Sat with a team a while back who'd done the observability part properly. Traces flowing, dashboards up, latency percentiles, error rates, the lot. A tidy setup.
Asked them one thing. The agents had been live three months. What had they changed about the server because of what they'd seen?
Long pause. Then, honestly, not much. They'd looked at the dashboards plenty. They hadn't known what the dashboards were asking them to do.
That's not a failing on their side. A dashboard shows you what happened, and it shows you well. A spike here, a slow tool there. What it won't do is tell you the spike is the model confusing two of your tools, and that here's the wording that fixes it.
There's a step missing between the data and the decision. On human-facing products a product analyst does that step. Reads the funnel, forms a view, comes back with: drop this field, the pricing page is scaring people, move the signup button. The chart was never going to say any of that on its own.
Agent-facing surfaces don't have that person yet. The data's there, more of it than human products ever got, and cleaner, because the protocol hands it over structured. The someone who reads it and comes back with change this, and this is why, mostly isn't there.
That's the gap. Most of what I work on now sits inside it.
If you build a surface that agents use, and you want to know how they actually use it, Vesta is in early access.