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
- 4:17:34
Skills Learned from Trying Dropshipping During COVID
A brief reflection on trying dropshipping during the COVID period and learning practical skills in ad management, product discovery, and landing-page creation.
- 4:18:23
Why Not Expand the Classifier Context Window Indefinitely?
A question-and-answer discussion about widening a classifier’s context from one message to several exchanges. The explanation is that the classifier runs on every message, so a wider window increases token cost; a practical comparison of the alternatives was requested, but this excerpt provides no clear quantified conclusion.
- 4:19:30
Validating an App Market and Starting with Organic Marketing
Advice to an app developer to first test whether they can grow the product themselves and to begin with organic marketing when marketing resources are limited. The Shopify App Store is cited as an ecosystem that can bring traffic and also offers advertising.
- 4:23:03
Turning Work into a Task and Running the Workflow
A practical excerpt asks for work to be prepared as a task, started through the regular workflow, and then requests its completion. Some command labels and execution details are unclear in the captions, so the tool and outcome cannot be stated with certainty.
- 4:23:56
Try to Monetize Before Moving to the Next App
A practical observation that many developers build an app, avoid taking it to market, and move on to another. The advice is to try to monetize the product and test the market instead of only repeating the build cycle.
- 4:29:54
Balancing Work Volume and Quality When Choosing a Coding Tool
The speaker discusses a practical trade-off between enabling more work and pursuing higher quality, while relaying an audience opinion that favors the higher-quality option when circumstances allow. The choice is framed as needs-dependent rather than a definitive verdict on any specific tool.
- 4:30:59
How Should AI Model Benchmarks Be Read?
The speaker distinguishes subjective impressions of models from benchmarks that attempt a more objective comparison. He says the referenced site combines multiple sources and that results still depend on how they are interpreted, while acknowledging that he needs more time to understand the benchmarks well.
- 4:32:46
Balancing Stream Timing with Productive Hours
The speaker discusses the tension between streaming when the audience is active and working in the morning, when he feels more productive even though community presence may be lower. The broader point is balancing audience habits with one’s personal work rhythm.
- 4:33:49
Closing Open Debugging Tasks One by One
After noticing several tasks open at once, the speaker begins reviewing them sequentially. A loading screen is mentioned separately, followed by garbled labels that appear related to request or catalog behavior, so the captions do not establish the fixes precisely.
- 4:35:06
Fixing One Bug Can Reveal More Bugs
The speaker reviews a progress calendar and a crowded debugging list, summarizing a familiar development pattern: fixing one bug can expose several additional bugs or tasks. He therefore wants visible daily progress while the debugging backlog continues to grow.
- 4:37:03
A Markdown File for an Agent Command Center
Luqman described sharing a Markdown file that could be placed in a project and given to its agents to set up a similar command center, while noting that the result would not be identical and saying he planned to explain the process later in a video.
- 4:41:12
Connecting AI Response Actions to Data and Mapping a Funnel
Luqman investigated a possible code duplication and requested a step-by-step testing walkthrough. He reviewed a task where buttons beneath AI responses had been placeholders, described connecting handler-computed actions to actual product data, and requested an SVG node map of the funnel for further checking.
- 4:44:48
Preparing an Audience-Requested OpenCode Trial
Luqman responded to an audience request to try OpenCode, asked which model to use, and began preparing a new session for the trial. The page does not show a completed outcome, and the spoken model names are too unclear to identify confidently.
- 4:47:20
Evaluating a Coding Assistant with Varied Tasks
The presenter tried simple, design, and backend tasks. For the harder task, the assistant was asked to inspect the codebase, gather findings, and return with a plan without executing anything until authorized.
- 4:51:31
Fallback Plan for Classification Timeouts
The presenter discussed a classification path that had not yet been implemented and reviewed a proposal to raise the timeout to 15 seconds while adding a keyword-based fallback when classification fails. He preferred the proposed option, but the captions do not establish a final implementation result.
- 4:58:50
Diagnosing Slow Chat Replies by Comparing AI Model Analyses
A chat UI was taking several seconds to answer simple greetings, with multiple possible causes and a temporary server outage complicating the check. Two model-generated diagnoses were compared, and a third model was asked to judge only the supplied analyses. The evaluation noted factual errors in some claims and favored the analysis identifying that the conversation handler itself might be unnecessarily heavy for greetings.
- 5:09:22
Speeding Up an Agent for Simple Conversational Tasks
A proposal to make an agent faster on straightforward conversational openings by delegating investigation to subagents and potentially using other models. Some technical wording in the captions is unclear.
- 5:11:40
When Should an Agent Stop and Ask?
A practical assessment of an agent taking a long time to diagnose a server failure: slower work can be acceptable if it produces a fix, but remaining in a reasoning loop without a result or clarifying question wastes time and tokens. The speaker also discusses limited visibility into what the agent is doing.
- 5:16:42
Long-Running Autonomy Still Needs Cost Oversight
A discussion of agents said to work for hours and self-correct without human intervention. Enthusiasm for that autonomy is balanced by practical concern about the cost continuing to accumulate during execution.
- 5:19:20
Choosing a Coding Model Based on Project Familiarity
The speaker argues that results depend not only on model capability but also on the user's familiarity with it and its understanding of the codebase. He prefers a model he already knows while keeping alternatives available for suitable cases, and says an agent should stop and ask for guidance rather than force a fix that could break other backend behavior.
Got a question about something in this stream?
Ask in the group