A progress letter from Equine Data Systems for investors, partners and friends. Eight days ago EqIQ was a plan on paper. Today it is a working website you can click through.

Before you buy a used car, you can check a CARFAX report. It tells you who owned the car, what got fixed, and whether anything bad happened. Horses have nothing like that. A horse can cost as much as a house, and people still buy them based on screenshots, paper files and phone calls.
EqIQ fixes that. It gives every horse a permanent record that follows the horse for life, no matter who owns it or which vet it sees. The people who know the truth about a horse, like its vet, its farrier, or the breed registry, sign the facts they add. The owner decides who gets to look.
The first version of EqIQ does three jobs:
Each horse gets a 12-digit EqueID, like a license plate that never changes. People, places and events get one too. The horse's microchip number is the physical anchor.
Every important fact, such as a vaccination, a competition result, a sale, gets written once into a chain. Each new entry is locked to the one before it. Change an old entry and the chain breaks in plain sight.
A step-by-step purchase flow that moves ownership when the deal closes. An auction room. A vet's office where a visit becomes a signed record on the horse's page the moment the vet finishes.
The build was fast because the thinking was already done. Here is how the idea turned into a plan.
The eqiq.io web address is registered. For a while the working name was EqueFacts. That name lives on today as the name of the horse report.
Recorded planning calls start in June. In the fall, the team talks with payment and blockchain partners, including Platinum Payment Systems, and settles on the pitch that stuck: a Carfax and Zillow for horses, with a record that lasts the horse's whole life.
One big writing pass produces the plan: who the customers are (serious buyers and sellers first, then breeders, riders, vets, insurers), how EqIQ makes money, which rules apply, and what could go wrong.
Market size, unit economics, a five-year model and the "wedge": start as the record-keeping tool that vets and registries rely on, and grow from there. The plan gets a pressure test against similar companies.
A deep look at everyone nearby: vet software, farrier apps, barn tools, show and auction platforms, horse haulers, and the FEI's new Equipass digital passport. The finding: each one holds a piece, but nobody keeps a record that survives a change of vet, barn or owner. The same month, the design brief for the drawings is written.
One Figma file, "EqIQ w Mobile," with about 60 screens: the home page, seven kinds of search, the horse page with its five tabs, the purchase flow, profiles for vets, farriers, trainers and athletes, stables, events and auctions, the account area, and a vet's office. Every screen drawn for desktop and phone, in Western and English looks. Last touched September 4.
In one day: the PRD (the product's rulebook, with about 130 numbered requirements and ten demo walk-throughs that define "done"), the data map (97 tables simplified down to 78), two big architecture decisions, and a build plan. Two independent reviews the same day found 63 issues before a single line of code was written. Then the build started.
Most of the work was done by teams of AI coding agents working in parallel, each on its own copy of the code, with a reviewer checking every bundle before it went in. Karl made the calls.
Screens below are real screenshots from the live demo, taken today. Every horse, person, price and ad is sample data. The ad artwork comes from the design file and does not mean those brands are sponsors.
Nothing you can click, and the most important part. Built first, on the first night, so that everything after it could be built fast and safely.
The heart of EqIQ. One page per horse with five tabs: Horse Data, Performance Stats, EqueValue, Health and Care.




Every attestable fact about a horse goes into a chain. Each entry is hashed to the one before it. A hash is like a fingerprint of the data: change one letter and the fingerprint changes.


Seven kinds of things to search: horses, veterinarians, farriers, athletes, trainers, stable suites and carriers. Plus events, competitions and auctions. One search template serves all of them.




A listed horse can be bought in a guided flow. When the sale closes, ownership moves in one step and the bill of sale is filed. Auctions have a live room with mock bidding. No real money moves anywhere in the demo.


A separate workspace for veterinarians. A vet writes up a visit, presses Finalize, and the record appears on the horse's Health tab as a signed entry in the chain. The invoice follows.



Youth riders are a huge part of the horse world. A parent links a child's account, sets daily and weekly time limits, chooses how the child's name appears to others, and approves anything that costs money.

My Horses, favorites, requests sent and received, your vet, farrier and trainers, membership and billing, security and preferences.




The demo is private. Invited testers sign in with Google or a one-time link. Owners of the demo manage the list. And EqIQ now sends real, branded email.


Only a platform admin with two-factor sign-in gets in. Every admin action is written to a security log that cannot be edited.




One website that fits every screen. The phone layouts follow the mobile drawings, not just a squeezed desktop.




Every time the plan was silent or two documents disagreed, someone picked an answer and wrote it in a log. These are the ones that made the biggest difference. The small codes point to the log entry.
Three questions people ask right away. How much was built, what did the AI cost, and what would this have taken the old way? Here are the numbers and how they were counted.
Counted from the code repository today. "Hand-written" means an engineer, human or AI, typed it on purpose. Files that machines produce on their own, like the API schema file and the sample horse descriptions, are listed separately and not counted as written work.
Over the eight days, 205,560 hand-written lines were added and 8,443 were removed or rewritten, in 138 commits. The app has 778 source files, 237 test files and 91 page routes.
The build ran on Claude, through Claude Code, Anthropic's tool for coding with AI. Every message it exchanged is on record, so the count below is measured, not guessed. A "token" is about three-quarters of a word.
What that would cost at pay-as-you-go prices: about $5,180 across the four models used. What was actually paid is less, because the work ran on a Claude Max subscription, a flat $200 a month, topped up with extra usage credits when the weekly allowance ran out.
What was actually paid, all in:
The bars show Claude usage by day at list prices. The busiest day was Monday the 22nd, when Discover, the horse record, the ledger, the vet's office, buying, auctions and the family area all went in. The Claude figures are measured from the transcripts; the Codex figure is Karl's estimate. Not included: the ordinary hosting bills for Vercel, Neon and Resend.
How long would this have taken a team of people? Nobody can know exactly, so here are three ways to estimate it, and they land in the same range.
1. Bundle by bundle. Each of the 129 merged bundles was sized the way an engineering manager would: a small fix at most of a day, a medium feature at two to three days, a large one at a week, the biggest at eight days, all including tests and review. Plans and docs at half a day to a day and a half. Total: about 380 engineer-days, or 3,050 hours.
2. Line by line. 174,000 lines of tested, reviewed code at a brisk professional pace of 200 to 400 finished lines a day is 435 to 870 engineer-days, or 3,500 to 7,000 hours.
3. Team and calendar. A four-person team (two senior engineers, one full-stack engineer, one designer who also writes the plans) would typically need six to nine months for this scope, or 4,000 to 6,000 hours.
Three to six thousand engineer-hours of work, done in eight days.
At a typical loaded rate of $120 an hour, that is $360,000 to $720,000 of engineering. It was bought for $1,900 to $2,500 of AI, all in, plus 26 hours of Karl's time: two hours on each of the five weekdays and two full Saturdays, during which he typed about 200 instructions and decisions into Claude Code. That is well over a hundred times cheaper than the human way. Hours saved: roughly the whole range above, minus those 26.
How the work was organized. A lead session reads the plan and writes a short brief for each lane of work. Builder agents take a lane each, in their own copy of the code, so they never collide. A reviewer agent checks every bundle against the plan before it can merge, and fixes what it finds. A watchdog task wakes a run back up after a usage limit. Every one of these rules lives in the repository as a written playbook, so a new session, or a new person, can pick up where the last one stopped. The numbers above are what that machine produced in its first week.
Every milestone in the original build plan is in except the last polish pass. Here is what is still open, in order.