TALKSTORE PIVOT
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 follows TalkStore’s public build after Shopify rejected its storefront domain: related code and references were removed, documentation and the roadmap were revised, and the app began repositioning around product operations. Proposed directions include an assistant that can act on Shopify tasks, understand product images and text, improve content in bulk, and potentially support GEO; these remain unproven ideas, while the product’s replacement advantage, defensibility, and market demand still require research and validation. In practical work, Luqman experiments with a command center that distributes tasks across Codex, Claude, and other tools while tracking dependencies, activity, and reviews. Unresolved issues include task-status synchronization, usage limits, and the cost of parallel agents; his tool comparisons are time-specific impressions rather than controlled tests. He also discusses separating engineering from debugging, QA, and manual pre-submission review, and establishing an adaptable billing foundation while plans and pricing remain provisional. Audience questions broaden the discussion to validating demand before building, software economics, distribution and B2B adoption, and dependence on platform rules. The main supported conclusion is that coordination, market validation, and distribution matter as much as build speed, while the product, workflows, and tool experiments remained works in progress rather than completed successes.
68 moments from the stream
- 2:14:44
Choosing an AI Coding Tool Under Usage Limits
Answering a question about subscribing to one model, the speaker said Claude was strong for software building but had particularly restrictive usage limits at the time. He noted that the issue had reportedly been under repair since March 23 and recommended starting with Codex until the limits improved.
- 2:17:31
Designing an Editable Billing-Page Warning
The work considered whether a warning banner should match the existing notification and where it should appear, then placed it on the billing page while accounting for possible future edits.
- 2:18:36
TalkStore Pivot After a Storefront-Domain Rejection
The speaker explained that TalkStore hit an external roadblock when Shopify rejected its storefront domain, which had been central to the app. The resulting pivot removed the storefront from the codebase and revised the app’s scope, mission, and vision, while the precise new focus remains unclear in the captions.
- 2:21:10
How TalkStore Helps Shopify Merchants
TalkStore is described as an app in the Shopify ecosystem that lets merchants ask an intelligent assistant to handle store tasks, such as improving products, images, descriptions, and collections. A broader idea for controlling the merchant’s website was not approved by Shopify, so the app retained a narrower scope.
- 2:24:05
A Command Center for Building Projects
A command center is presented as a way to organize project building and track the work, with a reference to a separate explanation covering what it is, why a builder may need one, and how one can be created.
- 2:25:50
Trading Review and Debugging for Speed
The speaker explains that he is speed-running the project and postponing detailed review and debugging until the end, while acknowledging that a more rigorous project would require checking and debugging each step as it is completed.
- 2:27:33
Usage Limits in AI Coding Workflows
A discussion of relying on coding tools such as Codex and how usage limits can interrupt the workflow. Switching among multiple tools or models is raised as a possible workaround, with the qualification that it can become expensive; private account, plan, and consumption details are omitted.
- 2:32:13
Separating Engineering, Polish, and Submission Work
A discussion of structuring the app’s final phase by separating backend work from polish, QA, documentation, and submission mechanics, while recognizing that debugging and UX refinement may require substantial time.
- 2:37:20
A Practical Codex–Claude Coding Comparison
The speaker argues that Codex and Claude differ in philosophy and infrastructure, with each stronger in different areas. Based on his experience at the time, he recommended using Codex temporarily despite generally preferring Claude.
- 2:39:25
Designing an On-Demand Debugger Agent
A design for an on-demand debugger agent that systematically reviews board tasks, connects user feedback to the relevant task, and uses browser navigation, screenshots, and console inspection to investigate behavior and UI issues. It would also provide clear steps for manually reviewing or testing a selected task.
- 2:49:51
A Full Manual Review Before App Submission
A proposed final-review milestone covering every section of the application before submission to Shopify, with the project owner following the checks manually instead of using the normal prepare-and-start task workflow.
- 2:51:55
Rapid AI Tool Releases and Usage Pressure
He discusses Anthropic’s rapid stream of releases and users moving between tools, while expressing concern about how demand may affect token limits and workflow continuity. His practical priority is understanding how to keep working amid those changes.
- 2:56:14
A Dedicated Review Protocol for Debugging Sessions
The captions outline a debugger-specific workflow rather than one general checklist: start a review session for a defined context, run automated checks, then guide the user through manual steps. Billing appears as an example context, not evidence that the protocol was implemented or validated.
- 2:57:38
Reviewing Progress and Milestone Task Structure
After finishing a set of tasks, he reviews application progress and how work is distributed across milestones, expressing dissatisfaction with too many tasks being grouped into one milestone. He says the displayed progress percentage understates his estimate because the plan includes additional stages.
- 3:00:18
A Hypothesis About API and Subscription Token Efficiency
He raises an unverified possibility that a token-consumption improvement may have reached the API before subscription accounts, and suggests comparing a bounded workflow to investigate. He explicitly qualifies his API experience, and the page does not show a completed test result.
- 3:02:54
Can Configuration Optimization Fix Rapid Usage?
The speaker says he optimized his setup and instruction files but still observed faster usage with Claude than in his experience with Codex. He presents this as a repeated personal observation rather than a controlled test, suggesting the issue may not be limited to the instruction files or degree of optimization.
- 3:05:34
A Command Center for Preconfigured Task Agents
The speaker describes a setup where commands such as starting or preparing a task deploy multiple agents, with different roles or models preconfigured for particular kinds of work. He says that even after this optimization, managing the workflow remains difficult.
- 3:07:27
Switching Model Providers and Evaluating Cursor
The speaker suggests distributing work between Anthropic and OpenAI instead of relying on one provider, viewing access to multiple models as practical. He then examines Cursor pricing and discusses a release that was recent at the time of recording, while remaining unsure about local availability and relying only on others’ reports when considering its potential value.
- 3:09:34
Estimating Workflow Cost Per Task
The speaker converts observed usage into a rough per-task cost estimate and concludes that the pattern would not be practical across a large project. He also informally compares the amount of work he obtained from Codex and Claude, presenting it as a personal impression rather than a controlled benchmark.
- 3:14:25
Reflecting completed work in clear task-board stages
He examines why completed work is not reflected in the task-board swimlane and proposes a logical milestone or node tied to a Kanban task. He also asks for a pre-submission review protocol to be checked and connected with the completed-task summary and related workflow stages.
Got a question about something in this stream?
Ask in the group