From the September 23rd, 2026, edition of How Teams Work
First UNBOUND for me. Seventh INBOUND.
Somebody made that joke on day one, and it stuck. A lot of people who'd been coming to the conference for years were doing the same math out loud: technically my first UNBOUND, but I've been coming since it was still INBOUND.
Yes, they changed the name. Fifteen years of INBOUND, gone, replaced with something new. I'm not going to spend six paragraphs on why. Honestly, it didn't change much about what actually happened in the room. What changed the room was something else.
HubSpot isn't the only one doing this. Salesforce is doing it too. Most of the big CRMs are quietly becoming something closer to headless infrastructure, a place that holds your data and feeds it to whatever tool needs it, instead of just the screen your team logs into every morning.
That's a real shift. And it raises an obvious question. If the CRM stops being the interface, why does it still matter?
It matters because somebody still has to hold the data. Somebody still has to be the place where a deal, a contact, a support ticket, and an email thread all point back to the same record.
HubSpot’s bet is that it can be that place: the layer everything else gets built on top of, even as the front door changes.
Good data still comes first
None of that works without clean data. That’s not a new idea; we've been saying it for years, but AI makes the cost of skipping it a lot higher, a lot faster.
The numbers back this up. Companies with excellent data quality are more than twice as likely to beat their revenue targets as companies with poor data quality, per HubSpot's own AI research this year. That’s not a small gap.
Good data in one place, actually being used, not scattered across four systems nobody trusts, is step one. Everything else here assumes you've got that.
Then comes context
Once the data's clean and centralized, the next question is whether your AI tools can actually see it.
This is where most people get stuck. They copy information into Claude or ChatGPT by hand every time because the tool has no standing memory of who the customer is, what the deal looks like, or what happened last week. Copy, paste, explain the background again, ask the question. Do it again tomorrow.
HubSpot CEO Yamini Rangan put it plainly: AI with bad context is worse than no AI at all. That's the whole argument in one line. A tool that's confidently wrong because it never had the right information isn't saving you time; it's creating work you don't see yet.
The fix is building context directly into the tools you're already using, HubSpot included, so nobody's re-explaining the business from scratch every time they open a chat window. And the payoff is real. Companies that give AI access to three or four layers of context see meaningfully stronger results than companies giving it none. HubSpot customers specifically saw some outcomes jump well over 100% once the context behind the AI was solid.
Where's this heading next? Multiple agents working together instead of one at a time. Orchestration instead of one-off use.
We had more than a few unplanned conversations at UNBOUND with people already piecing this together on their own using scrappy tools like Grok Bot.
But that's a topic for another day. Right now, the foundation matters more than the frontier.
Here's the part that surprised me the least and bothered me the most: Almost every company we talked to is using AI in some form. Depending on who you ask, the number's close to 90%.
But fewer than one in ten of those companies describe their use of AI as actually transformative; the kind that changes how the business runs, not just how fast one person gets through their inbox.
HubSpot's Co-founder and CTO Dharmesh Shah said something on stage that's stuck with me since: “Don't worry about the state of the art, worry about the state of the practical.” Not the newest model. Not the flashiest demo. What actually works for your team, right now.
It can be hard to know where to start. We point people toward reporting, because that's where we started ourselves.
Since the word gets thrown around a lot, let’s quickly define Agent. It is a system that does multi-step work using real context from your business, and can take action across your tools without someone driving every step by hand. Not a chatbot that answers one question and stops; something that keeps working autonomously.
We recently helped a client move off a manual process they'd built around Claude. Every week, someone pulled data out of HubSpot, dropped it into a chat window, ran it against a base prompt, read the output, and then manually adjusted the results for whatever it got wrong. The process worked. It just took time, and none of the corrections ever stuck. Next week’s work started from the same base prompt as the week before.
We moved that same workflow into an agent built inside HubSpot instead. Same reporting job, same weekly cadence. The difference is what happens to the feedback at the end. Instead of someone re-tweaking a prompt from scratch every Monday, the corrections get built back into the agent's foundation. It gets better because last week's mistakes are part of this week's starting point, not something somebody has to remember to fix again.
That's the whole idea behind starting agentic. Pick one recurring process, run it side by side with how you already do it, and improve the process over time instead of throwing away the learnings every week. Most people skip that last part. They build one decent prompt and never touch it again.
The data backs the instinct here too. Some of the highest-ROI AI use cases, things like pipeline prioritization and data analysis work, are also some of the least used. Most people default to whatever's easiest to set up, not whatever actually moves the needle.
Picking the right process matters just as much as your starting point.
One story from a keynote stuck with everyone in the room: A marketing leader was told AI would make her job easier. Instead, her workload tripled: writing prompts, checking outputs, and fixing what came back wrong.
Here's the part that surprised us. The people who rate themselves as AI-fluent report more anxiety about it than beginners do, not less. Knowing more doesn't make the feeling go away. If anything, it sharpens it, because you can see both what's possible and how much manual work still sits between here and there.
If that sounds familiar, you're not behind. You're doing it the hard way, because nobody's built the systems yet.
We think that gets solved soon, not eventually. HubSpot's already building toward it, giving the platform one real place to hold business context instead of scattering it across chat windows, and connecting that to the tools your team already touches. Once that's in place, the corrections you make every week stop resetting to zero. They start sticking.
What's the one thing on your team that still resets to zero every week?
Want to get the "How Teams Work" newsletter sent to your inbox? Subscribe Now.