5 TASKS IN 7 HOURS
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 work on a Shopify app and a multi-agent “Command Center” workflow that divides exploration, implementation, and auditing while keeping the human responsible for setup and sensitive decisions. Practical work centered on connecting AI APIs, aligning environment variables across Railway and Supabase, configuring Sentry for production monitoring, and troubleshooting deployment, authentication, trial, approval, and cancellation flows. Some deployments and checks were reported complete, but issues including an HTTP 500 error and parts of the user flow remained unresolved, illustrating the limits of automation and the value of clear logs, staged audits, and CI/CD. Luqman also discussed choosing coding tools by task, retained context, and usage limits; his comparisons of Codex, Claude, and Kimi K2 were framed as limited workflow impressions rather than controlled benchmarks. Broader topics included designing defensible scores for stores’ visibility in AI answers, the unsettled GEO-versus-AEO terminology relative to SEO, and validating markets, pricing, distribution, and product economics before accelerating development. Research-oriented discussion extended to LLM interpretability and the lottery ticket hypothesis, as well as Luqman’s views on Chinese and US model ecosystems, reliance on NVIDIA chips, and the need for accessible Arabic models and local pricing. Audience discussions covered AI’s effect on developer roles, code comprehension and debugging, parallel milestones and dependencies, agent-rule context costs, productivity habits, and peer learning. This portion also included additional practical discussions about workstation and operating-system choices, stream latency, development practices before AI, and programmatic video creation with Remotion. Preparing Shopify submission media and polishing the Command Center for a possible GitHub release remained future plans, not completed outcomes.
127 moments from the stream
- 0:03:06
Prioritizing the Shopify App Over a Standalone Website
The speaker prioritizes having the app and a strong Shopify presence, while the standalone website could be developed later alongside the build.
- 0:03:23
An Autonomous Command Center with Human Orchestration and Staged Audits
The speaker describes a command center that works with some autonomy after substantial setup, while he remains responsible for orchestration and system design. To reduce defect accumulation, one audit agent reviews each task and another reviews each milestone, informed by an earlier build that ended with a heavy debugging burden.
- 0:06:11
Marketing as a Bottleneck After Product Development
The speaker argues that many developers can build strong products but struggle with marketing. He therefore treats the transition from building to entering the market as an important practical phase to address later.
- 0:07:28
Why Use Auditing Instead of Debugging Every Task?
Answering an audience question, the speaker explains that deeply debugging every task would consume substantial time and resources. His current approach checks essential requirements after each task, stops work for fundamental issues, and sweeps each completed milestone, while still retaining final debugging as a necessary step.
- 0:08:54
Adapting a Coding Workflow to Tool Usage Limits
The speaker discusses switching between Codex and Claude when usage limits interrupt the workflow and pacing work around token and rate limits, while expressing a personal preference for Claude’s performance on this task.
- 0:11:12
Comparing SEO, GEO, and AEO Through Search Trends
The speaker compares terminology for visibility in traditional search and AI-generated answers, using Google Trends to inspect market language. In his reading of a five-year chart, interest in the newer terms began rising before 2025, climbed sharply in September 2025, and resumed growth after a Q1 2026 dip. SEO remained far larger, while GEO and AEO rose together, so he did not claim certainty about which term would prevail; his personal expectation was that AEO might become the earlier or clearer description. He also notes that the GEO and AEO acronyms can mean other things, limiting the precision of the comparison.
- 0:16:38
Direct AI Provider APIs or OpenRouter?
The speaker discusses integrating four AI providers directly—OpenAI, Anthropic/Claude, Perplexity, and Google—instead of routing all of them through OpenRouter. Based on a coding agent’s recommendation and the existing codebase, he chose the native provider APIs as the cleaner approach for automated prompting, then worked through project, key, and environment-file setup; the captions expose no secret values.
- 0:21:34
Adapting a Coding Workflow to Tool Usage Limits
The speaker explains that usage limits and reset timing may temporarily prevent relying on Claude, so he plans to use Codex for most of the work. Private plan and credit details are omitted.
- 0:22:30
Careful Setup of API Keys and Deployment Environment Variables
Practical work on placing keys in an environment file and following careful steps before updating an application deployed through Railway. The speaker notes that some deployment variables must change and that many relate to the Supabase URL, while some spoken details remain unclear.
- 0:34:16
Understanding the paid-tier setting and the app test step
A question was raised about why the Gemini paid-tier configuration line was needed and where it should be placed. It was described as related to documentation checks, followed by a plan to test an APK after another step whose wording is unclear in the captions.
- 0:37:09
Diagnosing Gemini API rejection across billing and workspaces
The paid-tier warning disappeared, but API requests were still rejected for insufficient credit. Troubleshooting distinguished adding a payment method from funding prepaid credit, identified a possible key-to-account or workspace mismatch, and mentioned pay-as-you-go as a longer-term option without establishing that the fix was completed.
- 0:42:19
Verifying an API Key and Workspace
He traces a smoke-test problem after confirming that both keys are in the same default workspace, then suspects the test is not using the intended key and proposes verifying it in the console.
- 0:44:10
Benefits and Risks of GLM’s Error-Fixing Persistence
Answering a question, the speaker stresses that he is not a GLM expert and says the tool appears designed to keep working until a problem is resolved. In his view, this can help when given an error screenshot, but it may take time or attempt broad changes that affect the backend, so he recommends asking more experienced users about controlling it.
- 0:46:38
Configuring Gemini Billing and API Keys
The work returns to Gemini to clarify prepaid versus pay-as-you-go billing and the auto-reload option, while placing the correct API keys in environment variables and hosting settings. Some details of switching billing modes remain unresolved in the captions.
- 0:51:50
Working on a Side Task While Parallel Tasks Run
The speaker explains that he prefers using waiting time for side work, so he moved to a lower-priority website task while other tasks were running in parallel.
- 0:55:25
Quiet Mornings and Workout Timing
The speaker reflects on waking and working earlier to benefit from quiet mornings, then asks the audience whether they prefer going to the gym in the morning or at night.
- 0:57:26
When Will the Work Series End?
Answering an audience question, the speaker says the series may reach roughly 150 days but has no fixed end date. Passing 100 episodes was a milestone, while stopping remains tied to achieving his main goals.
- 0:59:21
Balancing Coding-Tool Cost and Usage Limits
The speaker views Anthropic and Codex as important workflow investments, but criticizes Anthropic's cost and the disruption caused by usage limits. He considers temporarily relying more on Codex if it stays fast, completes the work, and resets its limits in time, while noting that the unusually low usage he observed may have been caused by a bug the previous day.
- 1:02:21
An experiential comparison of Codex and Claude
The speaker says Codex works well for him, while Claude seems to have a distinctive reasoning style and workflow that he cannot precisely describe. He frames this as a personal, habit-dependent impression and notes that user-community feedback helps surface other experiences and tools such as Qwen 3.
- 1:08:06
Handling outdated website-navigation guidance from AI assistants
The speaker explains that AI-generated website-navigation instructions can become outdated as menus and interfaces change, so he uses screenshots to give the assistant more current visual context.
Got a question about something in this stream?
Ask in the group