I grew up on an apple and pear operation in the Hood River Valley, and that’s where I first saw what happens when a farm’s records don’t agree. Reminder texts sent every week because there was no other way to follow up. Clipboards that traveled everywhere and still came back half-empty. Equipment repaired three or four times with no running log. Field complaints that vanished before they reached management. Everyone just absorbed the friction as part of the job.
“Every feature starts with something I watched go wrong in an actual orchard.”
None of it touched AI — not for checking a spray regulation before an application, not for a labor-law question, not for diagnosing why a machine kept failing. When I moved into AI work and got a close look at who was building ag tech, I understood why: teams in San Francisco, New York, and London who knew software but had never walked a block at harvest or rebuilt a spray record under pressure. The products showed it — beautiful dashboards that needed three days of setup, compliance tools that assumed a dedicated data-entry person.
Idunn is built from the other direction. Every feature starts with something I watched go wrong in an actual orchard, and the bar is simple:
- Thirty seconds from a phone, or it isn’t done.If a crew boss can’t note a performance issue that fast, the feature hasn’t earned its place.
- If it needs signal in the back block, it isn’t built for here.Records get captured where the work happens — gloves on, no bars, dust everywhere — or they don’t get captured at all.
- If it doesn’t work in Spanish, it’s only half a program.The people doing the work speak Spanish. Anything that’s English-only reaches the office and stops at the orchard, so every feature has to land in both.
