Show recap — Monday, September 14, 2026

Another Orange Morning Recap: September 14, 2026

Monday of Unbound week, and the show opened in purple. Chris Carolan hosted from a laptop instead of the usual desk, Joshua Oakes took the Monday seat, and Casey Hawkins was out for the morning. The release board was long for a Monday, with a run of betas flipping to live a couple of days ahead of the official announcements, which is what a conference week looks like from the product updates page. Before the board, a reminder: the Donut Happy Hour returns Friday morning, September 18, across from the convention center, organized by Madelyn Donovan while Mark Hans sits this year out.

Here is what we got into: image generation inside Breeze Assistant and the last-mile edit it still cannot do, granular tool permissions and a new batch of connectors, configurable CRM cards in the sales extension and the principle of keeping your tools where you work, one-line AI previews on the activity timeline and the trust question underneath them, the pipeline object tags API and why segments should carry the complexity, a pipelines API safety net, and a Will It Breeze that asked Breeze to fix Chris's navigation.

Image Generation Moves Into the Chat

Image generation in Breeze Assistant is live: generate an image from a prompt, then refine it across turns, make it more vibrant, change the background, use a reference image for inspiration or a faithful edit, and pull brand kit colors and fonts, all without leaving the conversation. Every result lands in File Manager. Generation was already possible over in Content Hub; editing an image inside the assistant chat is the new part.

The wish list showed up immediately. Brand templates would make this useful on the first day rather than the tenth, and the working theory in the room was that a Breeze project holding a set of reference images might be the way to get there until templates exist. Whether the assistant is files-aware enough to start from one was not settled on air.

The comparison everyone reached for was Canva, which is bad at generating images and very good at the thing AI is still bad at: composition, the fine tuning of how pieces sit on a card. That is the last-mile edit. A stakeholder asks to change one word, and a double-click in a layout tool beats a prompt longer than the change. Until generation can do that, the two tools live side by side.

Why it matters: this is a free feature that will almost certainly cost credits, and the room said so out loud. It is also a small preview of the pattern for the week: capability arriving in the assistant first, with the iterative edit as the real feature. One note for the product updates team, offered on air: the new Open tool button on this update went to the portal home page rather than the tool.

Permissions and Connectors, the Plumbing Under the Agents

Granular tool permissions for Breeze Assistant and agent connectors is live. You can decide which tools an agent is allowed to use, remove the write tools if you want a read-only integration, and trim unused tools to improve performance. The announcement is a little unclear on whether that lives on the connector or on the individual agent; the reading on the show was the connector.

Keeping up with connector scopes has been a pain, and not because the idea is wrong. The tools update constantly, the permissions and scopes change with them, and the current experience walks you into a page that looks like a checklist and then sends you somewhere else to turn it on. The hope is the same one that keeps coming up this month: soon enough you tell Breeze to set it up.

Alongside it, connectors for Breeze adds Fathom, Xero, Box, Canva, Gong and Slack to Agent Hub and Breeze Assistant, with Google Drive, Gmail, NetSuite and Zoom named as coming through the rest of the year. The closing sentence about providing a cohesive vision to all the tools your business thrives on read like a line lifted from Wednesday's keynote, and the show is going to hold it to that.

Why it matters: permissions are the boring half of agents and the half that decides whether a security team lets them run. A tool you can turn off is a tool you can turn on.

Configurable CRM Cards in the Sales Extension

Joshua's pick, and a personal one. Configurable CRM cards in the sales extension is in public beta: admins choose which cards appear in the sidebar, in what order, and which properties show inside each, separately for email views like Gmail and Outlook and for prospecting views like the Chrome extension. The settings path is long, and team-specific views would benefit from one place to manage them, but the capability is the point.

The history here is Outlook. Getting the sidebar to look like HubSpot for a rep who lives in their inbox has never reached parity, and you usually do not win the argument to move them. So the workarounds pile up: automated workflows, data put in places it does not belong, all so it shows up in a sidebar. This is a step toward putting any object there.

It also maps onto a principle Joshua has named on the show before, the second of his AI adoption guidelines: keep your tools where you work, because you do not put your hammer in the refrigerator. HubSpot data needs to be available in the other tools, and the other tools' data needs to be available in HubSpot. Some reps run on a task queue and want automation to surface the next step when a proposal is accepted; some ride the top of the inbox and clear it as their queue. A business needs parity for both work modes.

Why it matters: two cards are not configurable yet. The task card is fixed at the bottom, and the deals-on-this-email-thread card cannot be changed. The first one is fine. Tasks outnumber every other record type associated with a contact, company or deal by an order of magnitude, so keeping them at the bottom keeps them from pushing everything else nine scrolls down. Named on the same board but not opened: track leads directly from the sales extension sidebar, also public beta.

One-Line AI Previews, and the Trust Underneath

One-line AI previews for calls, emails, notes and meetings is in private beta. Every activity on the CRM timeline gets a one-line summary, so instead of reading a long email or a four-paragraph call summary you see "discussed bulk pricing for fifty-plus seats, send quote by end of day," scan the record in seconds, and click in only when you need the detail.

The appeal is obvious to anyone who has read a Breeze meeting summary that opens with what the customer requested in August of last year when the question was what happened in the last two weeks. Being able to skim the feed line by line is a quality-of-life change. Joshua is waiting on his beta invitation.

