How We Built Zaplit Using Our Own Agent Teams
We use our own Secretary, Research, and Support agents to run daily operations. Here's what we learned.
Eating Our Own Cooking
From day one, we've run Zaplit's internal operations using Zaplit agents. This was partly a forcing function—we needed to know our product worked before we asked clients to trust it—and partly genuinely the fastest way to get things done with a small team.
Here's what we actually use, and what we learned.
The Secretary Agent
The Secretary agent is the orchestrator. It handles task routing, follow-up tracking, and meeting coordination. When someone on the team says "we need to follow up with the Johnson account next Tuesday," Secretary creates the task, assigns it, and surfaces it at the right time.
What surprised us: how much cognitive overhead this removes. Not having to remember to follow up, not having to track down who said what would happen—it sounds small until you've lived without that overhead for a week.
What we learned: The Secretary agent is most valuable when it has context. We invested time in building good integrations with our communication tools so it has the full picture of what's happening.
The Research Agent
The Research agent handles competitive intelligence, market monitoring, and document analysis. When we're evaluating a potential technology partner, Research pulls together a briefing—product overview, pricing, customer reviews, recent news—in minutes.
What surprised us: the quality of synthesis. We expected to get data back and still need to analyze it ourselves. Instead, we get structured analysis with key takeaways and flags for things that need human attention.
What we learned: Research is only as good as its prompts and scope. Early on, we got reports that were technically accurate but missed what we actually wanted to know. Investing in clear briefing templates paid off immediately.
The Support Agent
The Support agent handles first-line responses to inbound inquiries—from potential clients, from existing clients, and from general contact form submissions.
It drafts every response. Nothing goes out without a human reviewing and approving it. This isn't slowness—it's us staying in the loop on every conversation that matters while the agent handles the drafting.
What surprised us: how much tone matters. We spent more time calibrating the Support agent's communication style than any other agent. Once we got it right, responses stopped feeling like they came from a bot.
What we learned: Training the agent with examples from our best human communications, not generic templates, was the key to getting tone right.
What Running on Your Own Product Teaches You
You find the bugs. The edge cases. The moments where the product asks too much of the user or gives too little context. Using Zaplit every day meant we shipped fixes to real friction points that might have taken months to surface through external feedback.
It also meant we were never shipping something we weren't confident in. If we weren't comfortable with our own agents doing something, we didn't ask clients to be comfortable with it either.
The Meta-Lesson
Building a company with AI agents isn't about removing humans. It's about deploying human attention where it creates the most value. Our team focuses on product, client relationships, and strategic decisions. The agents handle coordination, research, and communications volume.
We're a small team doing work that used to take a much larger one. That's the promise of Zaplit—and it works because we built it to keep humans in control the whole way.