Let's build an AI agent - Phil Nash - NDC Copenhagen 2026

Let's build an AI agent - Phil Nash - NDC Copenhagen 2026

by Phil Nash

Applied Sciences & Engineering AI — Models & Research UYQArtificial intelligence (AI)UYQFGenerative artificial intelligence
🎙 Phil Nash 👥 227K 📅 July 6, 2026 ⏱ 49 min 👁 5K 📄 tutorial 🧭 2026-08-13
Available in: English (current) Français

Keywords

agentLLMtoolsJavaScriptMCP

Summary

In this NDC Copenhagen 2026 talk, Phil Nash demystifies AI agents by building one from scratch in JavaScript. He starts with a simple echo bot, then progresses through basic text generation with an LLM, adding short-term memory via chat history, and finally implementing a tool-use loop that allows the agent to call functions. The core of an agent is a while loop that checks for function calls, executes them, and feeds results back to the model until the goal is achieved. Nash demonstrates adding a date tool and discusses the importance of tools in bridging the gap between the model’s text generation and real-world actions. He introduces the Model Context Protocol (MCP) as a standardized way to connect agents to services, and mentions his work on Open RAG and Docling at IBM. The talk emphasizes that the ‘magic’ lies in the model’s ability to generate tool calls, while the harness code orchestrates the loop. Nash concludes by showing how agents can be composed, with one agent acting as a tool for another, and encourages the audience to start building their own agents.

182 words

Critical Evaluation

Value of the Information & Strength of the Argument

The talk provides high practical value by walking through the actual code for building an agent, making the concepts tangible. The argumentation is clear and logical, building from simple to complex. Nash effectively explains the underlying principles, such as the agent loop and tool declarations, and supports his points with live demonstrations. He also references authoritative sources like Google’s paper on agents and Simon Willison’s definition, adding credibility. The presentation is engaging and accessible, making complex topics understandable.

87 words

Title / Content Match

The title accurately reflects the content: a step-by-step guide to building an AI agent, from basic text generation to tool use and multi-agent systems.

Quality & Reliability

8/10

The talk is a live coding tutorial that demonstrates building an AI agent from scratch, with clear explanations of core concepts and practical implementation. The speaker is a developer relations engineer at IBM with relevant expertise. The content is technically accurate and well-structured, though it lacks formal citations and in-depth theoretical discussion.

Key Moments

Markers derived by PSI from the transcript: the creator did not define chapters.

Cited Sources

  • NDC Conferences — Conference website mentioned in the description.
  • NDC Copenhagen — Conference website mentioned in the description.

Concurring Sources

  • Google's whitepaper on agents — Referenced in the talk as the source of the definition of an agent.
  • Simon Willison's blog — Referenced for the simplified definition of an agent.

Contribution & Novelties

The talk provides a clear, hands-on demonstration of building an AI agent from scratch, demystifying the underlying mechanics. It emphasizes that the core is a simple loop and that tools are essential for real-world interaction. The speaker shares practical insights from his experience at IBM, such as using Gemini and MCP.

Pour aller plus loin :

  • Model Context Protocol (MCP) — Official documentation for MCP, the standard protocol for agent-service interaction.
  • Open RAG — IBM’s open-source stack for building RAG applications, mentioned by the speaker.
  • Docling — IBM’s document processing tool, also mentioned.
  • Simon Willison’s blog on agents — Articles and insights on AI agents from a well-known developer.
  • Google’s paper on agents — The whitepaper referenced in the talk, defining generative AI agents.

124 words

Radar Profile

The radar profile shows high scores in information quantity, quality, and reliability, with a slightly lower technical level, indicating a well-balanced tutorial that is both informative and accessible. The speaker's expertise and clear explanations contribute to a strong overall rating.

Reliability 8/10

💬 No comments were provided for analysis.