// build note 02

The agent that renegotiated my insurance.

I knowingly overpaid for my policies for more than a decade — because doing something about it was a miserable manual job. So I built an AI agent to do that job with me. It found $1,500 a year.

LIVE $19.99 Insurance AI Analyst · insurance.contextinit.com
Learn more $19.99
fig. 01

The itch: a decade of knowing better

Here's an uncomfortable admission: for over ten years, I knew I was overpaying for insurance. Not suspected — knew. Five policies across home, auto, and more, each renewing on autopilot, each quietly drifting upward.

Why didn't I fix it? Because fixing it is an enormous manual undertaking: read five dense policies, actually understand the coverages, pull competing quotes, compare them line-by-line against what you have — coverage by coverage, not just premium by premium — and then negotiate. Every year. Nobody does this. The insurance industry's margin lives in exactly that nobody.

fig. 02

The design call: with me, not for me

The fashionable version of this product is full autopilot — "upload your policies, we handle everything." I deliberately didn't build that.

Insurance decisions are judgment calls about your own risk: what deductible you can stomach, which coverage you'd actually use, what your home office gear is really worth. An AI shouldn't make those calls for you — but it's superb at the part that kept me procrastinating for a decade: the reading, the extraction, the comparison grid, the homework.

So I built it as an agent that works through the whole process alongside you — it does the labor, you keep the judgment.

fig. 03

What it actually does

i.
Extract
Reads each policy and pulls out the coverage details that matter — limits, deductibles, riders, exclusions — out of dozens of pages of carrier prose.
ii.
Compare
Builds an agent-driven comparison across your current policies and competing quotes — coverage by coverage, so a cheaper premium hiding thinner protection gets caught, not celebrated.
iii.
Negotiate
Surfaces specific negotiation points — where you're over-covered, where a competitor beats your carrier, what to actually say — so you walk into the renewal call armed.
fig. 04

The receipt

I ran the full process on my own five policies first. The result, on real renewals:

$1,500+ / year SAVED ON MY OWN RENEWALS — BETTER COVERAGE, LOWER PREMIUMS

That number is the entire business case. A tool that costs a small fraction of what it recovers — every single year — doesn't need a clever pitch.

fig. 05

The open decision: free and open-book, or paid and packaged

This is the part of the note I can't finish yet, because the decision isn't made — and I'd rather show the decision honestly than pretend it's settled.

Ship it as a pluginFREE

Distributed as a Claude plugin — which means every prompt is readable by every user. Open-book by design, zero friction, and no commercial entanglement alongside full-time work.

Ship it as a servicePAID

An MCP/API-based product with a one-month license. Protects the prompt engineering, funds upkeep, and prices at a small fraction of the first year's savings.

WHAT I'M WEIGHING

The pricing question isn't really about revenue — it's about what this product is for. Paid creates obligations: support, uptime, a second job. Free creates reach and keeps the incentives clean. The deciding factor will be which version still makes sense a year from now, next to a full-time role.

fig. 06

What's next

It's in testing now — being run against real policies and real quotes until the comparison output is something I'd stake my own renewals on (because I literally do). Launch follows the pricing decision. If you want to know when it lands, the notify list on the homepage is the only place I'll announce it.

// build note 02 of 02 previous: storage you can actually own
Learn more $19.99