Shopify App Submission
Want to back the stream? The «داعم» membership on YouTube is EGP 60 a month: badges, emoji, and I answer your questions first on stream. Become a supporter
The gist
The July 7, 2026 session follows preparation of a Shopify app focused on improving merchant visibility through AI. The speaker walks through brand-context onboarding, discovery queries, product-data optimization, article and image generation, attribution, integrations, and a concept for progressively enabling automated publishing as merchant trust develops. Practical work centers on reinstalling and checking post-install behavior, clearly presenting pricing, trials, and usage limits, and preparing an English screencast and app listing. The discussion concludes that critical failures and broken workflows—not every remaining polish item—are the immediate review priority: issues appeared around generated-output persistence and smart posting, while some scheduled automation, export functionality, and product areas remained unfinished or untested. The stream also covers technical debt, isolating and verifying one task at a time, managing coding-agent context, and delegating implementation to fresh sessions, alongside qualified opinions about Codex and Claude rather than a definitive comparison. Audience questions address learning to build software with AI and an unclear VS Code run problem; the speaker stresses that diagnosis requires enough concrete context. On the business side, he sees growing competition in Shopify’s AEO category and expects marketing, service quality, and premium positioning to matter, while leaving room to adjust pricing or product positioning in response to market metrics. His historical expectation at the time was a Shopify review response in roughly two to four weeks, possibly five, not a guaranteed timeline.
49 moments from the stream
- 0:49:48
Preparing the App State for a Shopify Reviewer
Because the app depends on automation and merchant brand data, a reviewer installing it on a development store may encounter a slower or less populated initial state than the one being demonstrated. The speaker says this should be explained in the review presentation.
- 0:50:28
Discovery Queries and Post-Install Issues
An explanation of measuring merchant discovery queries and showing AI-model responses, mentioned brands, and presence or proximity scores, while acknowledging that bugs are concentrated after installation and some issues remain under repair.
- 0:53:57
A Brand Manifesto as Shared Context for App Features
The discussion describes researching a store to draft a brand manifesto instead of requiring the merchant to complete a long, drop-off-prone form. The merchant can refine elements such as the name, mission, audience, competitors, and supporting signals, after which the information serves as shared context for discovery queries, articles, FAQs, and AI behavior across the app.
- 0:57:48
Product Optimization and Progressive Autopublishing
The system evaluates the completeness and quality of product data—including descriptions, tags, variants, metafields, and search-engine titles—then proposes improvements and displays a projected score. Automation is introduced in levels: optimization begins without automatic publishing, and broader publishing capability is unlocked after the merchant reviews results and publishes some changes manually, building trust before automating sensitive product changes.
- 1:00:30
Continuous Autopilot for Product and Article Optimization
The speaker explains an autopilot concept that can take over after reaching a suitable level, then continue optimizing products and publishing updates over subsequent months. A similar, stronger concept is discussed for articles, with the qualification that parts still need work but do not appear to block the app submission.
- 1:02:10
Conversational Video Prototype for App-Listing Media
The speaker reviews an early video prototype intended for the app listing’s featured-media area rather than general marketing. The proposed format is a conversation that introduces a ChatGPT-related problem and presents the app as a solution, while noting that the graphics are still provisional and that the speaker wants design feedback on the presentation.
- 1:07:15
Mocking the Dashboard Before Implementing a Fix
Before implementing a proposed fix, the workflow first calls for deciding how it should render on the dashboard and mocking the dashboard’s current state with the widget mock, ensuring shared understanding before proceeding.
- 1:08:33
Attribution Pixel, Integrations, and API Access
The speaker describes an attribution pixel independent of Shopify’s own measurement for identifying traffic sources and relating them to revenue and order metrics, including traffic from AI services. The discussion also covers an analytics integration, off-platform access through an API, webhook subscriptions, and white-label options. During the readiness review, log export is identified as unfinished work needed before submission, alongside a concept for invitation-based multi-store management.
- 1:11:14
Creating a Concise Shopify App Presentation Video
A discussion of an approximately one-minute video that introduces a product quickly, sitting between a demo and marketing. The suggested approach emphasizes a clear, polished presentation of the product, distinct from either a shorter advertisement or a lengthy walkthrough.
- 1:16:43
Challenges of Debugging Scheduled Automation
An explanation of why automation that runs tasks on a cadence is difficult to test. A single failed scheduled step can create a bug that requires a failure-handling process, while verifying correct behavior may require waiting for future events rather than observing results immediately.
- 1:20:12
A Practical View on Reviewing Code with Multiple Assistants
The speaker says Codex completed a task well for him and noticed issues that another assistant—whose name is unclear in the captions—missed. He presents access to a different model for code review as a potentially useful advantage, not as a systematic test or universal conclusion.
- 1:20:51
Separating Product Gaps from Shopify Submission Blockers
He considers the activity section incomplete because some desired activity is missing, but not a review blocker because a reviewer would not encounter an obvious failure. A remaining submission step is to record an English screencast explaining the app’s features and send it to Shopify with the listing prepared.
- 1:22:52
Using AI to Write Code Without Traditional Coding Study
Answering a question about learning to code, he says his focus was learning how to communicate with AI tools and use them to produce code. For someone without a programming background, his advice is to explore multiple ways of working with AI and follow the field’s changing capabilities and direction.
- 1:25:18
Codex or Claude for Coding? A Use-Dependent Comparison
The speaker says the choice between Codex and Claude depends on the task. His view at the time was that Claude led for coding, especially “vibe coding,” while broader intelligence claims or benchmark results did not necessarily settle practical coding performance; he expected to reassess after an upcoming release.
- 1:27:19
Why Coding Assistants Can Seem Worse as a Project Grows
The speaker suggests that a model seeming less capable over time may not mean the model itself became worse. As a project grows and becomes more complex, simple instructions can produce less predictable results. He presents this as a possible personal explanation, not a proven finding.
- 1:29:50
Reviewing Risk Areas and Delaying App Marketing
The speaker expresses confidence in the catalog section after substantial work, while flagging concern about other sections whose names are unclear in the captions. He is eager to market the app but says he will wait for a further app-related step; the final word identifying that step is missing from the supplied captions.
- 1:30:45
Preparing the App and Funnel Before Ad Spend
The speaker says the app was still unfinished when submission was being discussed, and that it still needed polishing and marketing infrastructure before marketing began. The plan is to prepare the full funnel, email marketing, support system, and help center before testing digital ads, then use acquisition-cost and margin metrics to form an early view of market response. Changing pricing or product elements is presented as a possible response if the metrics do not support break-even, not as a proven outcome.
- 1:34:58
Delegating a Separate Task in a Coding Workflow
The speaker briefly demonstrates a tool that creates a separate task from a work tree and places that task in a debugging section. The command name and some tool details are unclear in the captions, so this remains a general example of splitting coding work into subtasks.
- 1:37:11
Avoiding Guesses When a VS Code Run Problem Is Unclear
An audience member asks about a program created with Claude that does not behave as expected when opened or run in VS Code. The speaker asks for clarification and says he cannot offer a solution or write code before understanding the problem, so the discussion reaches no confirmed diagnosis.
- 1:40:40
Getting Enough Context Before Diagnosing a Software Problem
The speaker explains that he cannot diagnose a program that is not working without specific information about the software and the problem. He suggests describing the bug to an AI coding assistant. He adds, based on his own recent experience, that repeatedly building something that does not work as intended can be a normal part of troubleshooting.
Got a question about something in this stream?
Ask in the group