# AI Agent Improvement Recommendations | Calvin Improve

> Move from knowing what went wrong to knowing what to change. Improve recommends changes to your agent’s prompt, configuration, or tools based on observed misalignment with the mission.

Canonical: https://www.usecalvin.com/improve/
Publisher: Value Metrics
Updated: 2026-10-25

## Start with the mission and the evidence

The mission defines what your agent should accomplish. Approved commitments make that mission checkable. Observe identifies where behavior or outcomes diverge from those commitments; Improve uses that misalignment to recommend changes.

Missing evidence stays unverified. A gap in access to a business system is not proof that the agent failed, and a proposed change is not proof that the problem is fixed.

## Recommendations for prompts, configuration, and tools

- Prompt: clarify instructions, constraints, or the order of actions so the agent follows the approved commitment.
- Configuration: recommend changes to agent settings or workflow rules that contribute to the observed misalignment.
- Tools: recommend changes to how the agent uses tools or handles their results so it can complete the intended task.

## Example: a confirmation without a booking

In this illustrative scenario, a voice agent says an appointment is booked, but Observe finds no matching calendar entry. The agent did not meet the commitment to create the booking before confirming it.

Improve could recommend changing the prompt to require a successful calendar response before giving confirmation, or changing tool-result handling so a failed booking routes to a human. The right recommendation depends on the evidence of what went wrong.

[Read the illustrative booking scenario](https://www.usecalvin.com/blog/appointment-booking/)

## Your team reviews and deploys every change

Improve recommends changes; it does not automatically rewrite or deploy your agent. Your team reviews the proposed prompt, configuration, or tool change against your business rules and decides what to apply.

Keep a named owner responsible for the change and the evidence needed to evaluate it.

## Close the loop with Simulate and Observe

- Review the recommendation and the observed mission misalignment behind it.
- Apply the approved change to your agent.
- Rerun the relevant scenarios in Simulate before release.
- Use Observe to check whether real outcomes align with the mission after deployment.

[Test the change with Simulate](https://www.usecalvin.com/simulate/) · [Verify production outcomes with Observe](https://www.usecalvin.com/observe/)

## Related references

[Calvin homepage](https://www.usecalvin.com/) · [Simulate](https://www.usecalvin.com/simulate/) · [Observe](https://www.usecalvin.com/observe/) · [Improve](https://www.usecalvin.com/improve/) · [Blog](https://www.usecalvin.com/blog/)