Multiplayer Underwriting: Building a Real-Time Workspace for Humans and AI

Here's a scenario that happens dozens of times a day at Oyyo-powered MGAs: An email arrives from a broker. The AI Organizer immediately starts reading the attachments. While the AI processes page 3 of the loss run, the underwriter opens the email body and starts reading the broker's notes. The AI updates field after field in real-time. The underwriter spots something the AI missed — a handwritten note about a prior claim — and types it into the same form. The AI immediately incorporates that correction into its premium calculation.

No handoffs. No "wait for the AI to finish." No separate screens. Humans and AI agents work on the same document, at the same time, seeing each other's changes in real-time.

The Technical Challenge: CRDTs and AI

Making this work requires solving a hard computer science problem: how do you let an LLM rapidly update structured data while a human simultaneously edits the same fields, without conflicts or data loss?

We use Yjs, a CRDT (Conflict-free Replicated Data Type) implementation, as the synchronization layer. Every field in an Oyyo form is a Yjs document. When the AI extraction agent writes "Insured Name: Acme Corp" and the human simultaneously types "FEIN: 12-3456789", both changes merge seamlessly.

The key insight: AI agents don't need special treatment. They're just another cursor in the multiplayer document. The same conflict resolution that handles two human editors handles a human and an AI editing simultaneously.

Why This Changes Underwriting Speed

Traditional underwriting software is sequential: the AI runs, then the human reviews, then the AI runs again. Each handoff adds minutes. Over hundreds of submissions per month, those minutes become weeks.

In Oyyo's multiplayer model, the handoff time is zero. The underwriter starts working before the AI finishes. They're reviewing field 1 while the AI is still extracting field 42. By the time the AI completes, the underwriter has already validated and corrected half the output.

This is why Oyyo submissions take 8 minutes instead of 45. It's not just faster extraction — it's parallel processing between human judgment and AI capability.

The Journal: A Shared Timeline

Every action — human or AI — appears in the Journal, a unified timeline for each submission. You can see exactly what happened, when, and by whom:

• 2:14:00 PM — AI: Email received from broker, 3 attachments detected
• 2:14:03 PM — AI: Started extraction on ACORD 125
• 2:14:08 PM — Human: Opened email, reading broker notes
• 2:14:15 PM — AI: Extracted 38 of 42 fields, 4 flagged for review
• 2:14:22 PM — Human: Corrected FEIN field, added prior carrier note
• 2:14:25 PM — AI: Incorporated correction, recalculated premium

This isn't a log. It's a living document that both humans and AI contribute to continuously. When a second underwriter opens the same submission, they see the complete story — every decision, every correction, every AI action — without asking a single question.

Building for Both Audiences

The hardest part of multiplayer AI isn't the CRDT implementation. It's the UX. How do you make an AI agent's rapid-fire updates not overwhelming to the human? How do you make human corrections visible to the AI without breaking its processing flow?

Our answer: presence awareness. Just like Google Docs shows colored cursors for each collaborator, Oyyo shows which fields the AI is currently processing (with a subtle animation) and which fields the human is editing (with their cursor). The underwriter intuitively knows: "The AI is working on the left side of the form. I'll start reviewing the right side."

That's what multiplayer underwriting looks like. Not sequential. Not parallel in separate windows. Truly simultaneous, in the same workspace, with full mutual awareness.