I finally had time to build my full recap of UNBOUND last week. Three days in Boston, a lot of product, and more hallway conversations than I could count. This is the long version of what I shared on LinkedIn: what HubSpot announced, what I saw firsthand, and what I would do in a portal this week.
What was UNBOUND 2026 actually about?
UNBOUND 2026 was about Growth Context: HubSpot's position that AI only delivers outcomes when it understands your business, and that this understanding is now something you can see, govern, and correct inside the platform.
The rename set the tone. After more than a decade as INBOUND, the conference became UNBOUND, and the product narrative followed. HubSpot no longer describes itself as a CRM with AI features. It describes itself as an agentic customer platform, built in three layers: a context layer (the Smart CRM and Context Home), an action layer (the hubs, agents, and Breeze Assistant), and a coordination layer (Agent Hub, connected systems, and governance).
Duncan Lennox framed the shift most precisely in the keynote: the CRM is no longer simply a system of record. Underneath it sits Aviator, HubSpot's new agentic harness, which orchestrates the agents and keeps them working from the same context.
"It's a system of context, the intelligence layer for your go-to-market."
Why does context matter more than the model?
Because HubSpot's own data shows that the same model, given bad context, performs worse than no AI at all.
Yamini opened the week with the statistic that framed everything after it: roughly nine in ten companies are using AI tools, yet only 6% describe their outcomes as transformative.
Her follow-up data, drawn from AI outcomes across more than 300,000 HubSpot customers, explains the gap. With bad context, the same AI model drove MQLs down 28%, deals won down 27%, and meetings booked down 49%, compared to using no AI. With good context, the same models drove MQLs up 264% and deals won up 197%. She wrote about it in The Outcomes Era is here, and the Fall '26 Spotlight press release carries the same argument.
That finding is the most important thing to come out of UNBOUND for anyone who operates a portal. A smarter model does not compensate for a portal whose lifecycle stages mean different things to marketing and sales, whose ICP was written two years ago, or whose activity data depends on reps remembering to log calls. The agent simply reasons faster over the wrong inputs.
It is also exactly why I built the AI Readiness Audit: to find where a portal's context is incomplete, contradictory, or stale, and put it in the right place before the agents start reading it. Context is going to matter more than ever, and HubSpot just made that case with its own numbers.
What stood out on the product side?
Four releases stood out to me, and each one reads from the same context layer.
The rebuilt Breeze Assistant
I got a live demo of the new Breeze Assistant, and it caught me by surprise how much it can already do compared to the previous iteration. The old assistant answered questions. The new one takes direction: you describe the outcome, it selects the agents, and it returns the deliverable. The release adds Projects for persistent context across conversations, a Slack integration, voice on mobile, the ability to @mention any agent from the chat, and formatted documents generated from CRM data. The assistant itself consumes no credits; the agents it calls do.
Knowing what Kolin Koehl and the team are working on, this is headed somewhere unbelievable.
HubSpot Work
HubSpot Work is the release I am most excited about. It is a new AI-first workspace built directly on the CRM: you describe the work, and it builds the project tables, documents, forms, and automations, all connected to live contact, company, deal, and ticket records. It has its own agent inside, and that agent is genuinely next level. It is in beta, and HubSpot has published nine use cases ranging from client onboarding to admin request management. For RevOps teams running shadow request queues in Slack threads and spreadsheets, this deserves attention immediately.
Context Home
Context Home is one screen showing what HubSpot currently believes about your business, customers, team, and processes, and the place to correct it. It supersedes the old AI data sources settings, values you edit are locked against automated overwrites, and uploaded documents produce suggestions for review rather than silent changes. Every agent reads it first, which makes it the most consequential screen in the release. Read it as a stranger would. If your ICP description there would not help a new hire, it will not help an agent either.
Agent Hub and Agent Builder
Agent Hub is the single place to see and manage every agent running in a portal, HubSpot's and your own. Agent Builder now supports custom event triggers (Slack, a spreadsheet, or a schedule), one-click conversion of existing workflows into agentic ones, and a simulation mode that estimates credit cost before you publish. Details on tiers and credit consumption are in Penguin Strategies' breakdown.
Also worth knowing
The self-updating Smart CRM. Calls, emails, and meetings are captured automatically. Manual activity logging has quietly degraded more portals than any broken workflow.
Prospecting Agent. It now monitors 40+ buying signals, assembles buying groups, and drafts outreach per contact.
Revenue Hub. Quotes are generated from deal context, and the new Revenue Agent brings AI to accounts receivable (public beta).
ChatGPT Ads. HubSpot says it is the first CRM with a native integration, including landing pages, lead capture, workflow enrollment, and attribution. Microsoft Advertising became a native channel as well.
The Breeze name is being retired everywhere except Breeze Assistant. When the whole product is AI, a separate AI sub-brand stops making sense.
What did Dharmesh mean by the Age of the Builder?
That the next generation of HubSpot is built for people and agents working together, and that operators, not only engineers, will build the agents.
Dharmesh's Age of the Builder talk was the session I kept thinking about afterward. In a follow-up post, he laid out HubSpot's 20-year arc in three phases: democratize inbound (inbound marketing), democratize CRM (the customer platform), and democratize AI (the agentic customer platform).
He also defined what "agentic" actually requires, and it is a useful checklist for evaluating any platform claim:
Pre-built agents for key use cases. Four new ones were announced at UNBOUND.
Custom agents that run natively on the platform, which is Agent Builder.
A platform that AI agents themselves can operate. Jon Dick noted the same week that HubSpot is the top CRM plugin for ChatGPT and the top CRM connector for Claude, which is how my own audit work runs today.
An underlying harness and agent operating system, purpose-built, on which all of the above runs.
The distinction that stuck with me: the last generation of the platform unified data so people across departments could work on a common system. This one unifies data and context so people and agents can.
What this supersedes.
Treat the CRM as a system of record. Turn on AI features individually and judge them by what they produce in isolation. Assume a better model will compensate for incomplete data.
Treat the CRM as a system of context that every agent reads before acting. Govern that context in one place with Context Home and manage agents in one place with Agent Hub. Accept that bad context makes AI perform worse than no AI, and sequence the work accordingly.
What should you do in your portal this week?
-
Read your context before you activate anything, then turn on one agent for one workflow and measure it.
-
Open Context Home and read it critically. Check the ICP, personas, products, and processes against how your business actually sells today. Correct what is wrong; edited values are protected from automated overwrites.
-
Inventory Agent Hub. List every active agent, what it does, and who owns it. Agents without owners are the new orphaned workflows.
-
Fix activity capture. If reps still log calls and meetings by hand, the self-updating Smart CRM is likely the highest-value switch in the release.
-
Govern credits before you spend them. Run metered agents in simulation mode first and set portal and feature-level caps.
-
Start narrow. One agent, one workflow, two weeks of measurement, then expand.
Who made the week?
The people, as always.
A sincere thank you to Sarika Garg for the shout out during her session highlighting my quote template. Sorry I missed it, I got pulled into meetings.
It was genuinely great to see so many HubSpotters in person, and to talk through the quote template with Ethan Kopit, Jack Coppersmith, and Arun Gunari from the Revenue Hub team. And a special shout out to Caleb King, who I ran into about six different times, every conversation more interesting than the last.
The custom RevOps Sins card decks were a huge hit. Four suits (Data Sins, Process Sins, Attribution Sins, and AI Sins), and watching people crack up as they opened their packs and read the cards was genuinely entertaining.
What comes next?
GROW in London on November 17. HubSpot's GROW Europe takes place at the Royal Festival Hall. If you are heading there, let me know.
