presented by

Hello there!

Two significant pieces of AI policy are set to make waves this summer: The first is the recently signed AI Action Plan, the second will be the long-awaited enforcement of the European Union’s AI Act in August. These two AI regulations share more similarities than initially thought. Both focus on evaluation and monitoring, although what they evaluate is slightly different. Enterprises might wonder what the rules would mean for them, though it is still early to really know the impact. What is sure for now, is that organizations should continue to consider their use cases before choosing a model.

Keep scrolling for more AI orchestration headlines.

A ChatGPT ‘router’ that automatically selects the right OpenAI model for your job appears imminent

Alibaba’s new open source Qwen3-235B-A22B-2507 beats Kimi-2 and offers low compute version

Early Anthropic hire raises $15M to insure AI agents and help startups deploy safely

How OpenAI’s red team made ChatGPT agent into an AI fortress

Open-source MCPEval makes protocol-level agent testing plug-and-play

Enterprises must evaluate the performance of agents. As more methods for testing agents emerge, some are leveraging other frameworks to bring evaluation to where agents are. Salesforce researchers developed MCPEval, based on the Model Context Protocol (MCP) architecture, that evaluates tool usage of agents in MCP servers. The researchers stated that their method differs from other performance testing frameworks in that MCPEval evaluates agents based on the tools and MCP servers they would use in real-life scenarios.

A message from WEKA

Your AI factory deserves better infrastructure

AI is no longer just R&D—it’s a full-scale factory for innovation and business value. But your factory can’t run on outdated infrastructure. It needs a modern data pipeline built to accelerate every stage of the AI lifecycle, from massive training runs to lightning-fast inference. WEKA and NVIDIA deliver a production-grade, high-performance foundation that eliminates bottlenecks, skips data copies, and scales seamlessly to power your AI factory at full speed.

Learn more

Interesting Stories

Anthropic researchers discover the weird AI problem: Why thinking longer makes models dumber

Why it matters: We tend to think of models as being smart, and for the most part, there is some intelligence there. As enterprises move to running models longer to facilitate agentic use cases, there’s an expectation that models continue to do smart things. However, research from Anthropic showed that the more a large language model (LLM) thinks about its answer, the less intelligent it becomes. The problem appears to lie in its training, as models can inadvertently reinforce problematic reasoning the longer the process continues.

White House plan signals “open-weight first” era — and enterprises need new guardrails

Why it matters: The AI Action Plan, signed on Wednesday, outlines how governments can contract with AI model and application providers and instructs companies to remove certain content from their models. While enterprises are not directly affected by the executive order unless they work with the government, the plan sheds light on how this administration will approach model and data evaluation, as well as its stance on open-sourced AI.

OpenAI’s ChatGPT Agent is haunting my browser

Why it matters: Browser use agents could soon be ubiquitous, but while many organizations are excited about this future, some individual users may find them slightly disturbing. These agents can be taught to access information on a website or finish transactions from online retail stores. Currently, it sometimes feels as though agents like the ChatGPT Agent are tracking you as you browse your browser.

AI Impact Series Returns to SF – Aug 5

Ready for the next phase of AI? AI is moving beyond copilots. Join leaders from Block, GSK, and SAP as they share how autonomous agent networks are transforming enterprise operations.

Here are the key takeaways you can expect:

• Scalable agent architecture

• Real-world deployment lessons

• Governance, cost, and compliance strategies

Claim your spot

Featured post of the day

—Thanks for reading! See you next week!

Emilia David

Unsubscribe from this message type