What if your AI agent could actually help?

Watch the live demo that connects docs, CRMs, and support tickets into one AI assistant.

AI-powered assistants are everywhere, but most still hit roadblocks when it comes to fragmented enterprise data. In his lightning talk at All Things Open, Kevin Blanco from Appsmith shares a live demo of how his team built an agent that integrates real-time data across tools like Salesforce, Notion, Confluence, and Zendesk—no complex MLOps required.

Kevin starts by calling out what’s holding most enterprise AI initiatives back: Messy data and app sprawl. Companies are juggling legacy systems, structured and unstructured data, and too many SaaS tools. Instead of forcing enterprises to clean it all up before they can even experiment with AI, Kevin argues for meeting companies where they are—building assistants that plug into their existing tools and workflows.

To show how this works in practice, he walks through a RAG (retrieval augmented generation) implementation that starts simple: A chat interface connected to docs and knowledge bases. Then he layers in more sources—Notion for product updates, Zendesk for support tickets, Salesforce for customer contracts, and Confluence for internal documentation. The assistant responds with grounded answers, citing sources and flagging when it doesn’t know something. With human-in-the-loop touchpoints, the assistant can even generate Jira tickets or respond to customers directly, all without switching tabs.

The key, Kevin explains, is speed and simplicity. The entire workflow—querying multiple systems, creating tickets, posting in Slack—can be built and deployed in minutes using Appsmith’s agent platform. It’s self-hostable, secure, and designed to embed directly into the tools your teams already use.

Read more: AI agents are here—are your skills ready?

Key takeaways

• Enterprise AI needs to work across stacks: Connecting to multiple systems like CRMs, docs, and support tools is essential to build useful agents.
• Grounded responses reduce hallucination: By tying AI output to real citations, agents can deliver more reliable answers and flag gaps in knowledge.
• Human-in-the-loop is a feature, not a flaw: Automated workflows work best when they include decision points and final reviews.

Conclusion

AI agents are only helpful if they can plug into the messy, real-world workflows teams already use. Kevin Blanco’s demo shows how you can skip the complex setup and start building practical, context-aware assistants that actually improve response times and reduce friction—no sunglasses required.

More from We Love Open Source

• What is Artificial Intelligence (AI) and the three things it does well
• AI isn’t scary—lack of knowledge is. Here’s how to stay ahead
• Discover Goose: Automate your developer setup with this AI agent
• Why AI agents are the future of web navigation
• AI agents are here—are your skills ready?