How to build Effective AI Agents by Diwank Tomer, Julep AI

Hey there! Welcome to the first episode of our series, “AI Agents Podcast.” In this series, we explore the world of AI agentic platforms and the companies leading this space. Today, we’re speaking with Diwank, the founder of Julep AI, a notable player in the field.




Genesis of Julep AI – Crafting a better AI ecosystem

Julep AI was established in November 2022, around ChatGPT’s launch. Diwank with his co-founder and wife, Ishita, started this venture to make computing more intelligent and human-like. Initially, they experimented with AI capabilities by creating an AI school receptionist. However, the rapid progress in AI models like GPT-3.5 and GPT-4 led them to pivot towards building a platform for creating AI agents.

Early Products and Evolution of AI agents

The first product from Julep AI was an AI school receptionist designed to manage calls efficiently. As AI models became effective, the team shifted focus to a sales assistant for Shopify stores, capable of handling customer queries and facilitating purchases within a chat interface. With the launch of powerful models like GPT-4 and Llama, Julep AI moved towards developing more effective human-like agents, drawing inspiration from the AI character Samantha in the movie “Her.”

Surprises in AI Capabilities

Working with AI agents often brings unexpected moments. While these models can perform complex tasks with remarkable precision, they sometimes stumble on simpler ones. For instance, Julep AI once witnessed its model trying an overly complicated method to calculate a delivery fee, highlighting the unique nature of AI intelligence.

Reliability and Planning

Ensuring reliability in AI agents involves a mix of deterministic steps and the flexibility of large language models (LLMs). A structured workflow approach ensures consistency and predictability, facilitating the management and troubleshooting of AI tasks. Julep AI focuses on defining clear workflows and integrating deterministic processes with LLMs for optimal performance.

Unique Selling Proposition of Julep AI

Julep AI distinguishes itself by emphasizing workflow-based systems rather than relying solely on multi-agent collaboration. While multi-agent systems have their strengths, Julep AI believes that many tasks are better handled through a step-by-step workflow approach. This method enhances reliability and ensures tasks are performed accurately and efficiently.

Building Effective AI Agents

Key elements in building effective AI agents include:

  1. Reasoning: Using advanced models with strong reasoning capabilities.
  2. Workflow Design: Crafting detailed workflows for task execution.
  3. Tool Management: Structuring and defining tools and context for optimal model performance.
  4. Continuous Improvement: Regularly updating and refining models to improve accuracy and reduce errors.

Future Features and Community Engagement

Julep AI is set to introduce new features, including tasks in general availability and expanded memory capabilities. These improvements will enable more personalized and long-running AI agents. The team also plans to enhance integration with platforms like Composio.

Vision for AI Agents in the Future

In the next five years, AI agents are expected to automate up to 95% of cognitive tasks, transforming many roles and creating new opportunities. While there is some concern about job displacement, the potential for innovation and productivity is immense.

Conclusion

Thank you for joining us, for this insightful conversation with Diwank. Stay tuned for more exciting discussions on AI agentic platforms and the future of AI technology. Watch the full podcast on YouTube.