Agentic system design emphasizes modularity, atomic agents, low-code tools, traditional tech stacks for complexity, and preparing for purpose-built LLMs and evolving AI UX.

Time to code is reducing by the day, and it is highly likely that the tooling or tech stack you use will change before you complete your project. Here are some principles to consider when designing agentic systems:

  • Leverage Object-Oriented Programming Concepts: Use abstraction, interfaces, stubs, implementations, and extensions to build scalable and maintainable architectures.

  • Modularize MCP Servers & Agents: Implement them as independent containers or serverless functions.

  • Make Agents Atomic: Each agent should focus on a single, well-defined function.

  • Leverage Low-Code Tools: Use tools like n8n and Langflow for rapid prototyping, much like Figma for UI design before moving to React.

  • Stick with Traditional Tech Stacks for Complex Workloads: For example, use React, Python, and PostgreSQL running on a hyperscaler, or ETL tools like AWS Glue or Azure Fabric to move data.

  • Prepare for Purpose-Built LLMs: Just as AWS introduced purpose-built databases tailored for different enterprise needs, expect purpose-built LLMs to emerge.

  • Continuously Learn from Innovators: Absorb content from sources like Y Combinator and AI tinkerers—not necessarily for app ideas, but to understand how designers and developers approach building them. Remember, the AI user experience is distinct from the UI user experience and goes beyond just chatbots.