Some questions deserve a moment's thought; most don't. Making you pick a "reasoning effort" for every message put that judgment on you. Today we're shipping Auto — the assistant sizes up each question and sets its own effort.
Execution plan
1. jira_search_issues — open tickets assigned to James
2. confluence_search — pages mentioning those tickets
3. Summarise by priority
jira_search_issues
confluence_search
You have 4 open tickets — 2 urgent. ONE-42 blocks the release and has a Confluence spec that changed yesterday. Want the diff?
2 tools · planned first
Tool turns start with a plan drafted against the real app catalog.
Effort that fits the question
Turn on Auto (it's the new default) and every message is judged on the fly:
Easy — a quick fact, a rename, a small edit — answered fast, with no wasted deliberation.
Moderate — a normal working question — gets a normal amount of thinking.
Hard — a thorny problem, a multi-step task — gets the deep, slow reasoning it actually needs.
Why it's a real win
Cranking reasoning to maximum on every turn is slow and expensive; leaving it low means hard questions get shallow answers. Auto gives each turn the right amount — so your everyday questions stay snappy and your hard ones get the horsepower, without you touching a setting.
It pairs with automatic model routing: an easy turn goes to a fast, cheap model with light reasoning; a hard turn gets a stronger model thinking harder. You ask the question; the right amount of effort shows up.
Still yours to override
Prefer to decide? The effort control still lets you pin low, medium or high for a turn. Auto is the default, not a cage — it just means the sensible choice is already made for the 95% of messages where you'd rather not think about it.
Try it yourself
Every model, one memory, and agents that work the way you do.