We are a town of eleven hundred with an elevator and a bank. Is custom software honestly worth it here?
Sometimes not, and we would rather say so early than take a deposit. The rule of thumb we use is whether the thing you want fixed is costing somebody an evening a week, or has already cost real money once — a disputed load, a tile run nobody could find, a file that missed a deadline. If it clears that bar it is usually worth a small, fixed-price build. If it does not, we will tell you what to buy instead.
Harvest is chaos and nobody is going to learn a new system in October.
Correct, and we do not ask them to. Anything intended for harvest goes live in July or it waits until next year. The other thing we do is build so the busiest people touch it least — a driver photographs a ticket, and that is the entire interaction. Learning curves belong in the office, not in the cab.
Could this recommend a rate or a product to a farmer?
No, and we would not build it that way if you asked. Rate and product decisions belong to a licensed agronomist or applicator, and putting a model in that chair is how somebody ends up in front of the Department of Agriculture. What software should do is check that what was ordered matches the acres and the label and raise a hand when it does not.
Our tile crew does not carry laptops and would not use one.
They should not have to. What we build for a dirt crew is a phone that takes a photograph, a machine that already logs where it went, and an office that turns those into an as-built without anybody typing. If a system requires an operator to sit down and enter something, it will not be used in November and the whole build is wasted.
What happens to our records — is any of it used to train models?
None of it. Grower files, tickets, borrower documents and pupil information stay in infrastructure held in your name, are never shared, and are never fed to anything for training. You can pull a complete export whenever you want it, in ordinary formats, and that right is written into the contract rather than promised on a call.
The district would need this approved and paid for out of public money.
Then it should come with a check the board can run. We scope school work at a fixed price against one measurable thing — usually forms completed without a follow-up call, or hours spent rebuilding routes after a move — and hand over the baseline before we start. We also write the handover notes for whoever inherits it, because staff change and a system that only one person understands is a liability.
Who is actually doing the work, and where are they?
A US-based team, working under an NDA if you want one, and the same people who mapped your process write the code. There is no subcontracting to somewhere you cannot ring. We put the process map in front of you before anything is built, and if it is wrong you say so then, when changing it costs nothing.
How much of this is genuinely artificial intelligence?
Less than the market implies. Reading a ticket, a certificate or a consent form off a photograph is real machine work and it saves hours. Scheduling, billing and record-keeping are ordinary software and we build them as ordinary software, because they must be exactly right every time and there is no sense paying a premium for arithmetic.