I’m writing this from Boston, with a lanyard around my neck, a lukewarm cup of coffee next to me, and a note-taking app full of exclamation points. HubSpot didn’t just add a few AI features this year. They took the entire platform apart and rebuilt it from the ground up, with AI as the starting point. That’s a big story, which is why this is a long read. Make yourself a cup of coffee.

I’ve paired each announcement with two questions. One: how does this benefit you as a HubSpot user? Two: what is the rest of the market saying about the problem HubSpot is trying to solve here? That second question is important, because none of the releases below were conceived in a vacuum. They’re all responses to metrics you’ll likely recognize from your own organization.

The order reflects how these features build on one another: first the foundation, then how you communicate about it, then the tools for marketing and sales, then how you keep everything under control, and finally how HubSpot integrates with the rest of your AI stack. At the bottom is what you should do first tomorrow.

Poor context is worse than no AI at all

HubSpot refers to the coming years as the “outcomes era.” In other words: the time of “look what AI can do” is over, and the time of “show me what it delivers” has begun. More than a trillion dollars has been invested in AI models. The value only comes when something translates those models into results. And for that, it’s not the model that matters, but the context in which the model operates.

This isn’t just HubSpot talk. The market is struggling with exactly this issue. McKinsey surveyed 1,719 organizations this summer: 88% use AI in at least one business function. Only 37% see any impact on profits, and that figure hasn’t changed in a year. Eighty percent of individual users say they’re more productive; the organization as a whole rarely sees that reflected in the numbers. McKinsey calls this—to put it politely—the central tension of the year. I call it “pilot purgatory”: everyone is doing something, but almost no one is reaping the rewards.

Gartner is even less subtle. According to them, more than 40% of all agentic AI projects will be shut down by the end of 2027. Not because the models don’t work, but for three mundane reasons: costs are rising, no one can explain what the benefits are, and the risks haven’t been managed.

HubSpot has compared this with its own figures, covering its entire customer base. Teams that use AI with good context perform better across the board: 264% more MQLs, 224% more closed deals, and 200% more scheduled customer meetings. Impressive. But the slide that silenced the room was the other one: poor context is worse than no AI at all. With messy context, the number of closed tickets drops by 70% and the number of connected customer conversations by 86%. Not compared to good context, but compared to no AI at all.

Why? Because people automatically compensate for hassle. You know that the
product catalog is from last year and that Sandra no longer handles approvals. AI doesn’t know that and confidently makes the mistake. These are HubSpot’s own data, and they’re correlations, not lab tests. But they point in the same direction as McKinsey and Gartner, and it’s the key to everything below: the difference between a pilot and a result lies in context, data, and governance. Not in which model you choose.

1. The new Smart CRM: no longer a filing cabinet, but a colleague who pays attention

The biggest change isn’t visible, and that’s exactly the point. HubSpot calls the new CRM a “system of context” rather than a “system of record.” Your CRM used to be a filing cabinet where you stuffed things in. Now it’s a coworker who understands what’s inside.

The architecture consists of four layers:

• Data: structured and unstructured, together. The foundation, as always.

• Trust: who’s allowed to do what (security), what rules apply (governance), what each agent did and why (observability), and is it accurate (quality). Now that agents are doing things on people’s behalf, this is no longer a side issue.

• Intelligence: the new brain, in two parts. Aviator, the agent harness that selects the best AI model for each task and provides the appropriate context. And the Growth Context

• Graph: which extracts insights from all your data in three categories: business (brand, tone of voice, product, positioning), team (roles, goals, approvals, workflow), and customer (ICP, communication, contracts, revenue). I wrote a separate article about Aviator; suffice it to say here that it doesn’t rely on a single AI model and is continuously tested on go-to-market tasks.

• • Action: the hubs you’re familiar with, HubSpot’s own agents, and the agents you build yourself. All powered by the same engine.

You’ll see why this is smart when you look at where projects usually fall apart. Analyses of the Gartner forecast show that data preparation consumes 60 to 75% of all work on an agent project, and that projects get bogged down by integration within a few weeks. HubSpot takes those two things off your hands: the data is already in the CRM, and the trust layer—which you’d otherwise have to build yourself—comes standard.

