Skip to main content
All case studies

American Growth Insurance’s app had a working AI assistant in 4 weeks.

I built an AI assistant into American Growth Insurance’s existing app as a 4-week fixed-scope project, while the company was still in stealth. They needed a product that agencies and investors could try for themselves.

Client
American Growth Insurance logo

American Growth Insurance An AI-enabled growth platform for insurance brokerages, backed by Atomic and Rockbridge Growth Equity.

Length
4 weeks
Industry
Insurance brokerage
  • 4 weeksto a working AI assistant inside their app
  • ~10design-partner agencies used it during stealthPer AGI’s launch post
  • ~$70Min committed equity at AGI’s July 2026 launchFrom Atomic and Rockbridge Growth Equity

The agency count and launch figures are from AGI’s launch post, opens in a new tab and PR Newswire release, opens in a new tab.

“During stealth, I could open our app in front of an agency owner and have the assistant build a sales deck while they watched. Investors saw the same thing. Fadee built it inside the app we already had.”

Luke Sophinos,American Growth Insurance

Where things stood

American Growth Insurance (AGI) works from one thesis: brokerages carry too much back-office work for every producer, the agent who sells, and AI can automate that “dirty work” across real agencies.

Before launch, while still in stealth, AGI needed a working product to show about 10 U.S. agencies that were working with them as design partners. Slides weren’t going to hold up in investor diligence. The product had to handle real agency work: reading documents, research, running code, and writing content. It also needed secure sign-in, isolated code execution, and audit logs.

How we worked

This was a Fixed-Scope Project: 4 weeks, with success written down before I started. Done meant a working demo, the infrastructure to run it, and a handoff.

I built it inside the app they already had, in their repo and on their team’s cadence, and kept it consistent with the app’s existing design.

What I built

I built the assistant, SuperAgent, into their app on two backends. The app’s own backend handles sign-in, users, and saved data, and a second backend runs the AI tools and workflows.

  • Users upload documents and the assistant reads them. It parses files in several formats and pulls out the data, and it can research the web live, with citations.
  • It builds PowerPoint decks, runs code, and scaffolds simple web apps. Code runs in an isolated sandbox, and the user sees the output live.
  • Tools live in a registry, and a workflow manager runs them. Adding a capability means registering a new tool.
  • It works with models from several AI providers. Answers stream to the screen as they’re generated.
  • The infrastructure is defined in code. It deploys through CI/CD, with monitoring, autoscaling, caching, and audit logging.
A user asks the assistant for an insurance sales deck, answers its follow-up questions, and gets the finished slides back.

What changed

  • The assistant shipped inside their app in 4 weeks. It matches the app’s existing design, so it looks like part of their product.
  • About 10 design-partner agencies used it during stealth. Source: AGI’s launch post, opens in a new tab.
  • AGI launched publicly in July 2026. Its launch post, opens in a new tab and PR Newswire release, opens in a new tab report ~$70M in committed equity from Atomic and Rockbridge Growth Equity, a first acquisition closed, and a path to $10M+ in revenue through a roll-up. Those are AGI’s results as a company. My part was the assistant.
  • The assistant was used to test the product and to raise capital. What AGI learned in that testing led to a strategic pivot, and a new team rebuilt the product.

Tech stack

Frontend

  • React 18
  • Material UI (MUI) with Emotion

Backend

  • Express.js
  • TypeScript
  • Drizzle
  • PostgreSQL (Supabase)
  • AWS Cognito

AI processing

  • Models from OpenAI, Groq, Anthropic, and OpenRouter
  • Streaming over server-sent events (SSE)
  • Daytona sandboxes for code execution
  • Dynamic tool registry and workflow manager

Infrastructure

  • Terraform and Terragrunt
  • CI/CD, monitoring, and autoscaling
  • Caching and audit logging

More case studies

Tell me what you’re trying to ship.

Free · no prep needed · proposal in 48h