Where does "agentic AI" break your production stack first?

Everyone's shipping agents. Few teams have a clean answer for what happens after the demo works.

Tool calls go live. Budgets get fuzzy. Compliance asks for evidence you can't reconstruct from logs. Evals live in one tool, traces in another, and "governance" is still a Notion doc someone updates after an incident.

We're building Traccia around a simple loop for production agents:

• Observe what ran

• Evaluate before you promote

• Control behavior at runtime

• Govern with audit-ready evidence

Same OpenTelemetry spans. Not four disconnected products.

Genuinely curious where this hurts you most today:

1. Visibility (can't debug a bad tool call fast enough)

2. Evaluation (no real gate before promote)

3. Control (observe after the fact, can't block in time)

4. Governance (auditors want proof you don't have)

Drop your stack and your pain point. LangSmith, Langfuse, homegrown, nothing yet. All useful.

We launched Traccia on Product Hunt today as an AI agent control plane. If you want context on what we're building before you comment, here it is: https://www.producthunt.com/products/traccia

Not looking for cheerleading. Looking for the honest stories. Those are the ones that shape what we ship next.