# Improvement proposals that show their evidence

**URL:** <https://community.solace.com/t/improvement-proposals-that-show-their-evidence/4944>\
**Category:** Solace Agent Mesh\
**Created:** [November 9, 2026, 4:00am UTC](https://community.solace.com/t/improvement-proposals-that-show-their-evidence/4944 "2026-11-09T04:00:14Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![Tamimi](https://yyz1.discourse-cdn.com/solace/user_avatar/community.solace.com/tamimi/32/2190_2.png) [@Tamimi](https://community.solace.com/u/Tamimi)\
**Post date:** [November 9, 2026, 4:00am UTC](https://community.solace.com/t/improvement-proposals-that-show-their-evidence/4944/1 "2026-11-09T04:00:14Z")

</div>

Hi all,

Agent Mesh reads your agents’ traces, and when the same problem keeps coming up, like a tool call that keeps failing, it records a finding. If a configuration change can fix it, a model drafts an improvement proposal: a specific edit to the text that caused it, with a short explanation of what happens, the likely cause, and the next step.

What’s new is how much a draft has to prove before you see it. It has to state a cause and cite the trace steps that show it. The drafter can search successful calls to the same tool for contrast, which tells it whether the tool is broken or just being called the wrong way. Then a second model reviews the draft in a separate conversation and looks for reasons it’s wrong. If it can’t hold up, it’s revised once or discarded.

So you’re not just told “change this prompt”. You can check why.

For the demo I gave a Northwind Freight support agent an instruction with a mistake in it, so ocean quotes failed. I read the finding, proposal and explanation through the platform API, applied the change myself, and the proposal was marked superseded. Just under five minutes:

[](https://www.youtube.com/watch?v=1B7iSW2-CRE)

Did you find the evidence format helpful? Is there anything you’d want a proposal to show before you’d trust it? Tell me in the comments.

To try it out, [download Solace Agent Mesh](https://solace.com/products/agent-mesh/). The GitHub repo is here:

> **[GitHub - SolaceProducts/solace-agent-mesh: Platform for building and running AI agents across...](https://github.com/SolaceProducts/solace-agent-mesh)**
>
> Platform for building and running AI agents across the full lifecycle — build, test, deploy, monitor, improve. Hosts your agents, runs their tools, and coordinates them over an event mesh, with sessions, artifacts, auth, evaluators, and observability built in.