This was the first year the event went by UNBOUND instead of INBOUND.
That name had been in place for fifteen years. Dropping it wasn't a small rebrand; it was HubSpot signaling a move away from being a conference anchored to the inbound methodology and toward something bigger, a platform that is at the core of how you run your whole business.
AI has been a big part of HubSpot's focus since they first announced Breeze at INBOUND back in 2024, so a change like that raised the bar on what we expected to see.
Since this was the first conference under the new name, we expected a full-on blowout with a wave of major new product reveals.
Here's what we actually walked away with instead.
Given the new name, we were bracing for a bigger show. What we got instead was less about the hottest new features the product team is working on right now, and more about where HubSpot thinks this is all headed.
Growth context wasn't just a phrase that came up a few times. It was the theme of the entire event. Every session and every announcement got tied back to it in some way: HubSpot's goal of giving your systems the actual business information they need to drive impactful outcomes.
That's worth sitting with for a second. A company that's built its reputation on splashy product launches chose to lead with a concept this year, not a roadmap of new upcoming features. That tells you something about where they think the real problem actually is.
Even what did get announced leaned more toward overhauls than brand new tools. Things like Context Home, Agent Builder, and the Breeze assistant saw a lot of updates to agents that already existed, with a few new ones sprinkled in.
2. You don't have a prompt problem. You have a context problem.
If you've ever rewritten the same prompt three or four different ways trying to get a decent output out of an AI tool only to give up and do it yourself, you already know this feeling.
That's not a skill issue. That's a sign the tool never had the right context it needed to get it right.
I sat in on a session about this exact thing at UNBOUND. The speaker asked the room whether they were "doing copy pasta" with their AI tools, re-explaining the same background information over and over instead of giving the tool something standing to work from.
What HubSpot's building toward instead is having the platform itself act as that context layer, pulling from the data already sitting in your CRM records, from your calls, and from the new Context Home space, instead of having to spend time giving the AI tools a fresh set of instructions every time.
We've watched this play out in our own work for a while now. Speed and quality with AI tools rarely come down to how well someone writes a prompt. It comes down to whether they handed the tool the right documents, the right call notes, and the right history before they ever asked it anything.
3. It's all about your quality data
We've been talking about data quality being at the core of effectively adopting AI for a long time now. AI tools are only as good as the foundation they're built upon. Without clean and consistent data inputs, all you will get back is AI slop.
A lot of that starts with reps. Whether they're following a consistent process, are properly logging their calls, and whether the data they enter while working deals is consistent.
But it's not enough to simply enter data in the system if the data they're entering isn't quality. As Yamini stated: "AI with bad context is worse than no AI at all."
If you have a sales rep that is going rogue and entering a bunch of nonsense, simply because a field is required and they want to get to the next screen, that's not something you want your AI referencing. That raises the real question: who's actually responsible for catching that before it becomes part of your context.
I've heard more than one partner land on the same idea independently. Businesses are going to need someone whose job is basically context architect, someone who owns whether the data feeding these tools can be trusted, and pulls together the different pieces, such as calls, documents, and CRM records, into something the rest of the team can actually build on.
The barrier to actually building something yourself, whether it's an agent, an automation, or a small internal tool, has basically disappeared. HubSpot's own push into agent-building tools this year is part of that story, and it's not the only place we're seeing it.
Dharmesh talked about this too: the gap between being a customer of a piece of software and being able to build your own version of what you need has nearly closed.
That's genuinely good news. It's also a little dangerous if you're not careful. Easier building doesn't fix a bad foundation underneath. It just lets you scale whatever you already have, good or bad, a lot faster than before.
For most of HubSpot's history, the CRM has been the place your team works. Based on a lot of the conversations we had at UNBOUND, including one echoed by another partner that same night, that's shifting.
HubSpot is becoming more of an infrastructure layer, essentially acting as a headless CRM, the place that holds your structured and unstructured data, the context that gets leveraged everywhere else, not just inside HubSpot's own interface.
That's key for how we think about a client's tech stack and how things get integrated. HubSpot stops being just the system your team touches directly and starts being the infrastructure other tools get built on top of.
There was a story from one of the keynotes about a marketing leader who was told AI was supposed to make her job easier. Instead, her workload tripled: writing prompts, checking outputs, fixing what came back wrong. It's a story that's going to resonate with a lot of teams.
New tools and models are coming out every day right now, and it's easy to feel overwhelmed trying to keep up. But what we heard at UNBOUND is that it's not about having the newest, greatest tools. Too many teams are already struggling just to keep up with everything moving this fast.
Dharmesh said something on stage that stuck with me: "Don't worry about the state of the art; worry about the state of the practical." What tools do you actually need to get the job done? It's not about being ahead of everyone else. It's about getting it right for your organization, on your terms.
If you feel like everyone else is further along than you, they're probably not.
One of the more useful sessions I sat in on broke down the different types of companies and how they actually use AI.
It splits companies into three types. Pathfinders take on what's coming at them, look at it, and decide deliberately whether to act now or hold off. Chasers jump at whatever's newest, then jump at the next new thing, whiplashing without any strategy behind it. Bystanders look at the same thing and decide it doesn't apply to them, so nothing changes. They're usually locked into rigid planning cycles, unwilling to question their own assumptions, and want guaranteed ROI before they'll move, which means they rarely move at all.
Most companies we work with aren't chasers by choice. They usually want to be pathfinders. They just haven't put in the work yet to get there.
If you've put real money into AI tools this year and haven't seen a return that matches the investment, you're not alone, and it doesn't mean you made a bad call.
The entire industry is still in an experimentation phase. New business models, new pricing structures, new infrastructure layers are all still getting worked out in real time. That's exactly what an early phase looks like.
The growth shows up later, once this stuff stops being novel and starts being standard. Right now we're all still building the road while we drive on it.
Every one of these eight things points back to the same idea, even though none of them announce it directly.
The companies getting real value out of AI right now aren't the ones with the most tools or the biggest budgets. They're the ones who did the unglamorous work of getting their foundation right before they started layering AI on top of it.
That was true before UNBOUND. It'll still be true next year, whatever they end up calling the event by then.
