← back
Building Applications with AI Agents — Michael Albada, Microsoft
Takeaway
Build agentic systems incrementally up the agency ladder, minimize tool surface area, and keep deterministic business rules out of the model.
Summary
- Microsoft Security Copilot principal scientist distills his upcoming O'Reilly book; YC agent-startup acceptances grew 254% in 3 years but production agents remain hard past 70% accuracy.
- Defines agency as a continuum vs efficacy axis; RPA is high-efficacy/low-agency, and adding agency should never compromise efficacy.
- Anti-pattern: don't map 300 APIs to 300 tools — more tools means more semantic collision and lower accuracy; group tools into single human-facing actions.
- Orchestration ladder: prefer chains first, then branching/router LLMs, then full agentic loops only when chain complexity exceeds maintainability (bitter-lesson reminder).
- Keep deterministic business logic outside the model — expose state-update tools so validation/sanitization stays in fixed code.
agentstool-usearchitecture
Original description
Generative AI has dramatically shortened the distance between ideas and implementation, enabling faster prototyping and deployment than ever before. But while language models can streamline individual tasks, true transformation comes from combining these capabilities into intelligent, autonomous systems—AI agents. This talk explores how to build and deploy foundation model-enabled agent systems that go beyond simple prompt chaining or chatbots. Drawing from real-world implementations and the latest research, it offers a clear and practical path to designing both single-agent and multi-agent systems capable of handling complex workflows with minimal oversight. Attendees will gain a deeper understanding of the core design principles behind agentic systems, the architectural trade-offs involved in orchestrating multiple agents, and the strategies required to develop tailored solutions that enhance efficiency and innovation. Whether just beginning or scaling up, participants will leave with actionable insights to navigate the rapidly evolving world of AI autonomy. ---related links--- https://x.com/michaelalbada https://www.linkedin.com/in/albada/ https://theneuralnexus.substack.com/ https://michaelalbada.com Timestamps 00:00 - Introduction by Michael Albada, Principal Applied Scientist at Microsoft. 01:14 - The Promise and Obstacles of Agentic Development. 02:37 - Defining What an AI Agent Is (and Isn't). 04:42 - Core Component 1: Tool Use and Function Calling. 06:37 - Core Component 2: Orchestration Patterns (Chains, Trees, Agentic). 08:47 - Core Component 3: Multi-Agent Systems. 09:43 - Common Pitfall 1: Insufficient Evaluation. 11:25 - Overview of specific Evaluation Tools. 12:57 - Common Pitfall 2: Lack of Observability. 13:50 - Other Common Pitfalls (Tool issues, complexity). 14:45 - The Critical Importance of Security and Safety. 15:15 - Conclusion and Future Outlook.