What you’ll notice: the CRM updates itself automatically. The demo example was wonderfully
everyday. An out-of-office email. The CRM reads: contact is out of the office until January 6; in the meantime, the CFO is the point of contact—here’s her title and phone number. Everything is automatically updated. No manual entry, no “oh right, I forgot that,” no outdated contact in your sequence. If you’ve ever led a CRM adoption project, you know how much hassle this saves.

2. Context Home: the dashboard showing what HubSpot knows about you

If context is so important, you’ll want to be able to see what’s there. That’s what Context Home is built for: one place with everything HubSpot knows about your company, neatly organized into Summary, Business, Customers, Team & Process, and Custom.

You’ll receive a context score that indicates how complete and up-to-date your context is, plus
recommendations on where you still need to add information. Adding information is intentionally simple: upload a file or specify a source, and HubSpot does the rest. There’s a change history, and “knowledge vaults” for sources you want to store in an organized manner.

Everything here is immediately available to every agent and every AI feature in HubSpot, including the agents you create yourself. HubSpot reported that customers with configured context see six times more MQLs and a seven-fold increase in ticket resolution compared to similar companies without it.

I’d go so far as to say this is the most important release of the week, even though it looks like a settings screen. Over the past year, the entire AI world has shifted from prompt engineering (asking the perfect question) to context engineering (ensuring the model has the right information at the right time). The major AI labs say it themselves: a simple, well-fed agent beats a complicated framework. Context Home is where that concept gets a button for marketing and sales.

And yes, it’s work. HubSpot asked 6,000 customers and partners what stands in the way of AI adoption the most. Number one: trust in AI. Number two: reliability of the results. Third: data quality. And only 20% said their data was ready to go. That means four out of five aren’t. If you’re thinking, “That’s us,” you’re in good company.

3. Breeze Assistant: Talk Instead of Clicking

Breeze has been around for a while, but what’s coming now is a complete overhaul. The idea: you no longer have to learn menus, figure out workflows, or know which agent can do what. You simply say what you want to accomplish. Breeze selects the right agents, does the work across the entire platform, and comes back with the result.

HubSpot calls it a go-to-market expert that does the work for you, not a chatbot that answers questions. The difference lies in the context from point 1. Because Breeze knows who you are, how your team works, and what’s going on with the customer, you don’t have to upload anything or write any prompts. The demo was a follow-up email after a trial. Breeze knew which product had been tested, how it compares to the competition—including the Gartner score—how the team typically follows up, what concerns the customer had expressed about ease of use and price, which colleague needed to be involved, and which pricing tier suits a company with 23,000 employees. Every sentence in that email had a reason to be there. Compare that to the average AI email you receive, where you know after just two lines that no one’s actually paying attention.

It’s a deliberate choice to go against the grain. Most agent platforms give you a powerful engine and say, “Good luck with that.” That’s exactly where it goes wrong. In 2025, MIT found that about 60% of organizations are looking into custom AI tools, 20% are testing them, and one in twenty actually puts them into production. Customers tell HubSpot they don’t have time to become AI experts. Breeze is built so they don’t have to.

4. Marketing Studio: From “Hmm, interesting” to a live campaign in a single conversation

For marketers, the diagnosis shared on stage was painfully familiar: organic reach is declining, buyers are asking AI for information instead of filling out a form, you have to create more and more content, and it’s yielding fewer and fewer results.

The numbers behind this are grim. In the first four months of 2026, 68% of U.S. Google searches ended without a single click to a website (SparkToro and Similarweb), up from 60% in 2024. On desktop—where B2B research takes place—that figure is approaching 80%. Bain & Company reports that click-through rates in B2B software categories have fallen by up to 30% since the introduction of AI Overviews, organic traffic is down 15 to 25%, and the most striking statistic: 85% of B2B buyers purchase from a vendor that was already on their shortlist before they even started searching. G2 found in March that 51% of B2B decision-makers begin their product research in an AI chatbot. And the traffic that does come in is different: Semrush reports that visitors arriving via AI convert more than four times better than organic traffic. Fewer clicks, higher intent—and it all depends on whether you’re mentioned in that AI response.

HubSpot’s answer is Marketing Studio: a single environment where you can use Breeze to go from insights to a running campaign. Three new agents do the work:

• Campaign Agent: Ask for a campaign plan, and you’ll get the message, the
  angle, the assets, and the segmentation.

• Content Agent: takes that plan and writes the assets in your brand voice, tailored to your ICP. That brand voice comes from Context Home, not from a prompt someone once put in a Google Doc.

