JustAI launch: Founder’s note

JustAI has changed a lot over the last three years. Some of what we learned was obvious, some ran against our instincts, and some was new to everyone. Before we talk about this launch, we want to share what we got wrong, because those mistakes explain every major decision in this release.

What we learned

1. People didn’t want 1:1 personalization. They wanted extreme customizability.

We started out believing that if an agent could truly understand a user, we’d be golden. That turned out not to be true. To do anything useful, we had to absorb a ton of product context. What features come with a Pro plan? How does a promo work? How many onboarding journeys can one user get? How many product teams are competing for the same inbox? Then we noticed something odd. Nearly every Slack conversation was a marketer sending us a detailed, custom stack of requirements. Flexibility beat autonomous agents every time. We had to change that.

2. Agents are only as good as the people using them, and the context they’re given.

Our product performed differently depending on who was using it. A few marketers were willing to be bad at using agents until they got good. They hacked the hell out of the AI stack to make it work for them. They had strong hypotheses and thought critically about when to use AI and when not to. They shouldn’t have had to work that hard, but this early on, only a few people could really make AI work in their favor. For everyone else, it quickly turned into slop: messages created and tested for no real reason.

Context was the other half of the problem. Half the time, our agents were dumb as a box of rocks. Why? They knew nothing about the company, and we hadn’t done enough to bring that knowledge into one place. So either the marketer had to fill the gap, or we were running on vibes. AI needed to be accessible, not something that demanded deep expertise and constant hand-holding. We had to change that.

3. Measuring revenue lift wasn’t enough.

Our hardest and most interesting pilots were the ones where we saw clear, statistically significant gains on our platform, and then went in circles trying to reconcile them with other platforms. The math was right, but everyone measured differently. Do you attribute a conversion to one message or to all of them? Within 24 hours or indefinitely? Do you account for event latency? The list goes on. We had to change that.

4. AI is about doing more with more, not more with less.

We assumed we were selling efficiency: plug AI into your current workflows and do the same work with less effort. That’s not what happened. We were giving marketers a new way of working that let them move much faster. But a new way of working comes with a real learning curve. You almost always spend extra time reviewing, verifying, and experimenting with AI. In exchange, you move faster than the average marketer. That extra work is real, and if marketers end up overworked, our job isn’t done. We had to change that.

What we changed

In April, we stepped back. We could keep speeding up, or we could get brutally honest about what was working and what wasn’t. We chose honesty, and we set out to fix each of these four problems.

1. Extreme flexibility:

We burned our old tech stack and rebuilt our infrastructure around AI, so marketers can customize it directly. You can now create your own agents and set your own guardrails. For example, you can set up a frequency optimizer with rules for exclusion hours and message caps (and more), and our infrastructure will scale your sends within those rules.

2. Shared context and one AI harness:

You can now store rich context about your company in the platform, prototype with it, and build memory over time, so agents stop working blind. We’ve also added onboarding and starter guides so AI is usable by everyone, not just people with the bandwidth to hack at it. We still have more to do here, and we’re committed to doing it.

3. Measurement that’s revenue-first and data-science native:

Instead of reconciling numbers after the fact, we ask for your measurement setup upfront so our results match how you already measure. We also keep a global holdout across your whole program and each individual journey, so you can reconcile at any point.

4. A QA agent constantly running in the background:

The whole product is backed by a QA agent that flags problems before you find them: volume dropping, a missing tag in an email, a badly configured user event. You stop spending hours verifying every detail.

—

Because why not? When AI can build everything, it’s our job to create a delightful and beautiful product that people want to use.

We still have work to do. As we do it, we’ll keep sharing what we see and what we build, because this isn’t about next year. It’s about the decades ahead. Salesforce wasn’t built in a day, and this launch won’t be our last.

- Neha
October 19, 2026