Agents moving into workflows

The old version of workplace artificial intelligence sat in a chat box and waited for questions. The new version is being wired into the software people already use, so it can turn a document into a chart, a prototype, or a draft app without anyone copying text into another tab. (atlassian.com) Atlassian showed that shift on April 8, 2026 inside Confluence, its document and knowledge product. It launched “Remix,” which converts page content into visuals like charts, infographics, diagrams, and summaries from the editor toolbar. (atlassian.com) Atlassian also added three partner agents for Lovable, Replit, and Gamma. Those agents take Confluence content and send it into those tools to make working prototypes, starter apps, and presentations without manual copy-pasting or custom integration work. (atlassian.com) That sounds like a product tweak until you look at where Atlassian says its agents already live. Rovo agents can run in chat, in automation rules, while you edit in Confluence or Jira, and in a separate Studio app, which means the agent is being treated more like a built-in coworker than a search box. (support.atlassian.com) Atlassian’s own support pages say these agents can use plugins and reach data from Atlassian apps and connected third-party apps, while still following the user’s permissions. Marketplace partners can also ship their own agents for admins to install. (support.atlassian.com) The plumbing behind that matters more than the demo. Atlassian’s public Model Context Protocol server is a cloud bridge that connects Jira and Confluence to external large language model tools, integrated development environments, and agent platforms, with OAuth 2.1 or API tokens enforcing the same access controls the user already has. (github.com) A leaked codebase from Anthropic’s Claude Code gave a second view of the same industry direction. Security researchers said the March 31, 2026 package leak exposed about 512,000 to 513,000 lines of TypeScript across roughly 1,900 files, revealing not just a model wrapper but a full agent system. (varonis.com) (zscaler.com) Those researchers described agent orchestration, tool-call loops, retry logic, multi-agent coordination, permission checks, telemetry, feature flags, and local execution hooks. In plain English, vendors are spending engineering time on the software that decides what the model can touch, what it should do next, and how humans can watch it. (varonis.com) (zscaler.com) That is why the center of gravity is moving away from “ask the bot a question.” The money is shifting toward agents that sit inside the workflow, inherit the company’s permissions, call other tools, and leave an audit trail when they act. (support.atlassian.com) (github.com) (zscaler.com) The practical change for workers is small on the surface and big underneath. A product spec in Confluence is no longer just a page to read; Atlassian now wants it to be the source file that can become a chart, a slide deck, a prototype, or a starter application from inside the same workflow. (atlassian.com)