• Nurture Agent: writes follow-up emails for each contact based on CRM data and
  engagement history.

What impressed me most in the demo was the AI Insights panel. Marketing Studio itself detected that a conversion path was broken (an “Apply” button was returning a 404 error, while 4,800 people were set to receive the next email in four hours—you can just imagine the campaign manager’s panic) and also provided an AEO recommendation: “You’re losing visibility in AI responses; competitors are being cited more often.” Answer Engine Optimization right in your daily dashboard. Given the numbers above, that’s not just a nice feature but a pure necessity, and I don’t know of any other marketing platform that has it built in at this level.

5. Sales: Less Typing, More Selling

For sales, the diagnosis was that reps spend more time on administrative tasks than on selling. Follow-ups are missed, and what was said in conversation one is forgotten by conversation three. This isn’t unique to HubSpot users: Salesforce has been measuring in its State of Sales for years that salespeople spend roughly 30% of their time actually selling. The rest is spent updating systems and coordinating internally. Thirty percent. Imagine if your baker spent 30% of their time baking.

Three roles address this:

• Prospecting agent: Tell Breeze you want to meet with a major prospect. The agent maps out the buying committee and writes a personalized email for each person, using the same context as above.

• Notetaker: After a meeting, dictate your updates—done.

• Deal progression: Automatically enter everything into the CRM so that deals move forward without anyone having to “quickly update the pipeline” on a Friday afternoon.

Together with the self-updating CRM, the promise is that prospecting, researching, coordinating, and updating the CRM require less manual effort in HubSpot than anywhere else. Test that claim in your own portal. But the direction aligns with what Stanford HAI summarized in April based on published studies: where AI is truly integrated into the workflow, productivity gains range from 14% in customer service to up to 50% in marketing output. The gains are there. The question is whether your organization will capitalize on them or let them slip by.

6. Agent Hub: A Single Overview to Combat Proliferation

HubSpot honestly acknowledged a problem I encounter with nearly every client: agent sprawl. The more agents there are, the less anyone knows what’s running, what it does, and whether it works at all. The result: higher costs, a cluttered customer experience, and agents working at cross-purposes.

The market is right at this point. Gartner places agentic AI on the “Peak of Inflated Expectations” in its 2026 Hype Cycle: 17% of organizations have actually rolled out agents, and over 60% plan to do so within two years. McKinsey forecasts that the share of large companies scaling up agents will rise from 27% to 40% in a single year. That’s a huge number of new agents in a very short time, built by people who weren’t capable of doing so just last year. Gartner has even coined a term for it: “agent washing.” Of the thousands of companies that claim to be building agents, Gartner estimates that only about 130 actually did anything worthy of the name. The rest are just chatbots with a new name. Without a clear overview, you as a customer don’t even know what you have.

Agent Hub is a single place to view, manage, and build all agents—both HubSpot’s and your own. Because they all share the same context, they truly work together, rather than being a collection of separate agents that get in each other’s way. If you’re responsible for AI governance in your organization, this is the first screen you’ll want to open. It doesn’t solve the problem of sprawl company-wide, but it does within marketing, sales, and service.

7. Agent Builder: Build agents yourself just by asking

HubSpot sees a new role emerging in go-to-market teams: the builder. Not the developer, but the marketer or sales ops professional who wants to quickly build something themselves to extend HubSpot’s capabilities. From simple (qualifying every lead before a rep sees it) to smart (monitoring the entire pipeline and flagging high-risk deals before anyone asks).

This fits into a broader shift. McKinsey predicts that by 2026, 32% of organizations will prefer to build software themselves rather than buy it, because AI has made building it easy. For decades, companies had to adapt to the software. Now, anyone can use plain language to create an agent that adapts to the company. That’s the world turned upside down—in a good way.

Agent Builder makes that possible, and you build by chatting with Breeze. The
key detail: your agents run on Aviator and use the same Growth Context as HubSpot’s own agents. So no second-rate agents—the same trust layer, the same context. And that’s the difference compared to that one colleague’s agent. You know the story: someone builds something brilliant on a Friday afternoon, shares it on Slack, everyone sends fire emojis, and a month later that person leaves for an AI startup. The agent leaves, no one knows the password, and the efficiency is gone. HubSpot told that story about its own team. I’ve seen it in various forms with clients.

