September 2026 Product Updates
Let Copilot find improvements in real conversations, create and test agents in one flow and follow up on customer conversations as threads.
Let Copilot learn from real conversations
Ask Copilot to analyze your agent’s recent production conversations — the last 30 days by default, or a date range you choose — and tell it what to focus on. Copilot reports what it found and, where the evidence supports it, suggests specific changes. Before you apply a suggestion, a diff preview shows exactly what will change.
You can start an analysis from the Agent Editor or while reading a conversation.
Create, test and improve in one flow
Creating an agent now takes three steps: choose a template, describe what you need, then review the generated draft. Codeer saves it as an unpublished first version with Live Test open beside it.
After your first test reply, Copilot can suggest improvements with that test in context. Apply the suggestion, save a new version and re-run the same question to compare.
Follow up on conversations as threads
Sessions from the same LINE, Messenger, Instagram or Slack conversation now appear together as one thread, so your team can read the history in one place. Threads show what’s unread for you, can be marked as pending for the whole team to follow up, and refresh automatically. When a thread is linked to a known customer, one click opens their CRM record.
AI drafts can use more of your tools
AI drafts can now use spreadsheet data and HTTP request tools when writing suggested replies. HTTP tools stay off for drafts unless you allow each one individually, and editing a tool’s request settings turns that permission off again.
Smaller improvements
- Website sources in the Knowledge Base show when they were last successfully synced.
- Tool instructions can now be up to 2,000 characters (300 is still recommended), so there’s room for conditions and exceptions.
- Spreadsheet answers can list up to 200 rows when asked, or a specific range such as items ranked 20 to 30.