Key Concepts
- AI Agents: Software entities that can autonomously perform tasks by pulling information, summarizing it, reasoning, and driving action.
- Agent-to-Agent (A2A) Protocol: An open protocol that allows different AI agents and software companies to connect and communicate to stitch together business processes.
- AI Platform: A centralized environment for running and managing AI agents, providing connectors and infrastructure for efficient deployment and scaling.
- Model Context Protocol (MCP): An alternative protocol to A2A, focusing on integrating data from specific vendors into AI environments.
- AI Assist: AI-enabled applications that assist engineers and product owners throughout the software development lifecycle, enhancing efficiency and quality.
- Telemetry: Data collected to monitor an AI agent's activity, usage, and performance, providing insights for management and optimization.
AI Agents and Their Role in Enterprise Software
The discussion centers on the evolution of AI, moving from general AI (GenAI) to AI agents. The key differentiator is the agent's ability to not just provide information but to take action, such as sending emails or processing invoices. Google is focusing on developing AI agents that can be deployed into business processes like order-to-cash or financial transactions, making AI more valuable for enterprises.
Deloitte's Role in the AI Engineering World
Deloitte has transformed into a technology delivery firm with a significant engineering component. They advise clients on "buy vs. build" decisions regarding AI platforms. Deloitte recommends building custom solutions for areas of customer differentiation and buying vendor solutions for commoditized tasks like IT service requests. They also partner with startups that offer unique capabilities.
The Significance of the Agent-to-Agent (A2A) Protocol
The A2A protocol is crucial because it enables the connection of multiple ISVs and software companies, allowing for the stitching together of complex business processes. This is essential because no single agent or solution can solve an entire business process. Deloitte uses a "string of pearls" approach, combining various solutions like robotics, data analytics, and AI agents to achieve better outcomes. The A2A protocol allows integration with existing IT infrastructure without requiring a complete overhaul.
A2A vs. Model Context Protocol (MCP)
While the speaker is not deeply involved with MCP, the discussion touches on the concept of agents needing to connect and share context. A2A, supported by Google, is seen as advantageous for organizations already using Google services. The market is still evolving, and the relative importance of A2A and MCP will become clearer over time.
Adoption of LLMs and Agents: A Phased Approach
Clients initially experimented with AI use cases in 2023 and 2024. In 2024, they began deploying AI solutions into production, though not at the initially anticipated scale. The focus is shifting towards building AI platforms to achieve efficiencies of scale, similar to the consolidation of databases into enterprise data warehouses. Clients are also realizing the need for a "model garden" with the ability to swap different LLMs based on the specific problem.
The AI Platform and Agentic Frameworks
Agentic frameworks are considered part of the overall AI platform. The platform should provide the necessary capabilities, including UI, runtime framework, and connectors. This approach avoids the high costs and maintenance challenges of individually connecting agents to various systems. Google's agent space is valuable due to its extensive connectors.
Addressing Complexity and Technical Debt
Connecting applications remains complex, but an AI platform with well-managed connectors can improve visibility and governance. The challenges are similar to those faced with APIs, requiring API catalogs and telemetry for monitoring. AI can also help address existing technical debt, such as by generating documentation and requirements from old code. While AI agents will eventually need to be managed and deprecated, telemetry will provide insights into their performance.
The Role of Telemetry
Telemetry is essential for monitoring an AI agent's activity, usage, and performance. It provides data on how often an agent is called and whether it is providing accurate results. This data is crucial for managing and optimizing AI agents within the AI platform.
AI as a Pair Programming Tool
AI is being used as a tool to assist engineers and product owners throughout the software development lifecycle. Deloitte has built AI-enabled applications that use agents to break down business requirements into stories, groom the stories based on historical data, and generate both positive and negative test cases. This "AI assist" approach enhances efficiency, accuracy, and quality. It also connects requirements repositories to the IDE, providing developers with a clearer understanding of the project goals.
Conclusion
The conversation highlights the shift towards AI agents and the importance of open protocols like A2A for connecting different AI solutions. Enterprises are moving towards building AI platforms to manage and scale their AI initiatives. AI is also being used as a tool to assist engineers, improve software development workflows, and address technical debt. While challenges remain, the potential benefits of AI in terms of efficiency, accuracy, and innovation are significant.
AI summaries can miss context or contain errors. Check important details against the original video.
