Mastermind · February 7, 2026
User interviews and API-shaped thinking
The reframe I came out of this week's call with is that I'd been shipping Pulse as an app and the bigger bet is Pulse as an API.
I walked Chinat through the new surface. The user interviews had been clarifying: founders are the sharpest early segment, and the jobs they want done are narrow and specific. One, automated cold outreach at real scale, where you specify who you want to reach and Pulse finds them and drafts the message. Two, presence across a few public surfaces, where Pulse can act on your behalf without constant supervision. Narrow, but demand is real. People were willing to pay for it.
What the conversation with Chinat surfaced is that the product shape I'd been defaulting to, an app with a dashboard, caps the ceiling. The app shape assumes a human in the loop. The API shape assumes that other agents might be the caller, which changes what the product is. If Pulse is an agent-callable capability, a different and larger class of product can be built on top of it, including my own. If Pulse is a dashboard, the ceiling is however many founders remember to open it.
The concrete version of this is a feature I'd been building underneath without naming: an authentication layer with API keys so developers can hit Pulse programmatically. A few companies working on consumer AI products in adjacent verticals had asked whether they could build on top of our context and memory layer. Those conversations had been running in parallel to the consumer app work. Seen separately, they looked like distractions. Seen together, they were the actual product.
Chinat framed it back to me with a question I liked: what does your product look like if half your usage comes from agents calling it, not humans? That forces a different design discipline. Rate limits, latency, reliability, and documentation become first-class instead of afterthoughts. The UX question becomes what does a good developer experience feel like, not what does a good dashboard feel like. The pricing conversation is completely different too. Consumer apps get priced on monthly seats. Infrastructure gets priced on usage, and the unit economics are more forgiving.
The reason this matters for the research I'm pushing on the network-of-agents side is that an API shape is the only shape that can credibly claim to be infrastructure for agents. A dashboard can't network with other dashboards in any meaningful way. An API can. So the research thesis and the product thesis start to plug into each other. The paper argues that networks of agents become intelligent through structured interaction mechanisms. The product has to be one of those mechanisms, not just a consumer surface.
The hard part is the cutting. I've been accumulating features on the app side because each one was individually plausible and individually demo-able. Every one of those features that doesn't serve the API shape gets deprioritized. Meeting notes that tailor to a user's context, fine, that becomes an API capability. A heartbeat that runs on its own clock, fine, that's infrastructure. Live transcription, not this quarter. Visual memory, not this quarter. Each of those might come back, but only if the core product is proven first.
The user interviews are what keep this honest. I'm running ten to fifteen personally this week, and my co-founder is running another ten to fifteen. The test is whether the people we talk to, unprompted, describe a shape that looks like what we're building. If they don't, we rework. If they do, we keep cutting.
The other conversation that came up is what kind of marketing Pulse needs right now. Chinat was direct about it: the first marketing video matters less than the first piece of documentation. If I'm pitching Pulse as infrastructure, developers need to be able to evaluate it on their own terms, which means a real API doc and a real developer experience beat a good video. I've been procrastinating on being on camera myself, and the honest version is that nobody cares about the video half as much as they'd care about a good API.
Chinat's side is moving in the opposite direction in a useful way. MentorMates just launched stronger agentic event features, which is exactly the kind of real context Pulse wants to plug into. A hackathon is a high-stakes, time-bound environment with real builders, real judges, and real sponsors. If Pulse can be the agent layer inside an event like that, we get live feedback that no synthetic usage would give us. The Pulse × MentorMates integration is starting to look less like a nice-to-have and more like the cleanest early test case for the API thesis.
The check I'm making on myself: API docs drafted and the first developer sandbox working by the next mastermind. If I'm still recording marketing videos instead, that's me avoiding the harder shape of work.