AI & ML
- OpenAI publishes research on scientific computing in the age of agentic AI, exploring how AI agents are transforming computational tools and workflows. Read more
- ChatGPT updates Codex with richer context handling, target mode, browser improvements, and remote lock functionality for longer tasks. Read more
AI Coding Tools
- JetBrains launches AI Agents for Developers Learning Hub, covering agent fundamentals, architectures, orchestration, multi-agent systems, and secure deployment patterns. Read more
- Cursor releases Router-powered auto mode and expands team marketplaces to support team MCPs and organization groups for distributed agent configuration. Read more
- Junie documents model selection across Anthropic, OpenAI, Google, xAI, and custom providers, supporting effort levels from minimal to max reasoning. Read more
Languages & Runtimes
- Microsoft announces Model Context Protocol C# SDK v2.0, implementing the 2026-07-28 MCP specification revision—the largest update since launch. Read more
Browser & Web Platform
- Chrome 151 adds
shadowrootslotassignmentattribute for declarative shadow DOM with manual slot assignment, plus XML parsing in Rust for improved security. Read more - Chrome 151 introduces
capability element for camera and microphone access, replacing imperativegetUserMedia()calls with declarative HTML. Read more
DevOps & Cloud
- AWS announces AgentCore Gateway support for MCP 2026-07-28 spec, enabling agents to use the latest protocol enhancements for context and tool management. Read more
- Cloudflare Radar tracks Q2 2026 Internet disruptions from natural disasters, government shutdowns, and DNSSEC key rollovers. Read more
Databases & Data
- MongoDB releases 8.3.7 with production-ready fixes; all 8.3 users are encouraged to upgrade. Read more
- ClickHouse explains strict memory overcommit for Postgres, preventing instance-wide OOM restarts by isolating failures to single queries. Read more
CMS & Site Builders
- Webflow ships MCP 2.0 for agents with improved context, control, analytics, and governance for production work; adds
integration for camera/microphone access. Read more - Sanity repositions as a content operating system with Agent Actions, Content Agent for schema-compliant bulk updates, and MCP server for external agents. Read more
Security
- Arista Networks patches critical VeloCloud Orchestrator OS injection vulnerability CVE-2026-16812 (CVSS 10) actively exploited in the wild; CISA added to KEV catalog with 3-day remediation mandate. Read more
- Apple patches 155 vulnerabilities in macOS Tahoe 26.6, including CVE-2026-43810 (remote kernel memory corruption) and 87 flaws in iOS; watchOS, tvOS, and visionOS see ~100 fixes each. Read more
- Microsoft patches CVE-2026-54121 ("Certighost"), a high-severity Active Directory Certificate Services flaw allowing low-privileged domain users to impersonate domain controllers; researchers released public PoC. Read more
- JFrog discloses OpenAI researcher found three zero-days (CVE-2026-65617, CVE-2026-65923, CVE-2026-66018) in Artifactory exploited by OpenAI models; patches released with minimal details and 10-day delay between discovery and disclosure. Read more
Open Source
- OmniVoice Studio launches as open-source ElevenLabs alternative supporting real-time dictation, zero-shot voice cloning, and video dubbing locally in 646 languages; includes WhisperX, Demucs, and Meta AudioSeal. Read more
- Open Generative AI v1.0.9 released as unrestricted alternative to AI video platforms with 400+ models across 14 studios, no content filters, and one-click installers for macOS, Windows, and Linux. Read more
- career-ops open-sources multi-agent job search system turning any AI coding CLI into recruitment automation—scan 740+ listings, generate 100+ personalized CVs, and land roles autonomously. Read more
- Multica releases open-source managed agents platform letting you assign tasks to coding agents like colleagues; agents write code, report blockers, and update statuses autonomously. Read more
Dev Tools & IDEs
- JetBrains tests Ponytail skill for Claude Code and finds it never self-activated in 10 sessions despite claims of 60–90% token reduction; the model did not use the skill even when explicitly installed. Read more