Zendesk Bridge brings trusted service wherever work happens

Service no longer has a single front door. It can begin in a customer’s AI assistant, an employee’s workspace, or an AI agent that a business has built for itself. In fact, Gartner recently found that customers are now 3x more likely to use their own AI tools of choice when getting support than a company provided bot.

Wherever the work begins, customers and employees want their problem solved or their task completed, with less waiting and less effort. For AI agents to do that, they need access to the right service tools and data to take action.

That access needs to come with the same safeguards businesses rely on in Zendesk: authentication to establish who is acting, permissions to define what they can do, and activity records to show what happened. Those protections need to hold wherever the work begins.

That is why we built Zendesk Bridge, Zendesk’s headless platform experience. It helps teams deliver better customer and employee outcomes by bringing Zendesk’s service tools, AI agents and context into the AI experiences they already use. Access stays governed, and the agent’s Zendesk activity remains visible.

Bringing Zendesk wherever work happens

To put those service capabilities to work, teams can start with prebuilt connections to tools such as Slack, Codex, Claude Code, Lovable and Notion. Teams building their own AI agents can connect those to Bridge too.

We’re also launching Zendesk plugins in the marketplaces for ChatGPT, Claude and Microsoft Copilot, so teams can discover and connect Zendesk directly from the AI tools they already use. Plugins for Amazon Quick and Gemini are coming next. Across these connections, admins control access, that respect existing Zendesk permissions, and activity is logged.

Bridge enforces rules an AI agent cannot negotiate

Giving AI agents access to service systems raises an important question: how do you give them the flexibility to get work done while keeping control over what they can access and which actions they can take?

An AI agent might receive a broad instruction such as “make this right for the customer.” To act on it, the AI agent still has to decide which account to access, what action to take, and whose permissions to use. A traditional interface narrowed those choices with buttons and forms. An AI agent has to translate intent into operations on its own, and understanding what someone wants does not grant permission to act.

The AI agent's instructions cannot be the enforcement mechanism either. An AI agent can encounter malicious text that persuades it to ignore a rule. If your access rules live inside its reasoning, a cleverly worded message can argue them away.

That is why Bridge enforces access rules within Zendesk. The AI agent decides how to approach its job. Bridge enforces access rules within Zendesk, outside the AI agent’s reasoning. The agent can decide how to approach a task, but Zendesk checks each requested operation against the agent’s approved permissions. Those checks apply even if the agent is prompted to ignore a rule.

You decide who acts and what can be accessed

Setup happens in the Admin Center. An admin registers a connection for the AI agent and describes the job it will do. Admin Copilot recommends Zendesk tools and AI agents for that job, and the admin reviews the recommendations and chooses what to allow. The admin also decides who can use the connection, including specific user groups. These settings can be updated later without changing the external AI agent.

The person using the AI agent signs in to Zendesk through OAuth, and their existing Zendesk permissions continue to apply. Every request is checked against both the capabilities approved for the connection and that person’s permissions. Connecting an AI agent does not expand their Zendesk access.

With Bridge, an AI agent can use more than 400 governed tools through the Zendesk Model Context Protocol (MCP) server, with coverage across tickets, messaging, knowledge and users. Bridge also supports Agent-to-Agent (A2A) protocol so external AI agents can delegate support to Zendesk AI agents when more specialized or complex service work requires it. The same permission checks apply to both. These open protocols give businesses flexibility to choose which AI providers they use with Bridge.

Here is what this looks like in practice

Let’s say a product team uses Codex to decide which recurring customer problems deserve engineering attention. The evidence sits in Zendesk tickets, and until now someone had to export them and assemble a summary by hand.

With Bridge, Codex can review those tickets using the Zendesk tools approved for its connection. It identifies a recurring checkout problem: customers change the currency and return to an empty cart. Links to the original tickets let the product manager check the reports and understand the customer impact.

Codex brings this evidence together in a draft Jira issue. The manager reviews and approves the draft before Codex creates the issue. Engineering gets a clear starting point for addressing the checkout problem so customers can complete their purchases. The review can become a weekly job, surfacing new patterns and preparing Jira drafts for approval.

When anyone asks what happened, you have the answer

As the review becomes a recurring job, admins need to see how the agent is using its Zendesk access. Bridge observability brings every connection's activity together in Admin Center. Admins see successful, blocked and failed transactions. Each transaction shows the AI agent, the person it acted for, the tool it called, the policy decision and the result. The log links to the associated Zendesk events, with the payload and timeline available for closer inspection.

Blocked calls are part of the record as well. You can see where Zendesk refused a request, not only what was completed. That history helps admins review whether an AI agent’s access still fits its job.

Service that helps the next customer too

The feedback review shows how customer requests can reveal problems in the product itself. Bringing that evidence to product and engineering gives them a chance to address the cause and spare future customers the same problem.

To entrust ongoing work to an AI agent, a business needs to know who is acting, control what it can access and inspect what it did. We built Bridge to give teams that control, so they can use Zendesk’s service capabilities in their chosen AI experiences and help customers and employees get the results they need with less effort.

The best support request is the one that never gets filed. Zendesk Bridge is a key part of this journey

Zendesk Bridge is in early access, and we are inviting teams to help us shape it. Put your AI to work for better service outcomes with the practical tools to identify relevant service opportunities and starting points for extending service through the AI tools of your choice.