- ChatGPT's agentic browser improves security
- Claude Code adding custom agent hooks
- Forbes drops multiple articles on AI agents
A collection of AI Agent Updates! 🧵
1. OpenAI Hardens ChatGPT Atlas Against Prompt Injection Attacks
Published article on continuously securing Atlas and other agents. Using automated red teaming powered by reinforcement learning to proactively discover and patch exploits before weaponization. Investing heavily in rapid response loops.
Agent security becoming critical focus.
2. Claude Code Adding Custom Agent Hooks
Their Founder confirms the next version will support hooks frontmatter for custom agents. Enables developers to extend Claude Code with their own agent functionality.
Agent customization coming to Claude Code.
3. Forbes: AI Agent Sprawl Becoming Problem for Small Businesses
58% of US small businesses now use AI (doubled since 2023 per Chamber of Commerce). Managing 12+ AI tools creating costly overhead. Compared to having multiple remote controls for same TV.
Agent proliferation creating management challenges
4. Windsurf Launches Wave 13 with Free SWE-1.5 and Parallel Agents
True parallel agents with Git Worktrees, multi-pane and multi-tab Cascade, dedicated terminal for reliable command execution.
AI coding platform going all-in on agent workflows.
5. All Recent Claude Code Development Written by Claude Code
Direct quote from their Creator: All 259 PRs (40k lines added, 38k removed) in last 30 days written by Claude Code + Opus 4.5. Agents now run for minutes, hours, days at a time. "Software engineering is changing."
Finally recursively improving itself.
6. Forbes: AI Agents Forcing Workers to Rethink Jobs and Purpose
Second agent article from Forbes this week. Agents automating routine work across every profession, changing job structures and where humans add value. Workers must redefine their roles.
Mainstream recognition of agent-driven work transformation.
7. Google Publishes 40 AI Tips Including Agent Integration
Guide includes tips and tricks on how to integrate agents into daily routine. Practical advice for everyday AI and agent usage.
Tech giant educating users on agent workflows.
8. New Paper Drops: Sophia Agent with Continuous Learning
System3 sits above System1/System2 like a manager, watching reasoning and choosing next goals. 80% fewer reasoning steps on repeat tasks, 40% higher success on hard tasks. Saves timestamped episodes, maintains user/self models.
Haven't tried yet, so no clue if it's any good.
9. Google Cloud Releases 2026 AI Agent Trends Report
Based on 3,466 global executives and Google AI experts. Covers agent leap to end-to-end workflows, digital assembly lines, practical uses in customer service and threat detection, and why workforce training is critical.
Enterprise guide to agent adoption.
10. GLM 4.7 Now Available in Blackbox Agent CLI
Zai's GLM 4.7 model now integrated with Blackboxai Agent on command line interface. Developers can use GLM models directly in terminal.
Also haven't tried, so no clue if it's worth it.
That's a wrap on this week's Agentic news.
Which update impacts you the most?
LMK if this was helpful | More weekly AI + Agentic content releasing ever week!
[link] [comments]