improved
AgentFlow
Edit your Playbook directly, it's exactly what your AI Agent follows

Before, the AgentFlow setup wizard showed two versions of your instructions: what you wrote in "Your playbook," and a separate "AI-optimized" copy the Agent actually used, rewritten behind the scenes. If a reply went wrong, there was nothing on screen you could trace it back to and fix.
Now there's just one Playbook, and it's the version your Agent reads. Everything stays editable, changes save automatically as you type, and nothing gets rewritten in between.
The Playbook is split into five sections, each with a one-line description so you know what goes where:
- Personality: how the Agent sounds, tone, style, manner
- Behaviour: background the Agent needs, your company, products, its responsibilities, and reply templates
- Unsupported Media Type Reply: what's sent when a customer shares something the Agent can't read, like a voice note or video
- Guardrail: situations to watch for and what the Agent must not say or promise
- Scenarios (previously "Playbooks"): the specific situations the Agent handles, step by step, the only section where typing "/" opens the actions menu
One heads-up: if you already had an agent running, your Playbook will now show the rewritten version it's been running on all along, not necessarily your original wording. Nothing is lost, but it's worth a quick read-through. Long instructions also need to be split across sections now rather than pasted into one box.