New features, improvements, and what is shipping across the platform.
With the AI Evaluation Platform, you can test Agents and models in batches against your own criteria. Knowledge base administrators can also require approval before updates to a company knowledge base go live. The AE CLI adds community evidence tools, version-pinned retrieval and Parquet ingestion.
Automations can now run on several days each week or several dates each month, including the month's last day, and send results to configured external channels. The AE CLI adds member groups and resource migration.
Shared workspaces keep a team's sessions, files, Agents, Skills, memory and automations in one place, with tasks, a message center and easier knowledge base upkeep.
MCP Management lets an Agent reach into the platforms and tools your team already works in, so it does the work inside a conversation instead of only describing it. Three built-in platforms are ready in a session, and integration runs both ways under system, company, and personal control.
A new A/B testing module lets you run several versions of your product at once, split users fairly, and get a plain answer on which one wins. Roll the winner out in one click, or revert instantly if something looks wrong.
You can now put any Agent, or a whole Agent Team pipeline, on a schedule: hourly, daily, weekly, or monthly. The recurring work runs on time on its own, so no one has to remember to trigger it.
Connect specialist Agents into one automated pipeline, in series or in parallel. Multi-step work that used to pass between several people now runs end to end from a single trigger.
Agentic SDK gives the agents embedded in your app a runtime trace of every skill, tool call, and cost, tied into the same funnel as your user behavior. Replay failures, attribute spend, and see where users drop off.
Upload your tracking plans, metric definitions, and playbooks into the Agentic Engine. Every in-scope Agent then answers with your team's own definitions instead of generic guesses.
The biggest Agentic Engine update yet: your Agents now answer using your company's own definitions, turn good practices into shared skills, run end-to-end workflows as a team, and give leadership the governance and cost guardrails to roll them out safely.
A new command-line tool that brings the full Agentic Engine to your terminal. Install a skill pack and your coding agent, Claude Code, Cursor, or Trae, can run AE commands on its own.
Agents now connect to third-party tools through secure OAuth, starting with Slack, and run recurring work like data inspections and reports on a schedule instead of only on demand.
The tools and methods your agents use are now a browsable, shareable library. Submit a skill, get it approved, and the whole company can reuse it, so no one rebuilds the same thing twice.
New admin controls for running Agents at company scale: unified member and permission management, an isolated sandbox for every user, and token-based cost caps at both the company and individual level.
