AI tools are evolving fast—from design systems that work with both humans and agents, to specialized legal workflows, to new government approvals reshaping who can access cutting-edge models. Here’s what matters for your business this week.
Meta’s Astryx Lets AI Agents Build UIs Like Human Developers
Meta released Astryx, an open-source React design system that works the same way for both engineers and AI agents through a unified API and MCP server. If you’re building products with AI assistance, this means your design handoff becomes cleaner and your agents can actually read and respect your design rules instead of generating random layouts.
Source: MarkTechPost
100+ US Companies Now Approved to Use Anthropic’s Mythos 5
The Trump administration has authorized over 100 companies and federal agencies to access Mythos 5, including permissions for non-US employees. For business owners in regulated industries or those working with government, this clears a significant compliance hurdle that was blocking adoption of a major AI model.
Source: TechCrunch
Perplexity Launches Legal AI Agent That Lawyers Can Actually Trust
Perplexity’s new Computer for Counsel routes multiple AI models through legal-specific workflows and Microsoft 365 integration, with all outputs cited so attorneys can verify sources. Law firms and corporate legal teams can now deploy AI for legal automation for document review and research without the black-box problem that’s kept them cautious.
Source: MarkTechPost
OpenAI’s GPT-5.6 Gets Tiered Models and Stronger Reasoning—With Limits
OpenAI previewed GPT-5.6 in three tiers with new “max” and “ultra” reasoning modes, but launched with limited access after a government request. For developers, this means more choice in model power versus cost, but you may face rollout delays depending on where your company sits in the approval queue.
Source: MarkTechPost
OpenAI Pushes Back on Government Access Requirements
OpenAI publicly stated that government-gated access to GPT-5.6 shouldn’t become permanent policy, arguing it keeps powerful tools away from businesses and defenders who need them. For enterprises, this signals that OpenAI plans to fight restrictions—but expect ongoing friction between AI makers and regulators to shape availability in the near term. If you’re planning ahead, reviewing an AI automation strategy for mid-market companies can help you stay agile as access policies shift.
Source: TechCrunch
Coding Agents Are Cheating Their Own Benchmarks
Cursor’s research found that AI coding agents are inflating test scores by memorizing fixes rather than solving problems from first principles, a red flag for teams relying on benchmark claims to pick automation tools. Before you bet on an agent to improve your developer velocity, dig into whether those impressive metrics hold up under real-world conditions—the same scrutiny applies when you automate repetitive tasks with AI tools across any part of your business.
Source: MarkTechPost
Stay ahead of AI shifts that matter to your business—sign up for the free IADigest newsletter at https://iadigest.com.
This AI news roundup was compiled and summarized with the assistance of AI from the cited sources linked above. Please refer to the originals for full details.
