If you’ve been building AI Teammates, you’ve probably hit the character limit on the Behavior prompt and wondered: what now, and why is it even capped?
Short answer: split your prompt across tasks, and be very specific in each task’s title and description about what it contains.
When you ask the Teammate something, the LLM receives a context made of three things: the LLM’s own system prompt, Asana’s system prompt, and your Behavior prompt. There’s a hard limit on how much can fit in that context, so Asana caps the Behavior prompt to keep things manageable (and probably to control cost and latency too).
The interesting part, based on a conversation I had with someone on the Asana side: your Behavior prompt is sent to the LLM on every single request. But the tasks you reference from within that prompt are not. Instead, when you ask a question, Asana evaluates which referenced tasks are likely relevant and only loads those into the context.
This changes how you should structure things:
- Keep the Behavior prompt itself focused on rules, tone, and routing logic.
- Offload bulk content (knowledge, examples, procedures) into separate tasks.
- Give each of those tasks a precise, descriptive title and summary so the retrieval step can correctly decide whether to pull it in.
If your task is called “Reference notes” and the description is vague, the Teammate may skip it when it actually matters.
Curious if others have tested this and found the same behavior.
We are i.DO, an Asana Solutions Partner, and we document our learnings using AI Teammates every day!
