Overview
You can now refine your AI Agent instructions with a single click. Improve instructions generates a rewrite of your prompt applying prompt engineering best practices, right inside the behavior editor. You stay in control: the suggestion comes fully editable before you accept it.
What's new?
- Improve instructions in the AI Agent behavior editor. With one click, you get a rewritten version of your prompt, clearer and more structured, following prompt engineering best practices.
- You can replace your current instructions with the suggested version or keep refining from it.
- Together with Test behavior, this completes the full AI Agent builder workflow: Build, Test, Improve, and Create, all in one place.
How does this help?
- Clearer, more structured prompts without starting from scratch: you begin from your own instruction and get an organized version instead of rewriting everything by hand.
- Faster iteration while building behaviors: apply prompt engineering best practices with a single click and test the result without leaving the builder.
- Full control over the final behavior: the suggestion is editable before you accept it, so you fine-tune every detail before applying.
Use cases
1. An IT ticket triage agent that wasn't classifying well. The team built an agent to categorize requests coming into the internal support process, but the original instructions were a loose list of examples with no hierarchy. Clicking Improve instructions, the agent gets a restructured prompt with explicit category criteria (access, hardware, software, network). Triage now correctly separates a password reset request from an infrastructure failure, without relying on the requester to pick the right category in the form.
2. A contract analysis agent written by someone who isn't a prompt expert. A Legal analyst described in plain language what they wanted the agent to extract from each contract connected to the process. The instructions worked inconsistently. With one click, the prompt is rewritten with a clear task structure and output format, and the agent starts extracting term, value, and termination clauses predictably, without the analyst needing to learn prompt techniques.
3. Iterative building of a lead qualification agent. While building an agent to score leads coming through the start form, the builder tests a first version in Test behavior, sees the agent confusing fit signals, clicks Improve instructions for a more structured base, edits the suggestion, and tests again. The Build, Test, Improve loop happens entirely inside the builder, shortening time to production.
4. Standardizing agents built by different teams. In a company running AI Agents across several processes, each team wrote instructions in its own style, producing behaviors that were hard to maintain. Running each agent through Improve instructions, the prompts converge toward a common best-practice structure, making maintenance and review easier for anyone who didn't build the agent originally.
Most relevant for
- AI Agent builders
- Pipe Admins
- Operations teams
- Users new to AI Agents
Documentation
For the full step-by-step, check the AI Agents article in the Help Center.
