Breeze Research: from question to action plan
Wednesday, November 4, 2026
What it is
Breeze Research is a deeper investigation capability in Breeze Assistant. Instead of answering a question in one shot, Breeze builds a plan, works through the relevant data step by step, and returns a structured report with findings and recommended next actions.
Why it matters
Surface-level answers are not always sufficient for questions that require synthesizing data across multiple sources, identifying patterns, and determining a course of action, which typically demands significant manual analysis. Breeze Research performs that analysis end-to-end and delivers a decision-ready report, not just a data point.
The Take
The approval needs to be at the start and at the end.
Per the panel, Breeze Research is a deeper investigation capability in Breeze Assistant. Instead of answering in one shot, Breeze builds a plan, works through the relevant data step by step, and returns a structured report with findings and recommended next actions. It is live for all hubs and tiers.
Chris cared about the plan part more than the report. Last week at the agent hub HUG he asked the HubSpot team whether Breeze conversations that build agents and workflows start in plan mode, and the answer he got was yes. He had asked for that a while ago: stop suggesting that we can do this in one shot. He also pointed out that the work management side of HubSpot already has conversations shaped like the screenshot we looked at on air.
The rest of his take was about where approval belongs, which is the real lesson for anyone running Breeze Research. He builds his own work around a plan command that has him review each section, and his conclusion is that the agents already have the expertise: “The approval needs to be at the start and at the end.” Piling on do-not instructions backfires. “Then the focus is on what not to do, instead of what to do.” The agent spends its effort worrying about the fences and then hands you a long report of things it did not do. His advice is to say what you want done and what good looks like, review the plan, and let it work.
Casey's version of the same instinct came from a conversation with Sarah Lane-Hawn: tell AI to show you first, before it does anything, especially when you are moving fast. Zach asked whether that belongs in your standing instructions rather than being repeated in every prompt. Chris said it works until it doesn't, and it comes back to whether you are putting people, or agents, in a position to succeed.
What to do: try it this week on one question that needs several data sources and a recommendation, not a single number. Read the plan before you let it run, and judge the report on whether the next actions are ones you would actually take. The tape did not cover setup steps or any credit cost for Breeze Research, so check the HubSpot product update page before you run it at volume.
Watch Us Discuss This
The full segment from the show, us working through this update start to finish.
FAQ
What is Breeze Research?
Per the panel, it is a deeper investigation capability in Breeze Assistant. Breeze builds a plan, works through the relevant data step by step, and returns a structured report with findings and recommended next actions, instead of answering in one shot.
Who gets it?
The panel lists all hubs and tiers, and the status is live.
What kind of question is it for?
Per the panel, questions that need synthesizing data across multiple sources, finding patterns, and deciding on a course of action, which usually takes significant manual analysis. A single data point does not need it.
Why did Chris care about the plan?
He had asked HubSpot for Breeze conversations that build agents and workflows to start in plan mode instead of suggesting a one-shot answer, and he was told at the agent hub HUG that they do.
What was the advice about prompting it?
Chris's take: put approval at the start and at the end, and tell the agent what to do instead of stacking do-not instructions, which breeds uncertainty and a report of things it did not do.
Produced by Value-First Media

