Skip to main content

August 12, 2026

Highlights​

One big idea lands this release: building an Agent should feel like describing it, not like filling in a form.

  • Agent Builders: create an Agent by describing it. A new Simple mode asks what you want your Agent to do, then builds a working first version for you. No settings to study before you start.
  • Agent Builders: refine it in plain language, and try it as you go. Tune your Agent through sections that read like questions: what it does, what it knows, what it can do, who can use it. Describe a change and let the AI apply it. A test chat sits right beside the setup, so you can feel every change. The Advanced editor is one click away when you want the full control panel.

Simple mode rolls out progressively, so not every workspace will see it on day one.

New features​

Simple mode: build an Agent by describing it​

Building a good Agent used to mean facing every setting at once. Simple mode turns that around: start by describing the job in a sentence or two, optionally give it a name, and you get a working first draft to refine.

Everything after that is grouped the way you would explain it to a colleague:

  • Identity: the name, what it is for, and how it greets people
  • What it does: your Agent's job description. Edit it directly, or describe the change you want ("always answer in French") and let the AI apply it, with a preview before you accept
  • What it knows: drag in files, add website links, or import from Google Drive, OneDrive, Jira, Confluence, and GitHub. You can also reuse a knowledge source another Agent already uses
  • What it can do: plain-language switches for searching the web, creating images, analyzing spreadsheets, the Workspace tool, and smarter file answers. Sensible settings are chosen for you
  • Connections: look through the tools your organization already offers, grouped by category, and enable one with a click or an API key
  • Skills: link reusable instructions from your library so your Agent can pull them in when a task calls for them
  • Model and Who can use it: the default model works well for most, and sharing is a single choice you can change anytime

A Try it tab sits beside the setup, so you can chat with your Agent while you build, save a change, and restart the test chat to feel the difference.

Simple mode is not a separate, lesser Agent. It is the same Agent, shown more simply. Anything advanced (API functions, custom inputs, the full toolbox, custom MCP servers) is still there in the Advanced editor, and each section tells you when something is being managed there. You can move between the two modes freely, and we remember which one you were last using.

Look for Simple mode as we enable it for organizations in production.

Improvements​

  • Honest status on every step. A step that fails now says Failed rather than showing completed wording, so you can tell at a glance what actually worked.
  • Steadier streaming in long conversations. Message history work no longer runs on every incoming chunk, which keeps long chats responsive.
⌘J