At the London AI Symposium Tour, Microsoft made a high-profile announcement of the Copilot Studio's slew of new features and teased a public preview of "Autonomous Agents" at the Microsoft Ignite 2024 conference. The move has certainly reignited the industry's interest in autonomous intelligences. Microsoft claims that these agents are able to "understand the nature of a user's work and perform actions on their behalf," which sounds exciting and as if it signals a leap forward in enterprise efficiency.

Since the emergence of generative AI and Microsoft Copilot, Microsoft's deployment in the field of AI has significantly accelerated. According to official data, 2.1 million users use Copilot in Microsoft's business apps every month. This is a significant number, which initially confirms the market's interest in Copilot. However, it's worth noting that the number of users does not equate to deep user engagement and productivity gains. The actual utility of Copilot, and the extent to which it actually improves users' workflows, remains to be verified by deeper data and case studies.

Leading organizations, including McKinsey & Company, Pets at Home, Thomson Reuters and Clifford Chance, have publicly expressed positive reviews of Microsoft Copilot Studio and Microsoft's modern business applications. "AI first" seems to be the new business mantra. The speed at which these companies are embracing AI is impressive, but it's too early to tell if "AI First" will really be the cornerstone of long-term competitiveness. The implementation of an "AI-first" strategy is not just about technology deployment, but also involves comprehensive changes in organizational structure, talent training and business processes.

Microsoft Ignite 2024: a "glimpse" of autonomous agents?

The Microsoft Ignite 2024 conference will be a key window into the autonomous agent capabilities of Copilot Studio. While Copilot is defined as a "personal assistant", agents are elevated to the level of "expert systems", claiming to be able to "operate autonomously on behalf of a process or company". Microsoft's blueprint is for a future in which every employee will have Copilot, supported by a multitude of agents. While this is an ambitious vision, organizations will face significant challenges in deploying, managing, and maintaining the "Copilot per person + agents" scenario. How to ensure that these agents work together and avoid information silos and wasted resources will be an issue that must be seriously considered in practical applications.

Copilot Studio emphasizes the "uniqueness" of its platform by allowing users to create custom agents for specific processes. While this flexibility is important, in practice, is it really easy for end-users to get started and build efficient, reliable agents autonomously? Low-code platforms lower the technical barrier, but building complex autonomous agents still requires a certain level of expertise and a deep understanding of business processes.

Autonomous agents: a "panacea" for efficiency challenges?

The pursuit of efficiency improvement, customer experience optimization and business growth is a constant theme for enterprises. Copilot Studio tries to solve these pain points through autonomous agents, and the direction is undoubtedly correct. However, autonomous agents are not a "panacea", and their applicable scenarios and effects need to be objectively evaluated.

The case of Pets at Home using Copilot Studio to build agents for the Profit Protection team seems to indicate that autonomous agents have potential for specific business process optimization. Automating data collection and initial analysis through AI frees up human labor to focus on more complex decisions, which sounds like sound logic. But it's worth asking whether this potential for "seven-figure annual savings" is universal. Can similar success stories be replicated in other industries and organizations?

McKinsey & Company claims to have reduced the delivery cycle time of its client onboarding process by 901 TP3T and the administrative workload by 301 TP3T, impressive figures. However, the consulting industry, with its relatively standardized and process-oriented business model, may be better suited for automation. Other industries, such as manufacturing and retail, have more complex and diverse business processes, and the results of autonomous agent application may vary. The case of McKinsey & Company perhaps demonstrates more the potential of autonomous agents in specific industries and specific scenarios, and its generalizability still needs to be further observed.

The case of Thomson Reuters building professional-grade agents for legal due diligence points to the promise of autonomous agents in expertise-intensive domains. Legal due diligence requires a high degree of expertise and experience, and Thomson Reuters' attempts have shown that Copilot Studio has the ability to build agents that handle complex specialized tasks. But this also means that building such agents requires a significant investment in specialized knowledge and domain data. Building similarly advanced agents may not be easy for organizations that lack the data and expertise to do so.

The application areas for Autonomous Agents are portrayed as very broad, spanning a wide range of sectors and industries. Microsoft also plans to introduce ten pre-built Autonomous Agents in Dynamics 365 to further expand its application scenarios. However, it is worth noting that the value of an autonomous agent ultimately depends on its ability to solve real problems. If the agent simply automates repetitive tasks without truly enhancing the intelligence of business decisions, its value will be greatly diminished.

New Copilot Studio features: a 'step up' in autonomy?

The new features released in Copilot Studio, such as Autonomous Triggers, Dynamic Agent Scheduling and Activity Overview, are all designed to enhance agent autonomy and manageability. Autonomous Triggers allow agents to "automatically respond to business signals and initiate tasks", which sounds smart, but in practice, how to avoid "false triggers" and "over-triggering"? How to ensure that the agent's trigger logic is consistent with the business goals?

Dynamic agent planning emphasizes the agent's ability to be "flexible and adaptive", able to dynamically adjust the execution path according to different situations. This is undoubtedly a key step in improving agent intelligence. However, the complexity of dynamic plans may also bring new challenges. How to ensure the rationality and effectiveness of dynamic plans? How to make users understand and trust the agent's dynamic decision-making process?

The Activity Overview feature is designed to enhance the "transparency and accountability" of agents. This is critical for enterprise-level applications. Through logging and operational monitoring, users can better understand the agent's operational status and decision-making process, and identify and resolve issues in a timely manner. However, this also places higher demands on the monitoring and management capabilities of the platform.

The Copilot Studio agent utilizes the latest models including the OpenAI o1-like family. More powerful models will undoubtedly improve the agent's reasoning and problem solving capabilities. However, increased model performance often comes with increased costs. Organizations need to consider cost-effectiveness while pursuing smarter agents. The specific performance and cost of the OpenAI o1 series models, as well as their practical application effects in the field of autonomous agents, need to be further observed.

Security and governance: the "lifeline" of autonomous agents

Microsoft emphasized the "security and reliability" of the Copilot Studio platform and described features such as enterprise-class data protection, security fencing and controls, and lifecycle management. These features are critical for enterprise-class applications. Autonomous agents are involved in core enterprise business processes and sensitive data, and security is their "lifeline." Microsoft's investment in security and governance deserves to be recognized, but the actual application results need to be rigorously tested and verified. Especially for large enterprises and highly regulated industries, how to ensure the compliance and security of autonomous agents will be a long-term and ongoing challenge.

The Future of "AI First": Opportunities and Challenges

Microsoft is positioning Copilot as "the new UI for AI" and believes that autonomous agents will make workflows "smarter and more resilient." It's an imaginative vision. The combination of Copilot and autonomous agents may indeed reshape human-computer interaction and propel organizations toward an "AI-first" future. However, there is still a long way to go from "vision" to "reality".

The release of Copilot Studio undoubtedly provides a new platform and tool for enterprises to embrace autonomous intelligences. However, to truly unlock the potential of autonomous intelligences, there are still many challenges to overcome, including technology maturity, application scenario exploration, security risk prevention, and user acceptance, among others. The Microsoft Ignite 2024 conference is coming soon, when we may be able to see more "dry goods" and practical application cases about Copilot Studio autonomous agents. Let's wait and see if Copilot Studio can really lead enterprises into a new era of autonomous intelligent bodies.