September 2026 — AI Employees can now hand off work to each other, and MCP Servers are fully released

This is an archived release note. See What's New for the latest release.

September 1, 2026: AI Employees can now hand off work to each other, and MCP Servers are fully released

This update makes it easier to build AI Employees that work together instead of one doing everything. An orchestrator AI Employee (one that coordinates other AI Employees to get a task done) can now call on another orchestrator, and the AI Employees it delegates to automatically get the right Tools for their job instead of having to guess. MCP Servers, a way of connecting Ema to outside systems and data, are now fully released and easier to set up. AI Employee summary cards, which give you a quick overview of how an AI Employee is performing, are also fully released with accurate numbers and no export limit.

Below are the headline changes:

Capability	What it means for you
AI Employees can delegate work to other AI Employees, including other orchestrators	If you build an orchestrator AI Employee (one that breaks a task into pieces and hands them to other AI Employees), it can now delegate to another orchestrator, not just to simple AI Employees. This lets you build deeper, multi-layered automations, like an AI Employee that manages several specialist AI Employees, each of which manages its own sub-team. Each AI Employee that gets delegated a task now automatically gets access to the correct Tools for that task, instead of trying to figure out which Tools it can use. And if a delegated AI Employee needs to ask the user something mid-task (a "connect prompt," used for example when it needs the user to log in to an integration), that request now shows up properly in the conversation instead of leaving the task stuck. See Core Actions for details.
Frontier apps connect to Autopilot automatically	Frontier apps are the custom applications you build on Ema. Previously, you had to manually connect a published app before Autopilot (Ema's in-app assistant) could use it. Now, any frontier app you have access to is automatically available for Autopilot to use as a Tool, with no setup step required. See Calling AI Employees from App Code for details.
New ⌘K search shortcut	A new ⌘K search shortcut lets you jump straight to any AI Employee, Integration, App, or Group by typing its name, instead of clicking through menus to find it.
MCP Servers are fully released, with a smoother, more reliable setup	MCP (Model Context Protocol) Servers let Ema connect to external tools and data sources in a standardized way. This connection type is now fully released (out of beta) in every environment. Setting one up is more reliable too: connection details register automatically across your fleet of AI Employees, you can edit and validate the server URL when you first create it instead of finding out later that it's wrong, and it supports vendors that require sign-in (OAuth) even when they don't follow the usual setup pattern. If a connection fails, the error message now tells you specifically what went wrong and how to fix it, instead of a generic failure. See MCP Servers for details.
AI Employee summary cards, better dashboard exports, and more accurate analytics	AI Employee Summary cards and Workflow-step cards, which give you an at-a-glance view of how an AI Employee performed on a conversation, are now available to every workspace. Summaries are generated automatically as soon as you publish, no manual step needed. When you export data from your dashboard, you can now download real XLSX (Excel) and JSONL files, and the previous limit of 10,000 rows per export is gone, so large workspaces can pull their full data. Analytics like session counts and ticket outcomes have also been corrected: some workspaces were seeing inflated numbers before, and those now reflect actual usage. See Chat Metrics for details.