and what it means for indie builders.