Product update / 27 Jul 2026
Launch deeper research agents
Some questions need more thinking than a quick pass can give. Deep Mode puts real compute and a team of agents behind them, and checks the answer before you see it.
Not every question deserves the same effort. A quick fact check is one pass. A technology deep dive across two semiconductor companies is hours of reading, comparing and checking. The new effort setting next to the message box lets you say which one this is: Fast, Normal, or Deep Mode.
Deep Mode is you telling Primer the question matters, and Primer putting more intelligence behind it. It breaks the question into pieces and puts a separate research agent on each one, all working at the same time. You watch them beside your chat, each task named and timed.
Nothing is a black box. Click any task to see the exact brief it was given, the searches and documents it worked through, and what it found, with sources.
When the work comes back together, an independent reviewer checks the finished answer against what you asked. Not the agent marking its own homework: a separate run whose only job is to judge whether you got what you wanted.
What changed
The effort selector next to the message box now has Deep Mode. Pick it and Primer puts more compute behind your question: parallel research agents working live beside the chat, and an independent review of the final answer.
Why it matters
The hardest research questions were always limited by how much thinking one agent could do in one pass. Deep Mode removes that ceiling. More ground covered, every piece shown, and a second pair of eyes before you rely on it.