From the show
Another Orange Morning Recap: July 27, 2026
Monday brought the show's HubSpot product managers back to the table. Davis and Becca returned the week Agent Hub went to public beta, and what was scheduled as a look at the new launch turned into a candid, hour-long working session about it, with Casey and Chris asking the questions a real operator asks and the two PMs answering in the open.
Here is what we got into: what Agent Hub actually changed about the navigation and where your Breeze context now lives, why Custom Assistants became Projects and the naming knot that created, how HubSpot is thinking about context and knowledge vaults, and a longer conversation about the good kind of friction between a person and an AI.
Agent Hub, One Week In
Becca opened by framing the launch the way the team sees it, as perfectionist product folks giving their own work a hard review, now building in public beta on the road to Unbound. Davis added the upside of being public: they can build in the open and talk about it in the open, instead of keeping everything under wraps.
The most useful part for an operator is the navigation. If you select Breeze expecting Breeze Studio, look again. Breeze Assistant is still the sparkle you know, but Agent Hub is now its own place, holding both HubSpot's first-party agents and the custom ones you build, and you reach it from a new icon. Davis's fun fact landed the moment: he did not know what the Agent Hub icon would be until about a week ago, and it turned out to be a small baseball diamond with a runner on second. AEO shows up here now as a content agent, part of the same reshuffle.
Why it matters: The furniture moved, and some of it got renamed. Budget a few minutes to relearn where things live before you need them mid-task, and know that Breeze Assistant and Agent Hub are two different doors, not one.
For more details: read our full take on Agent Hub.
Projects, Context, and the Naming Problem
Custom Assistants are now Projects, the scoped container for bespoke configuration. Becca made the case by pointing at how the rest of the industry uses the pattern, naming Claude and Anthropic as examples of projects being well loved as a way to hold a scoped context. The trouble is that HubSpot already has a Projects object in the CRM, so the same word now means two different things in the same product. Davis named it plainly as a collision, and Becca did not dodge it: naming, she said, is one of the hardest things they do. It was a rare thing to watch, a product team taking the feedback live instead of defending the choice.
The deeper thread was context. The group walked the difference between foundational context that applies across your whole portal, currently a private beta, and scoped knowledge vaults that a specific agent pulls from. Becca made it concrete with a competitive example: an agent that pulls the right battle card when a call flags a competitor, then links the matching case studies in the follow-up. Her framing for the two kinds of context stuck, always-on information as a pool you swim in versus a one-time instruction you hand over once.
Why it matters: The value of an agent is mostly the context you give it. Deciding what is foundational, what is scoped, and what is a one-off instruction is the real work, and it is worth doing deliberately rather than dumping everything into one place.
The Good Kind of Friction
The back half turned into the most honest part of the show. Casey shared that his first Agent Hub build felt like magic: he asked it to build a routing flow that sent sixty percent of leads to Tricia and forty to him, and it just worked. A more complex build that afternoon went differently, and the lesson was the same one that keeps coming up, that the payoff shows up when Breeze asks planning questions first instead of building on the first instinct.
Chris pushed the point into a thesis. Every interaction with an AI either kills a little trust or earns a little, and clicking build only to get something wrong is how the trust dies. The instinct to expect click button, get cookie is the trap; the real skill is the troubleshooting muscle that using AI well demands, and the context and expertise you bring are what produce a real result. His image for a build that starts slightly wrong and compounds was sharp: you were three degrees off, and then you went a thousand miles, so you end up a long way from where you meant to be. The PMs took it as exactly the product feedback they came for.
Why it matters: The move is not to hunt for the perfect prompt. As Chris put it at the close, come in and tell Breeze what you want, then have the conversation. The friction of planning and correcting is not a bug in working with AI. It is where the good results come from.
The Rest of the Day's Updates
The show stayed on Agent Hub start to finish, so the rest of Monday's release slate did not get pulled forward on air. If you want the full list, from the Revenue Agent to the new sequence task associations to the Pipelines API V1 deprecation notice, they are all logged with our read on the HubSpot updates feed.
The Sign-Off
You probably already own the value you are looking for in HubSpot. You just need to wake up to it. We will see you tomorrow morning.
Produced by Value-First Media