Show recap — Wednesday, September 2, 2026

Another Orange Morning Recap: September 2, 2026

Two weeks out from UNBOUND, and nobody in the room can say the word yet without reaching for "inbound" first. Wednesday brought Zach Hussion back after a week away, which meant the whole show ran as a catch-up: here is what shipped while you were gone, and here is the part of it that is actually going to land on your desk.

Here is what we got into: HubSpot renaming a score we already had, a new look for Breeze Assistant that finally ends a turf war, Customer Agent learning to write your knowledge base for you, Revenue Agent quietly becoming useful to people who do not bill from HubSpot, and a round of Will It Breeze that turned into the most interesting answer of the morning.

Almost Ten Updates a Day, and Some of Them Are Renames

The morning opened on a running joke that is starting to feel like a real observation. Earlier this year the theory was that once the AI work settled down, the update stream would slow. It has not. It is UNBOUND season and the release list is running close to ten a day, which is roughly the point where you want a second monitor just for product updates.

Some of that volume is genuine new capability. Some of it is renaming. CX Score is now Track Customer Experience Score, or CXS, because we all needed another acronym. The card in one portal had already updated and the card in the other still said CX Score, which tells you how fresh the change is.

The score itself is worth understanding even if the name keeps moving. It sits on the support ticket and reads customer satisfaction across every kind of interaction the ticket touched: rule-based bot, Customer Agent, and human rep, each scored separately and rolled up. That breakdown is the interesting part. It lets you ask a question that was hard to answer before, which is whether handing more volume to bots and agents moves the experience up or down.

Why it matters: Post-ticket surveys come back from about three and a half percent of tickets, and the people motivated enough to fill one out are usually the unhappy ones. A score the system derives from the whole interaction is a different instrument than a survey, and it is aimed at a real problem: managers drowning in data points with no way to read team performance at a glance.

Riding alongside it is Track Quality Assurance Score, or QAS, which scores the rep side of the same conversation. Both are the same bet: AI reading unstructured interactions at a scale humans never could, so somebody can take a breath and ask whether the thing is actually working. Half the battle in support is being everywhere at once. The other half is knowing afterward whether it went well.

Worth noting for anyone who wants the longer version: the CXS conversation on this show goes back weeks. Casey has the site set up so you can search anotherorangemorning.com for CX and land on the update page with the clips attached, including the fuller discussion from a few days after the first pass. That is the point of the update pages, and it worked live on air.

A New Look for Breeze Assistant Ends the Turf War

The Breeze Assistant redesign moved the panel from the right side of the screen to the left, and put an entry point next to the search bar. That sounds cosmetic. It is not, and Chris made the case for it on air.

The first thing you notice is suggested actions on open. Catch up on product updates. Look at these deals. They appear to be user-specific rather than portal-wide, which means the personalized HubSpot homepage people have been describing for a year is arriving in pieces, right now, through Breeze.

The second thing is the part that actually changes your day. The old right-side placement fought with the preview sidebar. If you were looking at a CRM record in preview and wanted to ask Breeze about it, the two panels were competing for the same real estate. Moving Breeze to the left ends that. You can hold a record and a conversation about that record on screen at the same time.

Why it matters: Most AI adoption problems are not model problems. They are moments where the product made it awkward to ask the question. This removes one of those moments.

Customer Agent Will Write Your Knowledge Base, and That Is the Easy Part

Customer Agent can now generate FAQ-style knowledge from your resolved support tickets and import it directly. No manual writing. PII is scrubbed automatically. You review each generated answer before anything goes live, and a brand new portal can get a working agent running without authoring a knowledge base article first.

Zach put his finger on why this lands. Tribal knowledge is one of the oldest problems in any business. People know the answer, or they know how to get the answer, and asking them to put it on paper is close to impossible. Marketing asks a subject matter expert for a few lines about the thing they know best, and the expert says who has time for that, and then writes a novel of an email an hour later. If the knowledge is already being created every day inside tickets, capturing it there and turning it into articles is the version of the task that people will actually do.

Then the show found the word that matters, which is review. Every one of these agents ships with a human approval step, and almost nobody has decided who that human is. It might be a marketing manager. It might be sales ops, or RevOps, or the HubSpot admin, or IT, or the partner doing the work. Different organizations put accountability for HubSpot in different places, rightly or not, and the ones that never name a person get whatever the agent guessed.

