AI Your Way: MCPs vs Skills vs SubAgents πŸ€–πŸš€ - Sam Basu - NDC Oslo 2026

AI Your Way: MCPs vs Skills vs SubAgents πŸ€–πŸš€ - Sam Basu - NDC Oslo 2026

Title history
  • AI Your Way: MCPs vs Skills vs SubAgents πŸ€–πŸš€ - Sam Basu - NDC Oslo 2026 β€” observed on 20261028-0955 UTC (YouTube)
πŸŽ™ Sam Basu πŸ‘₯ 230K πŸ“… October 28, 2026 ⏱ 59 min πŸ‘ 57 πŸ“„ tutorial 🧭 2026-10-28
Available in: English (current) FranΓ§ais

Keywords

MCPAI SkillsSubAgentsContext WindowAgent Workflows

Summary

In this conference talk at NDC Oslo 2026, Sam Basu, a developer advocate, explains how developers can bring contextual information to AI systems using Model Context Protocol (MCP), Skills, and SubAgents. He introduces a restaurant analogy: AI agents are chefs, MCP tools are raw ingredients, skills are recipe cards, subagents are sous chefs, and the developer is the restaurant owner. The talk emphasizes the importance of guardrails and context to prevent generic AI behavior and ensure productivity. Basu demonstrates practical examples using .NET and VS Code, showing how to create MCP servers, use instruction files, and set up subagents for code review. He discusses the significance of context windows, cost management, and pitching agents against each other for better outcomes. The presentation concludes that a combination of these tools enables developers to make AI work their way, enhancing automation and code quality.

142 words

Critical Evaluation

Value of the Information & Strength of the Argument

The talk provides valuable practical insights for developers, with clear explanations and a memorable restaurant analogy. The speaker makes a strong case for using a combination of MCP, skills, and subagents to enhance developer productivity. The argumentation is based on real-world examples and personal experience, though not heavily supported by empirical data. The emphasis on guardrails and context is well-argued, and the step-by-step demonstrations strengthen the credibility of the recommendations.

Scientific Rigor, Source Quality, Title Accuracy

The talk relies on the speaker’s professional experience and industry trends. It references the Model Context Protocol and its official SDKs, but does not cite specific research papers or detailed official documentation. The presentation is of moderate rigor, typical for a practitioner conference talk. The title is a good match: it clearly announces the comparison of three key concepts, which is exactly what the talk delivers. No formal sources are cited beyond the MCP website and conference links.

163 words

Title / Content Match

The title accurately reflects the content, as the talk directly compares and explains MCP, skills, and subagents.

Quality & Reliability

7/10

The talk provides practical, up-to-date information about MCP, skills, and subagents from an experienced developer advocate. The examples are concrete and the speaker demonstrates real usage. However, the content is based on the speaker's personal experience and industry trends rather than peer-reviewed research, and no formal sources are cited.

Key Moments

Cited Sources

  • NDC Conferences β€” Conference organization that hosted the talk.
  • NDC Oslo β€” Official event site where the talk was recorded.
  • Model Context Protocol β€” Official documentation and resources for the MCP standard mentioned in the talk.

External References

Contribution & Novelties

The talk provides a clear comparative framework for understanding MCP, skills, and subagents, with practical .NET examples and a memorable analogy. It underscores the importance of contextual engineering in AI agent workflows and offers actionable advice for developers. The original contribution lies in its synthesis of these concepts and its focus on developer-oriented guardrails.

Pour aller plus loin :

112 words

Radar Profile

The profile shows high information quantity and moderate quality/reliability, indicating a broad but not deeply rigorous talk. The technical level is adequate for a developer audience, with a balanced emphasis on practical application.

Reliability 7/10