Talkstore Begins Now! 🟢 Building to $100K
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 stream documents the public start of building Talkstore: a proposed Shopify-specific AI assistant through which merchants could manage products, storefront work, and commerce intelligence conversationally. Its capabilities and possible expansion beyond Shopify are presented as conditional directions, not validated outcomes. Practical work covered Shopify app and development-store setup, Supabase integration and Prisma migrations, secret handling, and repeated diagnosis of session, connection, preview, and Railway deployment problems. The preview and some setup steps worked, and the service later appeared active, but several database, network, build, and deployment issues remained intermittent or unresolved in the covered segments. This led to useful lessons about starting with logs, documenting a stubborn problem, seeking a fresh diagnostic perspective, and reviewing coding-agent compromises rather than accepting a result merely because it works temporarily. The stream also presents a multi-agent command center that separates research and task preparation from implementation, code, security, and compliance review, using dependencies, an adaptive timeline, and human approval gates. Luqman felt this structure accelerated and preserved the context of the work, while still questioning whether every agent was running and visibly logged. He emphasized that authentication, secrets, and nuanced decisions continue to require human intervention, and that task count is a poor measure of difficulty. Work later shifted toward pre-launch distribution: a simple but polished Astro waitlist site kept separate from the Shopify app, with email capture, analytics, social metadata, a teaser, and a distribution campaign. The first landing-page draft had missing hero content and styling or rendering failures, so it was treated as an unfinished iteration. Other useful discussions included audience questions about using AI to explore income paths from existing expertise, SaaS unit economics, choosing between ads and SEO/GEO, Arabic-market WhatsApp automation, and targeting external markets, plus local voice dictation, content automation, and Telegram community options. Stream lag and interruptions were also investigated without a conclusive cause, while music licensing and gym culture appeared as brief secondary discussions.
121 moments from the stream
- 0:57:41
Diagnosing a Shopify App Preview Preparation Failure
After the development server starts and loads the environment variables, preview preparation fails without exposing the complete error message. More detailed output is considered to locate the source, followed by a recommendation to install a Node version manager for easier version switching; the page does not confirm a final fix.
- 1:00:33
When a Coding Agent Needs Human Intervention
Luqman explains that an agent may handle much of a task smoothly, while steps such as signing in or supplying inaccessible codes and information still require human assistance.
- 1:01:33
Troubleshooting a Shopify CLI Dev Preview Failure
Luqman tries installing and selecting Node 22 through a version manager, opening a fresh terminal session, and reviewing the setup instead of continuing to guess. Some steps succeed, but the development-preview startup error remains unresolved.
- 1:08:58
Setting a High Bar for the Pre-Launch Teaser
Luqman describes the teaser video as an important part of the pre-launch marketing campaign, saying his concern comes from his high standards and desire for a result he finds genuinely strong.
- 1:15:35
Troubleshooting and Previewing Talkstore Locally
After suspecting the problem was in the backend rather than the terminal, a fresh session was opened and the preview worked. Luqman inspected Talkstore’s early interface and GraphQL interface, then asked how to launch the development preview reliably in future.
- 1:19:11
Refocusing Agents on the Original Task
After drifting into launch and preview work, Luqman asked the explorer and researcher agents to inspect the task and determine whether it was actually complete, preserving the original workstream.
- 1:21:49
Multi-Agent Task Preparation with a Human Review Gate
Luqman explained that he preferred observing the full workflow before immediately starting parallel work: researcher, explorer, and Talkstore agents prepare the task file, ask for operator input when needed, and pass a refined prompt to a builder agent. The task remains in review until the human operator approves it as done.
- 1:26:24
Human–Agent Collaboration for Railway Deployment Setup
A Railway deployment-pipeline task was treated as human–agent collaboration because some platform steps required operator action. Luqman said he had not used Railway before and that agents recommended it after assessing the project, while API-key and secret handling stayed off-screen.
- 1:31:55
Troubleshooting a Failed Railway Deployment
A discussion of a failed Railway build and uncertainty over whether Railway was using the Dockerfile instead of its default build mechanism. Updating the Dockerfile for the React Router 7 stack and redeploying were proposed, but the captions do not confirm success.
- 1:36:57
Starting with Output and Logs When Diagnosing a Build
A practical debugging approach that starts with build output because it reflects the effects of the changes, then steps back to inspect the code and logs when needed.
- 1:38:06
Calibrating an AI Coding Agent’s Commit and Push Permissions
A discussion of allowing a coding agent to commit and push to GitHub after it requests approval, alongside the observation that an earlier blanket prohibition may have been too restrictive and could need adjustment.
- 1:44:15
A Claude Permission Rule Blocks Commit and Push
Luqman diagnoses a failing GitHub workflow: a strict rule prevents Claude from committing or pushing before requesting permission, but appears to keep blocking the action even after permission is granted. He removes a hook he had added and looks for a configuration that can push and check automatically; the attempt is still unsuccessful in this segment.
- 1:48:29
Adding Guidance While Claude Keeps Working
He explains using the “/btw” command to send Claude an extra note or requirement in the middle of a task, allowing it to respond and continue working with the added guidance.
- 1:50:21
Reviewing Pipeline and Supabase Tasks
He reviews the next work items: a pipeline task followed by Supabase-related tasks. He notes that the tasks may be demanding because they involve collaboration, and that the required Supabase work still needs clearer understanding.
- 1:51:59
Why Coding-Agent Compromises Need Review
Luqman explains that a coding agent may make compromises to resolve a stuck task, with consequences that can surface later. He cites an example where the agent removed a metafield to get Shopify CLI working, emphasizing the need to understand what changed, why it changed, and whether it could cause future problems rather than accepting an immediately working result.
- 1:55:33
Handling Repeated Deployment Failures at the Image-Build Stage
During a live deployment attempt, Luqman contrasts the experience with an earlier setup that required fewer configuration changes and observes repeated failures around the image-build stage. He tries giving the agent broader context or more reasoning and suggests this could save time if it works, but the captions do not confirm a successful deployment.
- 2:00:25
Week-One Plan: Distribution Campaign and Waitlist Landing Page
Luqman briefly reviews the work board, noting that most of week one centers on a distribution campaign and a waitlist landing page.
- 2:01:16
Diagnosing a Railway Image-Build Failure
A repeated build failure was investigated despite logs that initially lacked enough detail. The error output pointed to the Prisma directory possibly being excluded by dockerignore and package-lock being ignored, but the build still failed after an attempted correction, so the final cause remained unresolved.
- 2:09:03
Breaking a Debugging Loop by Documenting the Problem
After repeated attempts stopped producing progress, the proposed strategy was to leave the current loop, review the Railway settings, and capture the problem and its context in a Markdown file so another model could continue the diagnosis from a different angle.
- 2:13:43
Handing Deployment Failure Context to Codex
A document summarizing the failure and repository state was used to give Codex context from a lengthy earlier attempt, while observing its reasoning and ability to make repository changes. The captions do not establish a final result.
Got a question about something in this stream?
Ask in the group