That is where it gets expensive. An answer a rep gave once, for one account, for a reason, gets saved into the knowledge base as if it were policy. Then AI treats it as policy. Your biggest customer talks their way into an out-of-warranty exception because they spend five million a year with you, somebody says sure, no problem, and five instances later the system has concluded that is the rule for everyone.

Why it matters: The answer is not to hand it all to the one person who is good at answering things, which is the reflex, and which the show named out loud as the wrong move. It is to switch what the humans are responsible for. The expert stops drafting and starts confirming. That only holds if confirming is easy. If review means copy-pasting into a doc and trading comments back and forth, no knowledge gets built anywhere.

The practical shape is monthly, not daily. Twenty suggested knowledge moments come in, you agree on one. You do not have to take all of them. And a fair warning that came with it: if your data quality and your processes are not in order yet, this is a shiny object. Valuable when done right, and close to impossible to do right on a weak foundation. There is a Knowledge Base Agent in private beta sitting right next to this one on the release list, and the two feel related even if HubSpot has not said so.

Revenue Agent Now Works If You Do Not Bill From HubSpot

The headline on Revenue Agent this week is one clause: it works even if you do not bill from HubSpot. Sync invoices from QuickBooks, Stripe, Xero, or a custom integration off any third party accounting system, and the agent manages that portfolio the same way it manages HubSpot native invoices. It watches every open invoice, decides which ones need attention and when, builds a personalized outreach plan per customer, sends reminders, and listens for replies and invoice changes to recommend the next step.

Zach had the timing on this one. The first email he sent Wednesday morning was to a client insisting they had already paid an invoice. They had not. They had paid last month's invoice late, and this was September's. His question was the right one: would the agent have caught that from the email itself, and how does it know the person writing to him is the billing contact rather than someone who happened to use the word invoice? That is the sandbox test worth running before you trust it with a customer relationship.

The phrase that got the room's attention was "listens for customer replies." That is the line between a reminder tool and something that participates.

Then the credits. Starting September 16, Revenue Agent begins consuming them, at five hundred credits when it starts collection outreach on one invoice. Five hundred sounds like a lot next to everything else in HubSpot. It is five dollars. If you can show finance that it gets invoices paid faster, that might be the least complicated ROI conversation you will ever have about credits, and it only breaks down if you are selling five dollar widgets, which HubSpot customers generally are not.

Why it matters: Two things, and the second one is bigger. First, HubSpot should put the dollar figure next to the credit figure everywhere, the way it already does on the knowledge base agent card. Two hundred credits, two dollars. Nobody should have to explain credits to a finance team. Second, cheaper than a collections person is not a reason to remove the collections person. That person now runs collections at a speed where nothing ages past thirty days instead of ninety. The move is repositioning the talent, not replacing it.

Will It Breeze: What Does Brand Identity Actually Touch?

The wheel landed on a prompt about brand identity, and the first spin produced the kind of question a real user asks and an AI struggles with: using brand identity, can you tell me if this is normal across this portal? Normal compared to what?

Breeze handled it better than expected. It found one brand kit, confirmed it was the portal default, and then volunteered the nuance that mattered: not every HubSpot surface visibly uses brand identity. Some areas are brand aware and some, including the product updates page they were standing on, are not. So a mismatch there is normal too.

Then it offered to list which assets in the portal actually use brand identity and which will not, and that answer came back with HubSpot Academy video clips embedded in it. Academy was only connected as a Breeze source a few months ago, and this was the first time the show had seen it surface at that depth, with sources visible.

Why it matters: There are a lot of black boxes in AI setup right now. You fill in settings, something downstream is supposed to use them, and when the blog output does not match the brand you configured, people leave HubSpot to get the job done and stop trusting the system. A Breeze answer that tells you which surfaces honor a setting and which do not is worth more than a cleaner output would have been. The screenshot-and-send-to-your-admin moment is a feature, not a failure.

The Rest of Wednesday's Updates

Plenty shipped that the show did not pull forward, including the AI Filing Agent for Segments, new Campaign API endpoints for cloning campaigns and listing asset types, a centered display option for the chat widget, two way SMS in Help Desk and Inbox, user import supporting updates to existing users, a Help Desk board redesign with new card settings, and sandbox creation copying blog pages from production. Everything else that shipped is in 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.