Just do it. (let your tools think for themselves) - Robert Chandler
Introduction to Agentic MCPs 00:00
- Robert Chandler, co-founder and CTO of Wordware, discusses building reliable AI agents.
- Emphasizes the importance of giving tools time to think and the promise of automated systems that can act on behalf of users.
Challenges with Current MCPs 00:28
- Current agents are often slow, expensive, and unreliable, lacking the ability to seamlessly interact with tools.
- An anecdote illustrates the inefficiencies of early MCPs, where a simple Slack message took five minutes and ended up misdirected.
- Many MCPs are merely low-level wrappers around APIs not designed for language models, leading to messy outputs and context pollution.
Improving Tool Agency 02:58
- Proposes adding more agency to tools rather than keeping them small and limited.
- Compares tools to a team of Avengers, advocating for specialized agents that can work together effectively.
- Suggests blurring the line between tools and agents to enhance functionality and reliability.
Demonstration of Wordware's MCP Toolbox 04:08
- Introduces Wordware's new MCP toolbox that allows users to build agentic MCPs.
- Demonstrates a competitor analysis tool that integrates with Twitter and Notion, showcasing the ability to tailor outputs to specific needs.
- Highlights the ease of creating and using multiple interconnected tools within the toolbox.
Conclusion and Future Potential 06:10
- The toolbox enables highly reliable, repeatable, and aligned tasks, enhancing the capabilities of generic agents.
- Encourages users to leverage the toolbox to build flows and utilize agentic MCPs for various applications.