Watch what a coding agent does across a week and the work is lopsided. A few changes that genuinely need a model to hold several files in its head. Then hundreds of renames, import fixes, one more test case, a config value threaded through three files. The same model priced for the first group handles the second.
Databricks shipped smart routing in Unity AI Gateway to deal with this. Their published numbers are more than thirty percent lower cost per task, thirty-five percent on internal benchmarks, and fifty-six percent on public ones while matching Opus 5. Those are their measurements on their benchmarks. I have not reproduced them.
What I find more interesting than the saving is how the routing decision gets made. A cheap fast model reads the task first and labels it: what parts of the code change, what evidence exists, what the failure mode looks like, whether the fix is localised or spread out. The router starts at a medium-sized model and moves up or down from those labels.
Most routing logic I have seen teams write is a length heuristic and a keyword list, which is why most of it gets switched off within a month. Difficulty is not a single number. A one-line change in an untested file with an unreadable stack trace is harder than a fifteen-file rename, and a scalar score cannot represent that. Labels keep the shape.
One limit worth stating, because coverage of this will blur it: smart routing today is scoped to coding tasks inside Claude Code and Codex. If your bill is a support assistant answering four hundred thousand questions a month, the idea transfers and the feature does not.
The traces matter more than the savings
There is a line underneath the cost headline that I would build on. Traces for coding agents can go into Unity Catalog, with tagging and access policies over them.
That means the record of what your engineers asked a model, what came back, and which model handled it sits in the same governance boundary as your customer tables. Same grants, same lineage, same audit surface.
I have sat in a lot of rooms over the last two years where the AI conversation stalls, and it usually stalls here rather than on capability or price. Somebody from risk asks where the evidence lives and who can read it, and a vendor dashboard is a bad answer to that question. A governed trace store is answerable with tools the security team already trusts.
It also makes the cost claim checkable in your own environment, which is what I would actually do. Route your own work for a fortnight and read your own traces.
The thing I would ask in a review
A router is a component that can be wrong, and it is wrong asymmetrically.
Routing a hard task down produces a worse answer that still looks like an answer. Somebody spends twenty minutes with it before deciding to start again, and the inference cost you saved you spent several times over in attention. Routing an easy task up costs a few cents.
So the router should lean toward escalating, and the number to watch is not average cost per task. It is how often routed-down work gets redone. A team that reports a saving without that second number has measured half of it, and I would want to know which half before I took the figure to a board.
None of that is an argument against turning it on. If you are running coding agents on Databricks, turn it on and record the traces. Two weeks of your own data will settle it better than anyone else's benchmark, including this one.