Does a Rovo agent's reasoning tier (Think deeper) apply to its subagents, or only the main agent?

David Villalobos_k6btal

In Studio, the Reasoning setting (Quick answers / Think deeper) only appears on the main agent's Details page — subagents have no reasoning control of their own. When the main agent routes a request to a subagent, does the subagent execute under the main agent's selected tier and model, or does it run at a fixed default regardless? This matters for agents where all the real tool use and classification logic lives in the subagent, so Think deeper needs to reach the subagent to have any effect. Is collapsing the subagent's logic into the main agent the only way to guarantee the tier governs that work?

Sami Shaik

Sharp observation @David Villalobos , and your reading of Studio is correct: Reasoning (Quick answers / Think deeper) is an agent-level setting, and subagents deliberately have no reasoning control of their own. Here is what the official record establishes, and where it stops.

What is documented:

1. The reasoning tier and model are selected per agent. Atlassian's launch note phrases it as choosing "the reasoning tier and model your agent uses," replacing the old behaviour where all agents ran the same model with no reasoning control: https://community.atlassian.com/forums/Rovo-articles/What-s-New-in-Rovo-Agents-Your-Bi-Monthly-Update/ba-p/3271694
2. A subagent's own configurables are instructions, routing description, knowledge, and tools, and nothing else. Reasoning is intentionally absent from that list: https://support.atlassian.com/rovo/docs/add-scenarios-to-a-rovo-agent/
3. The architectural detail that resolves your question: per the create-and-edit doc, the first instructions, knowledge and tools you give an agent become its "default Subagent", and added subagents only take over when their routing triggers: https://support.atlassian.com/rovo/docs/create-and-edit-agents/

Point 3 is the important one. In this model, essentially every response is executed through a subagent, default or added. If the reasoning tier applied only to some "main agent" layer above the subagents, it would apply to nothing, and the setting would be meaningless. The coherent reading of the documentation is that the tier is a property of the whole agent runtime, and delegated work runs under it.

The honest boundary: no documentation sentence states "subagents inherit the parent's tier" in those words. So before you restructure anything, verify it on your own agent in ten minutes: take a prompt that reliably routes to your specialist subagent, run it with the agent set to Quick answers, then again on Think deeper, and compare the behaviour in the Debug responses view in Studio along with the response depth and latency. The difference (or its absence) answers the question for your instance definitively.

And please do not collapse the subagent's logic into the main agent for this. You would be trading away the two documented benefits of the architecture, each subagent's own clean context window and the summarized returns that keep the parent's context lean, to solve a problem the documentation suggests you do not have. If you want the inheritance confirmed in writing, the in-product Feedback option or a support ticket will get you Atlassian's definitive word, and I would genuinely like to hear what comes back.