Show recap — Wednesday, May 20, 2026

Another Orange Morning Recap: May 20, 2026

Chris was out today, so the show ran as a duo with a first-time guest. Zach Hussion joined Casey for the hour, opened with a story about needing a spray bottle from his dad to wake up in high school, and was sold on coming back weekly by the end. Two updates from yesterday and a Will It Workflow that put the new Agentic Automation Builder through its first live test on the show carried the morning.

Here is what we got into: the personalization token update for rules-based chatflow welcome messages now in public beta, the Agentic Automation Builder going live for portals already running custom agents, a quick detour into Breeze data analysis versus Breeze reporting, and a Will It Workflow that landed on an invoice follow-up scenario with a data agent custom prompt step.

Two Updates, Both Familiar

Casey flagged at the top that the show has covered both updates before. The point of the segment was not the news but the second look. Both have shifted enough since the first coverage to be worth a fresh read.

Personalization Token Support in Rules-Based Chatflow Welcome Messages

You can now insert personalization tokens like first name, company name, or recent activity directly into the welcome message of a rules-based chatbot. When a visitor is recognized via tracking cookies, the token renders with their information. When they are not, configured default values keep the greeting natural. This is now in public beta across all hubs and tiers, which is the change worth noting since the show first covered this when it was private beta.

The conversation landed on the line between cool and creepy. Casey made the case that on a marketing page, a name-by-name welcome from a stranger's first visit lands as a "look over your shoulder" moment. Zach agreed it depends on the brand. Some companies can pull off a more personal opener earlier in the relationship. Most cannot. Where they both got back to "this is great" was the support context. If someone is logged in, with a real account and a real question, being greeted by name is the bare minimum, and being greeted with "Hi there" feels like the company has forgotten who they are.

Why it matters: Casey's add was that the feature falls apart on dirty data. If the first name field is full of all-caps entries, email addresses as names, or "Aaron" when the person is named Casey, a personalized welcome makes the data problem visible to the customer in the rudest possible way. The use case is post-consent surfaces with clean data. That is a narrower deployment than the feature description suggests, and it is the deployment that actually works.

For more details: Personalization token support in rules-based chatflow welcome messages public beta.

Agentic Automation Builder

The Agentic Automation Builder is the new app that brings workflows, AI agents, and data from across the business onto one modern canvas. It supports HubSpot CRM triggers, scheduled times, webhooks, third-party integrations, and other events. From there, you add traditional workflow actions, run AI agents as steps, branch on conditions, and reuse common processes. It is now in private beta across the Professional and Enterprise tiers of Marketing Hub, Sales Hub, Service Hub, Data Hub, and Smart CRM, and access is automatically enabled for portals already running custom agents.

Zach had not seen the builder before the show and opened it live, which made this the most useful kind of first look. The free-form canvas with drag-and-drop node placement got a real reaction. The new trigger options around buyer intent signals also caught the room. Casey pulled the conversation onto credits inside the new builder almost immediately. Triggering off a signal like "mergers and acquisitions" means buyer intent credits are being consumed every time the workflow runs. That cost is not surfaced in an obvious way today, and both hosts agreed HubSpot should make credit consumption more visible inside the builder itself, ideally at the property level on the record view too.

Why it matters: Casey's framing on this update was the one to take away. Treat the Agentic Automation Builder like a beta, because it is one. Breeze is the comparison she reached for. Breeze was rough at launch, has since matured, but a handful of bad early experiences still keep her from defaulting to it for things it now does well. Her advice for the builder: try it, and if it does not change your world today, do not write it off forever. Come back in a few months. The credits question is the second thing worth tracking. Powerful triggers cost real money to run at scale, and the bill is not in the builder yet.

For more details: Agentic Automation Builder private beta.

A Quick Aside on Breeze Reporting vs. Data Analysis

While Casey was talking about Breeze, Zach took a quick detour to ask whether anyone is feeling good about Breeze reporting specifically. The short answer in the room was no, not yet. Casey said she is still old school for reporting and reaches for the manual build by default. Where Breeze is shining is data analysis, where the model is asked to look at a set and tell you something useful rather than build a structured report. Two different muscles, and one of them is much further along than the other right now.

Will It Workflow: Inbox, Invoice, Send to External System

The wheel needed three spins. The first was calls plus workflows plus a use case, which Casey called a gimme. The second landed on cart, chatflows, and rescore. That one got stuck almost immediately because the cart object in HubSpot is tied to e-commerce integrations like Shopify, which Casey's demo portal does not have set up. The third spin landed on inbox, invoice, and send to external system, which is where the segment got real.

The use case Casey and Zach landed on: a vendor sends invoices to customers, and one of those customers emails the team asking about the status of their invoice. The workflow watches a one-to-one email trigger for body text containing "follow up" and "invoice," then runs a data agent custom prompt step to look up the most recent invoice on the contact and return its payment status, then sends a Slack notification to whoever owns accounting with the result.

The verdict was "it will workflow," with caveats. The trigger is the weakest part of the build as it stands. A one-to-one email body match catches too much, and the better version of this is a help desk ticket creation trigger filtered to a billing-specific inbox, which would scope the credit consumption to the right messages instead of firing on every email that happens to mention an invoice. The data agent custom prompt step was the part that earned a genuine reaction from Zach, and Casey shared a separate client example of using data agent custom prompts to backfill ten thousand missing company domain names from an ERP import. Different use case, same pattern: cheap AI step inside an automation, doing work that would otherwise be manual.

Watch the episode for the full build, including the trigger logic and the data agent custom prompt configuration.

The Rest of May 19's Updates

Full write-ups of both updates are on the updates blog: HubSpot Updates: May 19, 2026.

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, and Zach, we will see you next week.