Safely manage your Zendesk from the AI assistant you already use, via the Deltastring MCP. Beacon configuration platform
← Back to news

Service Incident - July 9, 2026 - Support | All Pods

Zendesk

A feature flag deployment failure in Zendesk's Side Conversations module rendered PDF attachments inaccessible across multiple pods for nearly five hours on July 9, 2026. Between 11:59 and 16:49 UTC, agents encountered "Failed to load PDF attachment" errors when attempting to open PDFs within Agent Workspace, though non-PDF attachments remained functional. The root cause was straightforward: a code change intended to be controlled behind a feature flag was only partially gated, allowing the modified PDF download/streaming logic to affect production traffic beyond its intended scope. Zendesk reverted the change at 16:49 UTC, restoring normal functionality with no data loss or exposure.

The incident exposes a critical gap in deployment discipline that should concern support operations teams relying on Zendesk's infrastructure. Feature flags exist precisely to decouple code deployment from feature activation, yet this incident demonstrates that partial gating—where only portions of a change are protected—can nullify that protection entirely. For teams already managing complex Agent Workspace configurations across multiple pods, this raises an uncomfortable question: how many similar partially-gated changes exist in production systems, waiting for the right conditions to fail? The mitigation roadmap Zendesk outlined addresses the mechanics (atomic gating, expanded test coverage, improved monitoring), but the incident itself suggests a process breakdown rather than a technical one.

The broader implication concerns operational resilience during a period when CX platforms are increasingly central to business continuity. A five-hour window where agents cannot access PDF attachments in side conversations—a feature commonly used for case documentation, evidence sharing, and customer communication—represents material disruption to support workflows. For larger support organisations running high-volume operations, this translates directly to SLA impact and agent friction. Zendesk's response was appropriately swift, but the incident underscores why CX teams should be stress-testing their fallback procedures and understanding their platform's deployment cadence, particularly as vendors accelerate feature releases to compete in the agentic AI space.