When Monte Carlo renamed itself an agent trust platform back in May, I read it as a conviction move rather than a hedge, and the open question was whether the product would catch up to the brand. Two releases this week start to answer that. Agent Lineage and a new Cost Agent both point the lineage graph Monte Carlo already runs at a fresh problem, and that graph is the most interesting asset they have right now.

Agent Lineage ties a bad agent output back to the data that caused it

Agent Lineage maps an agent’s run as a graph and connects it to the underlying data. You get the execution path of an agent, with its errors and latency overlaid on the steps, and you get bidirectional lineage that links the agent to the upstream tables it reads and the downstream agents a data change would affect. When an agent does something wrong, the hard question has always been whether the fault sits in the agent’s logic or in the data it was handed, and teams have mostly answered it by reading traces by hand. Tracing the bad output back to an upstream freshness or quality problem in the same view is the most direct answer I’ve seen to that question.

It’s also the part of the agent trust story that’s genuinely hard to copy. Plenty of tools can monitor an agent’s behavior and flag unusual output. Far fewer can tell you the output was off because a table three hops upstream went stale. After all, watching agents and tracing data have lived in separate products. Monte Carlo already owns the data half, so folding the agent half into the same lineage is a sensible use of what they’ve built. Where I’d like to see more is how it handles the messier failures, the ones where the data was technically fine but wrong for the decision the agent made, and I’d want a few production teams to walk me through how often the root cause it surfaces is the actual root cause.

Cost Agent puts Monte Carlo in the budget conversation

It’s the same engine Monte Carlo built to watch your warehouse for reliability, except now it’s looking for where you’re burning money. Some of that is storage. It finds the tables nobody reads and the dead-end pipelines still quietly running up a bill, then ranks them by how much you’d save by killing them off. The rest is compute, and the query history gives it away, the expensive queries and the warehouses that are just sized wrong. Either way, you come out with a ranked list of what to deal with first.

None of this took much new building, either. The data Monte Carlo collects to catch a reliability problem turns out to be the same data that exposes waste, and it’s been sitting there all along. The cost piece came close to free.

And it changes who’s buying. Reliability sells to data engineers. Cost is what gets the person who owns the bill to pay attention, and those people are jumpy right now, with AI and warehouse spend climbing. Now there’s a reason Monte Carlo is in both rooms.

You don’t even have to open Monte Carlo to use it. It runs in Claude Code and Cursor via the agent toolkit, so an engineer can request a cost fix directly in the editor. Nobody on that side of the house wants to babysit another dashboard, and Monte Carlo clearly knows it.

The lineage graph keeps Monte Carlo in front

Both releases work because Monte Carlo spent years building deep field-level lineage, and that graph is hard for the rest of the field to reproduce. Pure agent-observability tools can watch how an agent behaves, but can’t reach back into the data that fed it. The real test is the big data platforms, which own the data and are now building the agent runtime on top of it, so they’re the ones who could close the gap. That’s why Monte Carlo is racing to own the agent trust framing before the platforms settle it themselves. For now, it spans both the data and the agents, a position few hold. Selling reliability, cost, and agent trust as one platform opens a far bigger market than data observability alone.

Monte Carlo more or less built the data observability category, and now they’re going after agent trust the same way. The difference this time is speed. They renamed the company in May and had a real product behind it a month later. My guess is they don’t slow down from here.