Happy New Year everyone!
I wanted to share a new architectural pattern I’m exploring in Retool which was the focus of my Holiday Shipping Spree. I'm calling it Chasks+ (Chats, Tasks + workflows/agents/UIs, etc)
The goal was to move beyond the "static dashboard" and build a true Action-Oriented Operating System. While I’m currently using it for a Sales use case, the underlying engine is industry-agnostic.
- What if your AI agents could start a chat with you for those specific actions that move the needle in your business?
- What if your AI agents could leverage personalized UIs for the most important actions you need to take?
The Architecture: "The Headless OS"
Instead of a rep/user hunting for data, Chasks+ uses an end-to-end background process:
- The Engine: Retool Workflows, with agents within them, scan data sources (Postgres/Email/Messages) for high-priority signals.
- The Inbox: When a signal is hit, it creates a "Chask"—a grounded action item that carries all the necessary context (IDs, files, history).
- The Dynamic UI: This is the part I’m most excited about. Leveraging @abusedmedia's custom component, the AI doesn't just send text; it sends instructions to render specific UI elements, in the specif example: Discovery Plans, Risk Maps, or Battlecards, directly in the chat.
- The Pinning Pattern: Users can "promote" these widgets from the chat to a persistent section. This keeps the relevant tools visible during execution while the chat continues to provide context.
Here's the demo: https://youtu.be/ituIc7wRU0c
Why this matters:
We often build apps where the human has to find the data. This pattern flips that. The data finds the human, and the AI prepares the exact interface needed to handle it.
I’m currently running this for Sales Operations (Meeting Prep, Deal Nudging), but the the framework could swap in widgets for Legal Review, Inventory Management, or Customer Support without changing the core app logic.
