A grocery list that remembers the route

Some updates are especially satisfying because I know I will use them myself. My Next: List 1.3.2 has one of those.

A small thing for the weekly shop

I walk roughly the same route through the grocery store every time. A list sorted alphabetically, or by when I added each item, rarely follows that route. List can now learn the order I actually check things off at a saved store. The next time I open that grocery list there, it can suggest putting the items in the same order. I choose whether to apply it, and I can take the list onto my Apple Watch.

It is a small feature, but a nice one to build. I use List every time I go grocery shopping, and this makes it a little easier to keep moving instead of searching the list for the next shelf. Store recognition uses optional location access, checked only while I use a grocery list or a store feature. One saved store is included free.

Fazon is getting close to its first external beta

Since the Fazon website went live, much of the less visible work has been about making the iPhone app ready for someone else's phone. iCloud sync is enabled in release and TestFlight builds; there is a restore path when someone returns on a new or reinstalled device, and a way to delete their data. I have also been working through onboarding, App Store pricing, and the checks around the first purchase. The website now has its first posts in English and Swedish.

There are still beta gates to clear, including fresh-install and commerce checks. If those go well, I hope an external TestFlight beta will be available by the next journal. That is the aim, not an announcement that it is ready today. The waiting list is open for anyone who wants to hear when it is.

Jev is still the second reader

I wrote earlier about trying Jev beside Build Studio's existing text checks. It reads the same agent reports and records where it disagrees, but it cannot change a workflow decision. In the revised verification question, Jev agreed with the existing rule on 367 of 383 reports, or 96%. More useful than the agreement rate are the disagreements. It found about two real cases the rule missed, but most of its other flags were false alarms, often because a failed setup was followed by a clean rerun. It also missed two reports with an explicit marker that the rule catches.

The trial has now expanded, still in shadow mode, to three more decision points. It agrees on the first five builder-role readings. The early terminal-reading sample is more interesting: during a Codex outage, Jev sometimes saw that an agent had exited to a shell several minutes before Build Studio's dead-process check caught up. There are only 11 records for that question so far, though, and Jev's confidence in those cases was modest. The final-report question has no data yet. Across the trial, 644 calls have cost about four cents.

I will keep collecting disagreements rather than give Jev authority on a 96% headline. It needs to be better at the cases that matter, not merely agree often.

Elsewhere

Launch Studio now has an App Store listing workflow, with a review step before pushing changes, and work on search and AI-answer visibility. Välkomna can publish a staged blog post through Launch Studio; a new article about organising a julbord was published that way. Run Studio has added the beginnings of a Gmail connection for receipt work, but its finance workflow is still in development.

What's new in Build Studio

Build Studio has made the less photogenic parts of running agents sturdier. Owner checks can be saved in the queue for later, test-only changes run a narrower set of tests, and an agent can recover more of its context when a terminal session disappears or Codex hits a usage limit. Jev's broader shadow trial is part of that same effort: understand what the agents actually reported before letting a short pattern decide what happens next. The changelog has the full changes and upgrade steps.

This week's smallest release may be the one I notice most often. The next grocery trip is a better test of List than another feature list.