8. Headless and open: agents on HubSpot, and agents that power HubSpot

The last section covered how HubSpot opens up to the rest of your AI stack. There are two sides to this: agents can run on HubSpot (built using HubSpot’s data and intelligence as building blocks) and agents can run HubSpot (controlling the platform from the outside via APIs, without ever logging in).

To that end, the APIs are divided into three layers:

1. Data APIs: raw access, over a thousand public endpoints, with the promise that anything you can do in the interface can also be done via the API.

2. Context APIs: not raw data, but ready-to-use insights. “Ask HubSpot a question and get an answer instead of a spreadsheet.” The same layer that HubSpot’s own agents use.

3. Work APIs: call on an agent—whether from HubSpot or your own team—and get a complete result in return.

The example makes it tangible. You want to know which deals are at risk. Data API: you retrieve all deal data, drag it into an external tool, and figure it out yourself. A lot of work, a lot of tokens. Context API: “Give me my ten highest-risk deals”—done. Work API: “Find out which deals are at risk and give me an action plan for each deal,” and that’s exactly what you get. McKinsey notes that for 20% of organizations, operational AI costs are already a barrier to further adoption. Less raw data processed by the model means fewer tokens. That’s not a technical detail—it’s your bill.

And then there’s the part I, as an AI person, liked the most. Over a year ago, HubSpot was the first CRM with MCP (the open protocol that lets AI assistants communicate with tools, now embraced by all major model providers), has had an Agent CLI since May for bulk and scheduled tasks without human intervention, offers connectors for all four major frontier models, and is now the most widely used CRM connector in both Claude and ChatGPT, with 250% growth in weekly users.

Why this matters: the AI landscape in which your buyers and your team operate
is rapidly fragmenting. In 2025, 89% of measurable B2B AI referral traffic still came from ChatGPT. By the spring of 2026, that figure had dropped to 63%, with Claude at 18.5%, Gemini at 10.6%, and Perplexity at 7.3%. A CRM that only communicates with one of those ecosystems becomes an island. HubSpot has chosen to communicate with all of them. Good choice.

What to Do First Tomorrow

All the figures in this article point in the same direction. Using AI is no longer a differentiator—88% are already doing it. The difference lies in the small group that moves from pilot to results. And that group isn’t distinguished by which model they use, but by how they manage context, use cases, and governance. The order determines everything, and HubSpot’s own data shows that the wrong order actively worsens your results.

1. Context first. Open Context Home, check your score, and fill in the basics: brand, tone of voice,products and positioning, ICP and personas, team processes, and who’s authorized to approve what . This is work, not a setting. It’s also the work with the greatest leverage.

2. One use case, with a number. Choose one problem with a measurable outcome: qualifying leads before a rep sees them, flagging high-risk deals, or preventing a broken conversion path. Gartner and McKinsey both point to “unclear business value” as the primary cause of failure. Give the agent a number and an owner.

3. Governance from day one. Make Agent Hub your central dashboard and agree on who owns which agent. McKinsey finds that AI leaders are nearly three times as likely to have a fixed “human-in-the-loop” process as the rest (65% versus 23%). Preventing uncontrolled growth is much easier than cleaning it up.

4. Integrate with what you’re already using. Is your team already working with Claude or ChatGPT? Connect HubSpot via the connector and see where the context and work APIs can replace manual tasks.

5. Fact in the costs. Much of this runs on credits. Proof of value before commitment is the new normal—even for HubSpot itself. Start small, measure what an agent costs per result, and scale what works.

The UNBOUND Debrief: 45 minutes, your portal, one checklist

At Six & Flow, we start every HubSpot AI implementation with context and data foundation, and only then move on to agents. Since we’ve seen all the sessions this week and will be setting up the first portals on the new releases in the coming weeks, we’re offering a UNBOUND Debrief for teams who want to know where they stand.

In 45 minutes, we’ll do three things:

• We’ll walk through your portal to assess its readiness: what’s there, what’s missing, and what’s outdated.

• Together, we’ll choose the first use case that will deliver the fastest, demonstrable results for you—complete with a specific number.

• You’ll receive a list of up to five action items for the next ninety days, including which releases from this update are—and aren’t—relevant to you.

No slides, no pitch—just a concrete starting point. Schedule your debrief here. There are limited slots available over the next three weeks, since we’re handling them ourselves.