Short answer: On October 7 we said we'd ship one new feature of the sqrly Shopify app every day until October 20: fourteen features in fourteen days. We shipped 30 in thirteen days, each one live and on the changelog the day it landed. Speed held up. Quality didn't hold up fully: four problems got past our test store and into production. Each one was fixed the same day, but our goal was zero. This post covers both, and what we're keeping.
When we announced Shiptober, we promised this write-up for October 21, including the days that didn't go to plan. Here it is.
What we shipped
Thirty features, October 7 to 19. Grouped by what they do for a store:
Teaching the digital twin, with less work from the merchant
- It starts with what the web already knows about the shop, for the merchant to accept, edit or dismiss.
- It interviews the merchant: eight questions about the catalog, answered by typing or talking.
- Merchants can teach it by talking instead of typing.
- Fix an answer right in the transcript, so the twin learns it for next time.
- It tidies up its own knowledge, flags knowledge that disagrees with itself, and sorts the merchant's notes into facts and rules.
- It knows the store's Shopify discounts without anything set up twice.
Selling on the storefront
- It speaks the shopper's language, from knowledge written once.
- First questions to tap, taken from what shoppers really ask.
- When it doesn't know, the shopper can ask to hear back.
- An expert box as big or as small as the store wants, a quiet style that matches a design-led store, and a way to share the corner with a rewards button.
- Shoppers rate its answers.
Ready for AI shopping agents
- We sent AI agents shopping through the twin and fixed what tripped them up.
- Every store gets an "Ask our expert" page on its own domain, and the twin sits right where the choice happens.
- Agents can consult the twin directly, next to Shopify's own storefront tools, and get back the exact products for the cart.
Seeing what the twin does, and whether it sells
- Watch it work, live, readable at a glance and ready to present.
- This week at a glance, the sales funnel from question to order, and confirmed orders without access to the store's order history.
- One health score, which answers help, and a win of the week to share.
- A name and a face for the expert, and a first moment that shows it working right after install.
- Fifty attacks to put the twin through before shoppers do.
The full list, with dates, is on the changelog.
What changed from the plan
We built ahead, then changed the rule. By day three, ten features were done. We'd planned to hold finished features back and post one a day. That felt like theater, so from October 9 every feature went on the changelog the moment it was live. Some days had one feature; October 14 had seven.
The lineup moved. Six of the original seven planned features shipped. "Usage and limits" didn't: with no paid plan yet, there's no limit to show, so it waits until there is. Most of what we added came from testing with AI shopping agents and from talking to shops. One shop has a rewards button in the corner of its store, another a quiet, editorial design. Both requests became features within a day.
October 20 had no feature. We spent it getting sqrly ready for the Shopify App Store instead: listing, review video, data protection answers. As of today, sqrly is submitted for review.
What we measured
We said we'd track three things.
Idea to live: under a day. We mostly hit it. Most features went from "let's build this" to live in production the same day. A few slipped to the next morning, usually because a slow build or a merchant check came at the end of the day. The rhythm that worked: a short spec the evening before, build in the morning, check on the test store at midday, release in the afternoon.
Problems getting past our test store: zero. We missed this one. We counted every problem that got past our test store and into production, and there were four:
- Rate limits that didn't work the way we thought. On the storefront, Shopify passes shopper requests to sqrly without the shopper's address, so a limit per visitor treated every shopper of a store as one person. Busy stores would have been throttled. We found it in a testing pass on day three. Limits now count per conversation and per store.
- Two security problems, found in a full review on October 14. Both were already in production and could be triggered with a crafted link. We fixed both that evening, and a full review of the whole app is now a regular step, on top of checking each change.
- A gap during every release. For a few seconds while a new version starts, the twin couldn't answer. A shopper who asked at that moment got an error. Now the storefront retries quietly for about half a minute, so a release goes unnoticed.
- Going public broke store access. Switching sqrly to a public App Store app changed which access keys Shopify accepts, and admin features on our own stores stopped working. We caught it on our review store within hours and converted every store's access the same day.
The test store did catch a lot. A shopper's email saved twice as a Shopify customer, a spoken answer turned into a non-answer, interview questions in the wrong language: all three stopped there. But four problems in production is four more than we wanted. Speed only counts if shops can trust what we ship, so the review step stays.
What merchants use: too early to say. Two weeks isn't long enough to see which features become part of how a shop sells. We'll come back to this once there's enough to go on.
What we're keeping
- Two branches, test store first. Every change goes to a staging app on our own store before it reaches anyone else's.
- The changelog entry as the definition of done. Writing it in plain language kept catching features that were harder to explain than they should be. It also left us a public record of every one of the 30.
- Small enough to finish in a day. When something didn't fit, we cut it into a piece that did.
- Leveraging AI coding where possible. It's a big part of how a small team kept this pace. The judgment stays with us: what to build, how it should feel, and when it isn't ready.
- A regular full review. That's new, and it's the lesson from the second problem.
What we're dropping is the daily feature count. Thirty features is a lot for a merchant to take in, and the next stretch is about the shops using them.
What's next
App Store review, and our first pilot shops. sqrly is live on Shopify today, with more platforms to come.
If you run a store where good advice sells (gear where fit, compatibility or use case decides the sale), we're still accepting a few pilot shops: free for 60 days, and set up on a copy of your store first. Email florian@sqrly.ai with your store's URL and what you sell.