Binder

An AI-native CRM for independent insurance agencies. Not an AMS replacement. The producer workspace that sits next to your AMS and runs the quote, bind, and renewal workflow with autonomous agents instead of a pile of tabs.

Why I am building this

I am an independent AI researcher. Most of my writing is on autonomous agents, MCP, and Claude Code. Over the last year I kept running into the same conversation with independent insurance agency owners:

"Our AMS is a filing cabinet. Our CRM is a spreadsheet pretending to be Salesforce. Every producer has their own Gmail folder labeling system. Every renewal is a fire drill. The AI stuff everyone is talking about has not shown up yet, at least not in a way that actually moves work through the agency."

That is a shaped problem. Independent agencies are a fraction of a percent of the Fortune 500 budget but they run a huge portion of how insurance actually gets distributed in the US. And the tools serving them were designed for a world where faxes were the fast channel.

Binder is the working prototype of what I think an AI-native producer workspace should look like. This page is the public version of that research.

The research question

If you rebuilt the producer's day from scratch, using agents instead of checklists, what would it look like? Specifically:

1. What parts of the quote-to-bind flow are judgment calls (stay with the human) versus data gathering (agent).
2. How does a renewal actually happen today, and which 80% of renewals could run on autopilot with a human checkpoint.
3. Where does cross-sell die, and what information would have kept it alive.
4. How do agencies onboard a new producer today, and what does that look like when the agent knows every prior policy, note, and email thread.

What I have found so far

From 14 interviews so far with independent P&C agencies (3 to 40 producers), the pattern is consistent:

• Selling is a minority sport. Producers spend roughly 10-15% of their time on the thing they are paid to do. The rest is orchestration, documentation, and chasing.
• Renewals are the loudest fire. Agencies process 30 to 70 percent of their book on renewal each year. Most of that work is sequential, deadline-driven, and done by hand.
• The AMS is for after. Policies get recorded in the AMS after the deal is done. The actual deal lives in emails, shared drives, and the producer's head.
• Existing "AI features" are summarization at best. A small number of agencies have tried carrier portals that summarize quotes. Nobody I have talked to has a system that can read an email thread, detect a quote request, assemble the submission, and route it to the right carrier.
• Tech fatigue is real. Most agencies have been sold 4-6 pieces of "game-changing" software in the last three years. Adding a seventh is not interesting; replacing two or three with one thing that actually works is.

What Binder does (prototype)

Four modules in the current prototype:

1. Inbox intelligence. Binder reads the shared inbox (Gmail or Outlook, OAuth, no forwarding setup). It classifies every incoming message, pulls out quote requests, renewal questions, and service items, and drafts the next action for the producer to approve or edit.
2. Submission builder. When a quote request lands, Binder assembles the carrier submission from the thread, prior policies in the AMS, and the agency's underwriting notes, then routes it to the carrier of best fit.
3. Renewal orchestrator. Binder watches the renewal window on every policy in book and starts the renewal workflow 60 days out without being asked. Pulls loss runs, flags rate changes, and drafts the renewal conversation.
4. Cross-sell radar. Continuously scans the book for coverage gaps (has a home policy, no umbrella; has a commercial auto, no workers comp) and surfaces a one-click outreach sequence when gaps look worth the call.

What Binder is not

• Not an AMS. Your AMS stays the system of record. Binder reads from it and writes back, but it does not try to be it.
• Not a generic sales CRM. HubSpot and Salesforce are general purpose. Binder only works if it understands policies, carriers, LOBs, renewal dates, and loss runs. The insurance model is the whole point.
• Not a chatbot. The work gets done by agents running in the background, not by a person asking an AI to do it one task at a time.

The pilot cohort

I am running Binder with a small cohort of independent P&C agencies. The goal of the pilot is not revenue, it is research: to see what breaks when you try to replace a producer's tab-switching with autonomous agents.

What pilot participation looks like:

• Free use of Binder for 90 days.
• Weekly 20-minute check-in with me.
• Access to raw findings from other agencies in the cohort.
• Your agency shapes the roadmap; the features that ship are the features the cohort actually needs.

What I ask in return:

• Honest feedback. I would rather hear "this is broken" early than "this is fine" politely.
• Willingness to let Binder read inbox traffic with appropriate access controls.
• A 15-minute intro interview before we start.

Request a research interview

If you run or work at an independent insurance agency, I want to talk to you. Even if you are not interested in piloting Binder, the interview itself is how this research gets built. 15 to 25 minutes, on video, zero pitch.

What we will cover:

1. How your agency actually runs day to day.
2. What the last three producers to leave or join taught you.
3. The one workflow that if it broke tomorrow would shut the agency down.
4. Where AI has and has not earned its keep so far.
5. What the 2026 version of the agency looks like, if you got to build it.

FAQ

Is Binder shipping yet?

No. Binder is in research and prototype stage. A small pilot cohort is running now and I am opening a few more slots.

Who is Binder built for?

Independent P&C insurance agencies with roughly 3 to 25 producers. Agencies that feel stuck between an AMS designed in 2005 and a generic sales CRM that does not understand policies, carriers, or renewals.

How is this different from an AMS?

An AMS is the system of record for policies. Binder is not trying to replace your AMS, it sits next to it and runs the producer workflow: lead qualification, quote coordination, cross-sell, and renewal orchestration, all driven by agents that read the same email threads and shared drives your team already uses.

What does a research interview involve?

15 to 25 minutes on video. You talk about how your agency actually works, what is painful, what you have tried, and what a fixed version would look like. I take notes, ask follow-ups, and share back patterns I have heard from other agencies. No pitch. No recording unless you ask.

Do you pay for interviews?

No payment, but pilot participants get free access to Binder for the duration of research and heavy weight in shaping the product.

What is your stack?

Claude, Anthropic API, a custom agent harness built on the Claude Code patterns, MCP servers for AMS and carrier integrations. The design of the agent system is public on this site; the product code is not.

Where does my data live?

Agency data stays in your environment. Binder reads via OAuth and writes back to your AMS over its existing API. Nothing is used to train foundation models.

Who are you?

Samuel Ochoa. Independent AI researcher and writer. See about and the framework for how I think. Past serial entrepreneur, currently focused on research into autonomous AI systems, with Binder as the applied half of that research.

Related reading on this site