It's been some time since we did a basics of AI agents. This is a refresher session on the same.
The session is conducted by Bala Panneerselvam. Bala is a founding member of Applied AI club, founder of ZORP with 17+ years in technology and product.
If you've missed the session or if you'd like to go through it again, here's the session video - https://youtu.be/UPY77vYjlv0
Here's the notes from the meeting:
Meeting Purpose Introduce AI agents and demonstrate how to build a basic research assistant using N8N
Key Takeaways
- AI agents differ from AI assistants by their ability to take actions using tools, not just provide information
- Agents typically consist of an LLM, memory systems, and tools/integrations
- Building agents is simpler than traditional development, with platforms like N8N enabling no-code/low-code creation
- A basic research agent was demonstrated using N8N, integrating Google search, document creation, and email notifications
Topics AI Agents vs AI Assistants
- AI assistants (e.g. ChatGPT) provide information based on internal knowledge and memory
- AI agents can take actions using tools and integrations, not just provide information
- Agents typically consist of an LLM, memory systems, and tools/integrations
- Examples: meeting schedulers, research assistants, coding assistants
Use Cases for AI Agents
- Open-ended problems requiring multiple steps and back-and-forth (e.g. scheduling)
- Multi-step processes (e.g. researching and summarizing many articles)
- Self-improving systems that can learn from feedback
Building AI Agents
- Simpler than traditional development
- Key steps: Define clear objective, choose platform, start with small use cases
Platforms: Langchain, N8N, Relay.app, Agent.ai (HubSpot)
- Can be built with no-code/low-code tools or programming frameworks
- Research Assistant Agent Demo
- Built using N8N platform
Components:
- Outline agent to structure research topics
- Research agent to gather information on each topic
- Google Docs integration to compile results
- Email notification of completed research
- Demonstrated basic workflow, though some refinements needed (e.g. looping through topics)
N8N Platform Overview
- Visual workflow builder for creating automations and agents
- Connects various tools and services (e.g. LLMs, search, document creation, email)
- Good for personal/small-scale use, may have limitations for large production deployments
- Workflows exportable as JSON for portability
Next Steps
- Refine the research assistant demo (implement looping through topics)
- Explore human-in-the-loop options for agent workflows
- Watch previous session on N8N for more advanced usage examples
- Consider appropriate tools based on scale (N8N for personal/small, custom code for production)
Here's the entire recording of the session.