Vibes’ SVP of Product Management Joe Catrambone has been defining, shaping and evolving the company’s Vibes Platform, Connect and RCS Studio offerings for over five years from his home base in Chicago. He’s a veteran leader who’s been working with AI dating back to the 2010s days when we all called it that – but it wasn’t anything close in power or daily utility to the AI that he and his organization are using to rapidly build Vibes products with today.
In this second of our AI-focused interviews (read the first with Vibes VP of Engineering Steven Mastandrea here), Joe provides a bit of a curtain-lift for Vibes' move from individual to institutional AI, and what sort of innovation this is bringing to Vibes customers in the weeks and months to come.
Jay Hinman
Joe, I’m first interested in understanding how AI and LLMs are shaping your team’s product development processes, and how they allow you to learn from the market about what to build next.
Joe Catrambone
As the head of product here for over five years now, the most profound and valuable application of AI in that time is what the industry calls the distinction between individual versus institutional AI. AI branding like a “.ai” at the end of your web domain really doesn’t mean anything. It's a marketing ploy.
Individual versus institutional for me, especially when you talk about developer habits, is really the coordination over chaos. You know, as someone who's responsible for directing our development investment every year, it's great that we have something that employees can use. Individual AI means that every employee has their own habits, their own prompting styles, their outputs, you know, but they don't necessarily connect to anyone else's. It's good to get familiar with it but imagine thousands of agents rowing in different directions. That's the chaos that individual AI can bring.
What sets us apart is that we are very much invested in the movement to institutional AI, which is that coordination layer. We’re likely to turn more into Agentic managers than content creators, if that makes sense. We're going to make sure that the AI has the context that it needs, the structure that it needs, and the guardrails that it needs.
That's the institutional AI movement. I think that's foundational; number one, you have to do that. You get what everyone has been talking about, the increase in efficiency. You have an AI factory pumping out code, and that frees your developers, your product managers, and your designers up to make a move from the discovery side of the house.
What I mean is that back in the day, you'd go and you'd do heuristic studies, software usage studies with user groups and how they interacted with your software. You had data collected from the software behaviors themselves, and you analyzed all that, then you decided, okay, based on what we're seeing here and with the industry trends, let's take a look at what we should be doing. All of that took a long time - doing the observation and doing market research. That could take weeks.
Jay Hinman
That’s the model I’ve operated under my whole career, pretty much.
Joe Catrambone
Now the market research takes hours, if that. Most of the time is reading and validating the output of what the AI has put together. So no longer are you going from site to site and to webinar to webinar, taking notes, then doing all the analysis of that, and then making your best guess. You can have AI, you know, comb all that stuff and provide you with an output, and that frees you up to actually experiment. And the whole goal of experimentation is to fail.
Now, you learn that success doesn't really teach you anything. Success just validates that you had the right thing. Failure teaches you something, when you get to go to x number of customers, and you can experiment with a specific concept. That's where Vibes’ evolution is really coming into play. You get this candid feedback on something that isn't just a high-fidelity prototype, it is really true working software.
Jay Hinman
That they can even play with at the end. You basically can do usability testing in a much more iterative manner than you've been able to do.
Joe Catrambone
That's right. I know people think that AI is dehumanizing things. The way I see it as a product leader is that it's actually allowing us to be more human in our approach to building software, because we can get our heads out of the mundane. We can farm that out to an agentic agent, or we can farm that out to our agentic coding machine.
Jay Hinman
So you see your team being able to be more strategic with the time that they're saving.
Joe Catrambone
Yes. You don't get there, in my opinion, unless you move from individual AI to institutional AI. You've got to have that common goal, that common framework. Everyone needs to purposefully execute with AI for this specific use case, instead of, you know, Joe has his tool, Jay has his tool, we're building this piece of software, but we don't have a common way of doing it. I think that's a recipe for disaster, and a lot of money spent. I don't also know how you measure ROI if you have 25 steps in your process, and I have 106.
Jay Hinman
To make it more institutional, what sort of governance process has Vibes put in place so that it isn’t this individual collection of people?
Joe Catrambone
Right now, that's what we're doing with our product development program. We're trying to figure out number one, what are the guardrails that we need to give to the AI? What are the things we're going to let it do, and what are we just not going to let it do? It needs to be auditable, so we're looking at ways to do that. We want to make sure that we have really good checkpoints within the agentic process.
What we’ve found is it's not really great at just going from concept to real thing in one fell swoop. You do need to chop that up. The frontier models are trained into what could nicely be called sycophancy. Like every single time you give it a prompt, it's like, “oh, that's a great idea!”.
There's no real pushback, so you do need to inject those things into the process as well. The frontier LLMs have been programmed to please. Part of the creative process is that back-and-forth tension that you get between, you know, commercial, product, and engineering, where they challenge each other's assumptions about a specific market problem and how to solve it. You don't get that with agentic coding unless you specifically put it in there, and still, I don't know that we trust that.
AI also optimizes for broad usage, so as we create this engine, we'll want to make sure we optimize it for our competitive edges. You know, it needs to be domain specific. There's no reason in the next 5 years for us to be relying heavily on the frontier models. I think sovereign models are the way to go. We're not there yet, but building the Vibes LLM specifically for our software is probably right.
Jay Hinman
I'm really interested in this sovereign model idea, and how Vibes in a pre-agentic world has been set apart from our competitors by our human touch and human element, and the fact that customers that become Vibes customers usually stay Vibes customers. How does that sort of play into your thinking and your team's thinking in how you develop AI, and a sovereign model for Vibes?
Joe Catrambone
We aim to understand exactly which elements of the human touch are the most valuable, because I don't think that that's going to be able to be mimicked by AI. However, I think we're going to be able to do more of that, because we can start to figure out how we automate some of the processes that removed our ability to spend more time with our customers.
For example, if I don't have to spend 3 weeks on market research, and I don't have to sit down with a designer and go through, you know, 2 weeks or 1 sprint of iterations, and we can literally just chunk this stuff out and go to a Customer Advisory Board, that’s great. Imagine the amount of trust that we're building with those folks as they feel like they're part of our development process, on a cadence that they have never been able to operate on before. As we are doing this organically, we are maintaining great relationships, and rapport with those customers. The idea has always been that product teams need to be in front of their consumers. That always had to be balanced with time in the office and developing and leading those dev teams. AI's going to make it so that as a product manager, you’ll spend 100% more time in front of customers than you used to.
