Yesterday Brian Halligan said he and Dharmesh Shah are refounding HubSpot as the “Customer Harness Company.”
That sounds right to me, and it’s super exciting. HubSpot has made a big platform call like this before. They named inbound marketing as a category and then owned it for a decade.
But what does “Customer Harness” mean in practice? Right now at least, we’ll all have different takes. We’re all at different stages of our agent journeys, in general and in GTM specifically. Some teams are piloting their first SDR agent. Some have rebuilt their whole revenue org around agents. Most are somewhere in between.
So this is what it means for us at SaaStr, with 3 humans and 20+ AI agents in production. Some of it will change in six months.
#1. We Are 100% Headless on Our CRM and Never Going Back
No one on our human team logs into our CRM anymore. Including me.
10K, our AI VP of Revenue, and our other agents come in through MCP and the API. They read from it, write to it, and keep it updated. The CRM still matters a lot. It’s our system of record. We just don’t go there.
For 20+ years, B2B software was priced, designed, and sold around the human who logs in. Seats, dashboards, login frequency as a health score, adoption as the metric your CSM chased every quarter. For us, that model is over.
We will never log into B2B software again. Our agents will, and we’ll talk to our agents.
If you sell B2B software, it’s worth asking how much of your product’s value sits in the UI versus in the data and actions behind it. For a growing share of customers, the UI part is headed to zero.
#2. We Moved Off Traditional Marketing Automation Entirely in the Last 60 Days
We still do a lot of marketing. We just don’t use marketing automation software to do it anymore.
Our list is 450,000+ people. We send millions of emails, plus everything around our events, content, and products. Over the last 60 days, all of that moved to being run almost entirely by agents.
What changed:
- No more templates. 10K built all our email and promo templates, and rebuilds them dynamically for each send.
- Real personalization at scale. Many of our emails are now 100% personalized to each recipient, written for that person based on what we know about them, across a 450k list. This goes far past merge fields.
- No journey builders. The drag-and-drop workflows, branching logic, and hand-maintained nurture tracks are gone.
Consent is incredibly important, and so is deliverability. At our volume you can’t be sloppy on either one. But you can get both headless from a number of top-tier vendors now, as services you call via API. They no longer lock you into a marketing automation suite.
I don’t think a lot of vendors have priced this in yet. The old lock-in was that your templates, workflows, list logic, and sending reputation all lived in one place. Once an agent owns the templates and logic, and sending is an API, most of that lock-in goes away.
We will never buy marketing automation software again. It was built for humans to operate, and we don’t have humans operating it.
#3. Claude, ChatGPT, and Slack Are Great, but Our Main Surface Is 10K
Everyone has a strong view on the “killer interface” for agents right now. I don’t know which one wins, and I’d be skeptical of anyone who says they do this early.
For us, running GTM inside our agents, using their own custom interfaces, is still easier. A big reason is that it’s multiplayer: our humans and agents all work in the same place on the same context. Claude, ChatGPT, and Slack are great surfaces for GTM work. For us, at least for now, they won’t be the primary one.
What works best is that we just talk to 10K like a human. A human that never stops working and makes plenty of mistakes. 10K has all our context, all our history, and our data from multiple sources, including QuickBooks, Bill, and the rest of our stack, beyond just the CRM.
That’s our best surface. An agent that already knows the business and sits on top of every system we run.
If “Customer Harness” means one thing to us, it’s that whoever holds the most complete context on the customer and the business becomes the harness.
#4. We Use the Best Agent for the Job, Whether or Not It Comes From Our CRM Vendor
This may change. It’s all in flux. Today, though, we and 10K really don’t care whether an agent comes from our CRM or our other “core” providers. If our CRM vendor ships an agent and it isn’t the right tool, we don’t use it.
Our process:
- Pick the best-of-breed option for that specific job, at least for the moment.
- See if we can deploy it very quickly. If it takes a quarter to stand up, it’s probably the wrong one.
- Test it for real, with evals on our own data and workflows.
- If it passes, go.
That’s a very different motion from “standardize on a platform and use the modules that come with it.” The platform still matters as the system of record and the data layer. The agents on top get picked one job at a time, on performance.
For incumbents, this is the real risk. Owning the data used to mean you’d own every new workflow built on top of it. Now owning the data can just mean someone else’s agent calls your API.
#5. There Is No Agent Loyalty, but It’s Hard to Get Us to Look at Other Agents Once We’re Happy
This is where the urgency is, and why the timing is right for HubSpot.
All of us were loyal for years to our favorite B2B vendors. We were Salesforce shops, HubSpot shops, Shopify shops. In many ways we still are.
Our agents aren’t. They do pick favorites and they do pick leaders, but they aren’t remotely loyal.
We had a core vendor go down for 48 hours. We were loyal and would have waited it out. 10K routed around it.
We never went back.
No one on our team made that call in a meeting. There was no RFP and no renewal negotiation. The agent found a path that worked, the work kept getting done, and the old vendor dropped out of the loop. For the vendor, that kind of churn won’t show up as a cancellation at renewal time. Usage just falls to zero once an agent finds something better.
The other side of this works in a vendor’s favor. Once we’re happy with an agent, it’s hard to get us to look at alternatives. We don’t have the time, and every switch means new evals, new context, and new failure modes. So the winners in each job will get sticky quickly.
The window to be the one that gets picked is short. HubSpot is entering that race now, and every other B2B vendor is in it too, announced or not.
What “Customer Harness” Means for SaaStr in November 2026
Here’s our working definition today:
- The CRM is the system of record, accessed headlessly. Humans don’t log in. Agents do, via MCP and API.
- Marketing is agentic, and sending is a commodity. Consent and deliverability are services we call.
- Our primary surface is our own agent with full context. That’s 10K, with our CRM, finance systems, and history behind him.
- Agents get picked job by job, on evals. Being bundled doesn’t earn the deployment.
- Our agents decide where loyalty goes. They route around whatever breaks, and stick with whatever works.
HubSpot calling this out publicly and refounding around it is the right move. The vendor that went down on us for 48 hours learned how this works from its usage data.
It’s still early for agents in B2B. But it’s not that early. It’s Now or Never. HubSpot’s doing the right thing. Just in the nick of time. Let’s see how it plays out with the cofounders running the place again. It almost always goes faster. And that’s what everyone needs today.
