№ 61
Friday, July 24, 2026
AI/Tech Brief — 2026-07-24
№ 61
AI/Tech Brief — 2026-07-24
Claude Code 2.1.218 improves background tasks and safety
The /code-review command now runs as a background subagent to avoid cluttering active conversations, and Auto Mode adjudication for potentially dangerous commands (like rm) has been enhanced to reduce unnecessary permission dialogs.
Source
OpenAI introduces API hard spend limits
Developers can now set monthly hard spend caps at both the organization and project levels. Requests will return a 429 error once the limit is reached, giving teams better control over costs.
Source
DeepMind introduces computer use for Gemini 3.5 Flash
Google has expanded the capabilities of the Gemini Flash family, announcing new computer use features for Gemini 3.5 Flash to enable more agentic workflows.
Source
Black Forest Labs releases Flux 3
Flux 3 and Flux 3 Mimic have officially launched, quickly rising to the top of Hacker News discussions as the community evaluates the new generation of open-weight models.
Source
Claude Cowork gains screen memorization skills
Superhuman AI’s daily newsletter highlighted that Claude Cowork has received updates allowing it to utilize screen memorization for better contextual awareness.
Source: Superhuman AI newsletter
Gemini CLI v0.52.0 improves tool reliability
The latest release bypasses LLM correction for JSON and IPYNB files in write_file and replace tools, significantly improving reliability for these file types.
Source
Claude Cookbook launched
A new cookbook resource has been published, offering developers practical examples and guides for building applications with Claude.
Source
(Sources checked: Claude Code changelog, Claude Support release notes, Gemini CLI changelog, OpenAI API changelog, OpenAI Codex changelog, Superhuman AI, ByteByteGo, Dario Amodei’s blog, Sam Altman’s blog, DeepMind blog, and Hacker News front page.)