FIXED 1 BUG. CREATED 3 MORE.
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
A wide-ranging working stream focused on reviewing a Shopify app before launch, combining chat and merchant-UX refinement, fault diagnosis, and conversion of review findings into a debugging and testing queue inside a custom command center. Practical work covered crashes caused by products without images, disappearing text or attachments, slow replies, alignment problems, contextual actions, and confirmation states. A recurring conclusion was that a local fix is not proof of system readiness: human review and realistic scenario testing remain necessary. Luqman also explored agent-driven debugging through backend tests and log inspection, context management through documentation, state files, and fresh sessions, plus confirmation gates and Git checkpoints to constrain autonomous changes. Audience questions led to substantial discussion of Claude, Codex, OpenCode, OpenRouter, local models, and agentic coding tools. The practical view was that model choice depends on the task, quality, cost, and familiarity with the project, while benchmark tables must be interpreted through their methods rather than a single ranking. Many multi-model and cheaper-routing ideas remained proposed experiments rather than completed findings. The stream also covered validating product ideas with landing pages, taking products to market instead of repeatedly building, and moving from a useful first release toward scaling based on real usage. Visual brainstorming and interface design, SEO and GEO, precise product descriptions, and an Arab knowledge-sharing community were additional useful threads. Open questions remained around some latency and timeout causes, classifier-context cost trade-offs, automation reliability, and several fixes and tool trials that were not conclusively validated.
155 moments from the stream
- 0:01:24
Turning Debugging Into a Trackable Workflow
He explains that an app looking complete does not mean its backend is ready, and fixing one issue can reveal several more. The earlier build workflow therefore needed a separate review process covering UI polish, UX flows, and backend bugs, supported by command-center pages that show completed, active, partial, and unstarted work.
- 0:03:45
Planning Releases From Launch to Scale
He outlines a staged roadmap: define the launch scope, add the next version’s features, and postpone scaling until there is real usage. Later decisions would be informed by user reports, recurring issue patterns, and feature requests.
- 0:05:11
Estimating Submission Timing and Launch Readiness
Speaking on April 13, 2026, he considers April 17 probably too early and presents April 20, 2026 as a possibility rather than a firm commitment. Submission depends on completing UI, UX, and bug reviews and preparing suitable screenshots or a marketing video, after which Shopify’s review could result in acceptance or rejection.
- 0:08:04
Validating the Market Before Building
He introduces validating market demand before building an app as a practical experiment worth exploring. He starts looking for an example apparently involving multiple landing pages, but its name and method are unclear, and the page ends before the explanation is completed.
- 0:11:06
Testing a Product Idea Before Building It
He explains a proposed approach: create simple landing pages for different business ideas, promote them, and compare the cost of generating interest or sign-ups before investing time in building the strongest candidate.
- 0:12:55
Planning Multi-Model Experiments with OpenCode and OpenRouter
He describes a setup that gives OpenCode access to multiple models through OpenRouter and connects it to his command center through an MCP server. The stated plan is to try paid and free models and compare their time, cost, and reasoning quality with Claude and Codex, while acknowledging that extensive use of stronger models may bias his expectations of cheaper alternatives.
- 0:17:50
Strong Models for Building, Cheaper Routing for Runtime
He offers an architectural opinion: use one or two strong, trusted models to establish an application's foundation, then route user tasks to different models according to the task to reduce backend costs and improve monetization. He presents this as a strategy to test, not a completed experimental result.
- 0:19:29
Starting the Fix Agenda Review
He begins reviewing the work agenda, noting critical and smaller issues and observing that diagnosing one problem often reveals additional work. The page ends before the specific issues are explained.
- 0:20:40
Evaluating AI Models with Benchmarks and Practical Use
Responding to an audience question, the speaker said Claude’s hallucination rate appeared higher but still considered it his strongest current option for build quality based on personal experience. He urged careful reading of benchmark methods and numbers instead of relying on broad claims, while acknowledging uncertainty about one chart’s interpretation. He also noted that other models may be faster and cheaper and expressed interest in trying multiple models through OpenCode when Claude cannot solve a problem or produces an unsatisfactory design.
- 0:26:17
Combining Local, Low-Cost, and Frontier Cloud Models
The speaker and audience discussed using free or lower-cost models for lighter tasks while accounting for what the local hardware can run. A practical proposal emerged: plan and execute tasks incrementally with a local model, then pass the result to a frontier cloud model for review or final refinement. This was presented as a workflow to explore, not as a proven result.
- 0:29:10
Learning from Audience Suggestions and Exploring OpenCode
The speaker said the audience discussion prompted him to revisit tools he already knew and that he would begin exploring OpenCode and some suggested models. The exchange shows how audience questions and suggestions expanded the experimentation path, without claiming that the later tests were completed.
- 0:30:19
Balancing AI Coding Model Quality and Cost
The speaker discusses a practical view that paid subscriptions may provide more capacity and fewer usage constraints, while local models and open tools can reduce costs. He prefers optimizing margins and expenses when a product begins to scale, while respecting the alternative approach of prioritizing cost efficiency from the outset.
- 0:31:39
Evaluating an AI Model on a Small Coding Task
After an audience discussion about local-model workflows and output quality, the speaker proposes testing a model on a small task that removes unwanted chat behavior. The evaluation criteria include context and token use, modification quality, and avoiding regressions; this page does not establish a confirmed final result.
- 0:36:09
Sharing AI Coding Experience in an Arab Community
The speaker and audience exchange experience with local models and programming backgrounds, alongside an invitation to a small community where people interested in vibe coding can help one another and share updates and workflows across Egypt and the Arab world.
- 0:38:19
AI-Assisted Building Does Not Replace Marketing
The speaker compares vibe coding to a powerful tool that can accelerate building a project, but stresses that it does not market the product automatically. In his view, some developers create strong projects yet stall because they do not know how to reach the market, leading them to abandon the work or move to another idea before completing the journey.
- 0:40:27
Cleaning Up Output Formatting
A brief practical pass on making output cleaner by adjusting its presentation and removing unwanted formatting elements; the product context and some captioned words remain unclear.
- 0:41:57
Plan First and Keep Debugging from Becoming a Rabbit Hole
The speaker describes configuring an agent to plan a task and wait for approval before starting. He also explains how newly discovered bugs can pull debugging away from the original issue, arguing for clear structure and separate conversations or tasks to preserve scope.
- 0:47:03
Specialized Debugging Agents with Human Review
The speaker discusses simplifying a multi-agent workflow by skipping a broad orchestration cycle and deploying a focused debugging agent when needed to reduce usage and move faster. In his view, multiple agents can improve output quality, but each completed session still requires opening the application, reviewing the changes, and testing them manually.
- 0:49:24
Subscriptions and APIs for Coding-Model Access
In response to a question about subscriptions versus APIs, the discussion covers subscriptions for primary work and more recent exploration of APIs, noting that connecting OpenCode through OpenRouter can provide access to a broader model selection. Personal account and plan details are omitted.
- 0:50:48
Agent Skills for Brainstorming and Design
A brief tour of available skills and plugins, including the speaker’s opinion that the Superpowers brainstorming skill was among the tools he used most, plus references to research and frontend-design skills.
Got a question about something in this stream?
Ask in the group