In a significant leap towards better governance of AI agents, Microsoft has unveiled a new skill called run-assert-eval aimed at small business developers who rely on AI for customer service and operational efficiency. This innovation comes on the heels of tools such as ASSERT and Agent Control Specification (ACS), which were designed to enforce policies in AI systems. As AI becomes more integrated into business operations, understanding how to govern these technologies can enhance productivity and minimize risks.

Microsoft’s efforts center around two critical assumptions that often fall short. First, businesses may not have a comprehensive understanding of the risks their AI agents pose. Many failures stem from unforeseen circumstances—issues not listed in existing requirements. Second, there’s a common reliance on manual processes that may introduce errors and inefficiencies when connecting findings to policy adjustments. As small businesses increasingly adopt AI solutions, these challenges can hinder their operational efficacy and reputation.

The introduction of run-assert-eval aims to alleviate these pain points. With just a single prompt in Visual Studio Code, developers can now discover potential risks related to their agents. The skill not only measures fail rates but also automatically generates runtime policies based on identified risks. This all-in-one solution simplifies the process of validating whether a fix is effective, significantly benefiting small business owners who need to ensure their AI tools are safe and helpful.

For instance, a billing-support agent utilizing run-assert-eval revealed a shocking data exposure rate of 30% across a sample set. After applying the new governance skill, that rate plummeted to just 5.9%, demonstrating the effectiveness of runtime oversight. This capability to verify real-world implications of changes can be transformative for businesses that prioritize customer data safety.

One of the stands out features of run-assert-eval is its ability to rely on evidence-based governance rather than assumptions. It begins by threat modeling the AI agent, identifying both known and unknown risks before any metrics are assessed. This ensures that important failures are not overlooked, allowing businesses to plug potential vulnerabilities before they become liabilities.

However, small business owners should also be mindful of the potential challenges that come with this technology. The workflow might still require some level of technical expertise and investment in training staff to navigate the complexities of AI governance effectively. While automation reduces manual labor, understanding the underlying processes is critical to leverage its full capabilities.

Moreover, businesses that choose to adopt this technology will need to set aside time for team members to adopt the new system while ensuring that existing customer service processes remain unaffected. Transitioning to new tools can sometimes produce short-term disruption—an aspect that must be managed carefully, particularly in customer-facing roles.

Feedback from real users has shown that the seamless integration of run-assert-eval can streamline workflows significantly, addressing pain points in governance and compliance. Future enhancements aim to develop these tools further, allowing teams to work with even more streamlined and customized experiences tailored to their specific needs.

As small business owners gather insights into these emerging AI governance tools like run-assert-eval, the opportunity to improve customer interactions and ensure operational efficiency is clear. By investing time in understanding and implementing this technology, they can stay ahead of compliance challenges and risks associated with AI deployments.

To explore more about the capabilities and immediate applications of this new skill, visit the original press release by Microsoft. As AI technologies continue to evolve, staying informed and proactive about their governance could define the success of small businesses in an increasingly digital landscape.

Image via Google Gemini

This article, “Microsoft Unveils run-assert-eval to Enhance AI Governance and Safety” was first published on Small Business Trends