# Incident Report: Degraded Binder and AI Performance

**Date:** August 6, 2026
**Status:** Resolved and monitoring

Today, we experienced degraded performance across several Binder services, particularly AI Prep, AI Planning, and AI Review. Moderate-to-complex Binder workflows were the most affected.

## What happened

During a scheduled feature update, a software bug caused excessive requests to our file-storage system whenever a customer opened a Binder. As more Binders were opened, the unexpected storage traffic placed the system under significant load.

This initially slowed storage-dependent features, including AI Prep and AI Review. The impact then spread to other services as requests accumulated and processing queues grew, causing broader sluggishness across the platform.

## Customer impact

Customers may have experienced:

- Slow or unresponsive AI Prep, AI Planning, and AI Review workflows
- Delays when opening or working with Binders
- Greater disruption in moderate-to-complex workflows
- Increasing latency across related services as requests queued

Services with limited dependency on file storage—including Analyst, Chat, and some simpler Binder workflows—continued to operate with less disruption.

## Cause

The incident was caused by a bug introduced during a scheduled feature update. The bug generated excessive file-storage activity each time a Binder was opened. The resulting load degraded storage performance and created cascading delays across dependent services.

## Resolution

We resolved the issue by revamping how our systems read and process Binders. Previously, parts of the system could attempt to rebuild a Binder or read every file in it at once. Under increased demand, this behavior placed excessive load on file storage and contributed to the cascading slowdown.

We deployed a performance upgrade that prevents all Binder files from being read simultaneously. Binder content is now accessed incrementally, reading only the files required for the operation being performed.

## Prevention

To prevent this issue from recurring, we also redesigned the underlying Binder-processing architecture. Processing no longer depends on loading or rebuilding an entire Binder before work can begin. This reduces storage demand, improves performance for large and complex Binders, and limits the effect that any single Binder operation can have on the wider platform.

We continue to monitor system performance following these changes. We apologize for the disruption and appreciate our customers' patience while we restored normal service.