Build voice AI that runs a business. Sub-second latency. Real conversations. Production scale.

The Problem

Voice is the lifeblood of insurance. Businesses call brokers. Brokers call underwriters. Carriers call everyone. The entire industry runs on phone conversations—millions of calls per day, most of them repetitive, most of them waiting on hold.

36 million businesses in America need coverage. 77% are underinsured. 40% have no coverage at all. Most would rather just talk to someone than fill out forms. Voice is how trust gets built. Voice is how deals get closed. Voice is where the industry lives.

And voice AI is where most companies fail. Too slow. Too robotic. Too fragile for real conversations.

The industry is 90%+ human-led. We’re building the inverse: 90%+ AI-led, inching toward the higher 90s. Voice is how we get there.

The Thesis

Every industry with human-bounded distribution consolidates rapidly once someone makes it computational. When distribution becomes computational, Jevons Paradox kicks in: increased efficiency leads to increased consumption.

Voice is the hardest part to make computational. Getting voice right is what separates AI demos from AI that operates. Companies like Giga (DoorDash) and HappyRobot (DHL, Ryder) are proving that voice AI can run core business operations at enterprise scale. We’re building the same for insurance.

We’re not building voice features. We’re building an insurance distribution engine where voice AI is the primary interface.

The Role

You’ll own our entire voice AI infrastructure—architecture, implementation, scaling, continuous improvement. Voice isn’t a feature at Harper. It’s a core system that powers sales, operations, and customer service.

Today we have working voice agents. They call leads, collect information, handle customer service. But we’re at 1% of where we need to be. We need to scale from hundreds of calls to thousands. From functional to exceptional. From single use cases to every voice touchpoint in the business.

We’re a growing engineering team looking for a high-agency operator who can own voice end-to-end.

What You’ll Build

Outbound at scale. Lead qualification agents that call thousands of prospects daily. Follow-up systems that never forget. Personalization that makes every call feel human.

Inbound customer service. 24/7 agents handling policy questions, certificate requests, claims intake. Intelligent routing when humans need to get involved.

Operations and underwriting. Form-filling agents that conduct 20-30 minute insurance applications over the phone. Underwriter follow-up systems. Document collection workflows.

Platform infrastructure. Move from low-code platforms to custom infrastructure as we scale. Sub-700ms latency across the voice pipeline. Architecture that handles 10,000+ concurrent calls.

The systems that compound. Capture every conversation as structured decision traces. Build feedback loops that improve agents based on outcomes. Every exception becomes searchable precedent. Design patterns that let us ship new voice agents in days.

You Might Be a Fit If…

You’ve built production voice AI. Not demos. Real systems handling real calls—significant production volume.

You know both fast deployment and custom infrastructure. You’ve used platforms like VAPI or Retell when speed mattered. You’ve built custom systems with Pipecat or LiveKit when you needed more control. You know when to use each.

You understand telephony. SIP, RTP, WebRTC—you know how voice actually works over networks.

You’ve achieved low latency in production. Sub-second response times on real calls. You know where latency hides and how to eliminate it.

You write code with AI. Cursor, Claude Code—you manage multiple coding sessions simultaneously and multiply your output.

You’re 3-6 years into your career. Senior enough to own systems end-to-end. Recent enough to still be hands-on every day.

Compensation

Salary: $170,000 - $275,000

Equity: 0.10% - 0.50%

Location: San Francisco, in-office. We build together.

The Process

1. 15-min founder call — Alignment on mission and pace

2. If in SF: Super Day on-site

3. If outside SF: Technical phone screen, then on-site

The Super Day: You’ll show how you rapidly gain context on a new codebase, use AI code generation to architect and ship, and manage multiple coding sessions simultaneously. We assess voice AI architecture, real-time systems design, and how you think about latency.

To Apply

We’re building a vertically integrated AI platform that connects go-to-market, sales operations, customer service, and retention under one architectural roof. That integration creates compounding through feedback loops—every interaction makes the system smarter.

Voice is the highest-leverage problem in our stack. It’s where most AI companies fail—and where we can build something truly defensible. The AI you build will power real revenue, serve thousands of customers, and feed the loops that make everything smarter.

If that’s the kind of challenge you want—send your resume and links to voice systems you’ve built.

We’re a championship-minded team. We push each other. We move fast. We care about craft. If that sounds like where you belong, let’s talk.