Then the room went looking for the how-it-works section and found a what-happens section instead: enroll, open a record, scroll the timeline, summaries appear. The considerations are more useful. Enrollment is per user until public beta, the initial load can take up to a minute, only text longer than a line gets summarized, and sensitive data portals are excluded. What the announcement does not say is how the summary is made, and it cannot be a black box or it will not work. Show it to a user, have it be wrong once, be unable to explain why, and the feature is over.

The failure modes are the familiar ones from every meeting summarizer: a statement attributed to the wrong speaker, a rhetorical question read as a decision, a nested hypothetical about a timeline recorded as a commitment. Communication is hard, and this is not unique to Breeze. The right posture is a guidepost, something that tells you which activity to expand, never something you decide from.

Why it matters: the fix is context, and the phrase Joshua used for it was sculpt your data. Eighty percent of any business is the same as every other business; the twenty percent that is unique is where summaries go wrong, and you have to tell the AI what matters, which two similarly named products to keep straight, what a word like development means in your industry when it can mean ten things. Transcripts arriving through the public API, covered Friday, are part of the same story. The show also took a passing look at the transcripts API as the structured-data side of that path.

Object Tags on Every Pipeline, and the Case for Segments

The CRM pipeline object tags API went live on September 11 and had not come up with Joshua in the room. He has projects where it will get its first real workout. The two things it delivers are the expansion of tags beyond deals to other pipeline objects, which brings them up to parity, and the ability to deploy and update tags by API, which takes tag setup off the manual punch list on a deployment.

The recap of what tags are for was worth the segment on its own. Tags normalize complexity, both simple and complicated rules, into elements that are easy to query and easy to see, so the data model stays clean while humans still get the shortcuts they need. Once a deal is tagged high priority, Breeze or Claude or an API call can ask for high-priority deals without reconstructing what high priority means from a close date, an amount, a priority field and a target-account list.

The stronger point was about where the criteria should live. Most people configure the same complex conditions inside the tag, then again inside lead scoring, then again somewhere else, and have to remember to update all of them. Any time a rule has to be maintained in two places, it should be a segment. Segment membership can be a tag's inclusion criterion, a scoring input, and a mailing list filter, which makes the definition permanent and portable. Segments are the meta layer for the messy dimensionality of what counts as a high-value deal, a high-risk project, or a high-value listing.

Why it matters: an implementation that can rise above its properties and reason with these meta tools is on the right track, and the API makes that repeatable across portals. The caution from the show's earlier deal-tag episodes still holds: build a staging view so you can see what the tags look like, and do not go crazy with them.

A Safety Net in the Pipelines API

Pipelines API: validate pipeline and pipeline stage usages took two readings to understand and was welcome once it landed. Deleting a pipeline or a stage through the public API now follows the same validation as deleting one in settings: if it is in use, the request is blocked.

Why it matters: a few recent betas have retroactively enforced constraints on the API that the UI already enforced, and this is the good kind. Nobody on the show had deleted a stage with records in it by API, and now nobody can.

Will It Breeze: Fix My Navigation

The personalized navigation that arrived as Friday's major update is now in Chris's portal, and it arrived heavy: the top block that used to be bookmarks, a long tail of items filled in from usage, and three wallet icons in a row because a design system that reuses icons sensibly was never asked to survive a user rearranging it. So the test was the standard one. Any time there is a new feature, ask Breeze if it can do it for you. Full segment on the personalized navigation page.

The first ask, what do you think of my navigation, got the party line, and a detour into the product updates page. The second ask worked: based on what you know about the update and how I work in this portal, recommend a lean configuration. Breeze came back with a lean operator's navigation rather than a show-me-everything one, organized into five working zones, starting with a daily control layer of dashboards, projects, tasks, appointments and calendar. Projects meant the projects object, which is where a lot of agent work gets coordinated in this portal. It could see the saved bookmarks, which is more than expected, and its suggested groups were better named than the defaults. Where it lost the room was demoting forecast, a reminder that its context is not yet the whole picture.

Two practical things for anyone who has this beta. If you remove items from the personalized bar, the More menu at the very bottom is the old sidebar, drill-downs intact, and the pin on any item adds it back. And if you are coaching someone who just got excited about this, do not start from the menu. Walk them back through a week of their work, ask what the three biggest things were each day, and the patterns fall out; winnow those to five, make them the bookmarks, and see how it goes.

Why it matters: the classic battle here is navigation versus search, and search is not a clean answer yet. Faceted global search has improved, but changing modes between contacts, deals, settings, tools and apps is too many clicks for a rep, and searching across everything gets slow. The likely resolution is the one showing up everywhere else on the board: wherever you click, you end up in a brief conversation with Breeze.

The Rest of Monday's Board

Monday shipped more than the show could open. Also on the board dated September 14: the Reddit Ads integration in public beta, the Microsoft Ads integration now live, asking Breeze Assistant for customer agent deployment recommendations, adding the prospecting agent card to the sales extension sidebar, and using Breeze Assistant to create plays for the prospecting agent. Everything else that shipped is in the HubSpot updates feed.

The Sign-Off

Next Wednesday, September 23, at 11 a.m. Eastern, the Hubsessed team is doing Unbound Unboxed, what they learned in Boston and what they think matters to you. Friday is the Donut Happy Hour if you are at the show. 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.