Inside the Core Engine
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 centers on an in-development Shopify app intended to improve stores’ visibility on AI platforms, with a capability audit exposing gaps between the published plan matrix and actual implementation—especially APIs, webhooks, and multi-store agency support. Luqman demonstrates a review workflow combining specialist agents for security, compliance, and UX with hands-on testing, human-controlled merges, and automated deployment; mobile inspection revealed visual defects, confirming that the product was not finished. The discussion also covers free-versus-paid report design, A/B testing the placement of email capture, and an audit-led acquisition funnel whose channels should be measured before ad spending. Audience questions explore Claude, Codex, agent skills, and conversation-context compaction, with advice to master reliable tools instead of chasing every new model. On GEO, the speaker distinguishes it from SEO while leaving the effects of AI-generated articles, backlinks, and llms.txt unresolved because the evidence is not conclusive. The practical conclusion is to align public claims with real capabilities, understand integrations, and test the complete experience before Shopify submission; the completion date, best conversion flow, and need for external monitoring remain open.
109 moments from the stream
- 2:22:37
Keeping Pull-Request Merges Under Human Control
A workflow where coding tools assist with the work, while the project owner personally merges the pull request and pushes the changes to reduce the risk of problems.
- 2:25:04
Aligning Currency for Meta Purchase Event Values
A practical question about sending a purchase-event value in a currency different from the Meta dashboard currency, followed by a review of the current behavior indicating internal conversion with an exchange rate before accepting the approach.
- 2:26:45
Resisting Distraction with a Clear Project Vision
The speaker argues that a clear vision, a stable goal, and a timeline help sustain focus on one project instead of chasing every new model, video, or tool.
- 2:28:58
A Small PR Before Reviewing Phase Two
A request to complete a small pull request first, merge and push it to the main branch, and then provide a concise rundown of what Phase Two contains.
- 2:30:19
Turning Broad Knowledge into a Focused Toolset
Advice to a broadly knowledgeable developer to move from exploring many models and tools toward selecting and mastering a smaller set that supports their path. Praise for another tool is explicitly presented as hearsay rather than verified experience.
- 2:32:16
Keeping a coding task from turning into spaghetti code
A practical reflection on how prolonged coding can blur a task’s scope and produce tangled changes, followed by completing the work through a commit-and-push workflow.
- 2:34:06
How competition benefits coding-tool users
An opinion that competition between coding tools can encourage improvements for users. Codex and Claude are cited as an example, with private account and usage-limit details excluded.
- 2:34:56
Designing meaningful enterprise capabilities
An exploratory discussion of understanding the value of APIs, webhooks, and automation before offering an enterprise tier, including how such capabilities could support organizations managing multiple stores. The ideas still require deeper understanding and validation and are not presented as finished features.
- 2:36:53
Benchmarking a first release against competitors
The discussion uses competitor research to set expectations for a first release: it should be credible against existing products at launch, while a later version can go deeper. Competitors are framed as useful benchmarks, whereas having none can be a double-edged situation.
- 2:40:50
Clarifying responsibility for currency conversion
While reviewing a merged change, a technical question arises about converting an amount from USD to CAD with a different value, despite an earlier assumption that the receiving platform would convert it automatically. The excerpt raises the issue but does not establish a resolution.
- 2:41:52
Gemini added to Antigravity
A brief note says Gemini was officially added to Antigravity, followed by an attempt to locate or apply the update. The excerpt does not clarify the feature details or version.
- 2:43:11
Testing a Model’s Knowledge Freshness Without Web Search
The speaker described a test in which someone asked a model about the latest iPhone without web search; the response dated its knowledge to May 2026. He suggested fresher knowledge could reduce the need for searching, while also saying his overall trial was not the best. The model’s identity is unclear from the captions.
- 2:44:34
Learning GEO as an Emerging Field Adjacent to SEO
The speaker said he is still learning GEO and does not consider himself an expert. In his view, it is a young field where strong SEO experience may provide an advantage; this was presented as his opinion rather than an established finding.
- 2:46:32
Finding AI-Built Application Errors Through UX Testing
The speaker explained that small inconsistencies are easier to catch in domains he understands, raising concern about similar errors in unfamiliar areas. His practical approach is to exercise the application and inspect its UX when direct code review is not feasible for him.
- 2:48:00
Specialized Agents Versus Model Fine-Tuning
Answering an audience question, the speaker distinguished configuring agents for a narrow task and reasoning approach—such as auditing, exploration, security, or UX—from fine-tuning an open-source model itself. He said he had implemented specialization at the agent level inside his system but had not personally fine-tuned a standalone model, while expressing interest in local open-source models.
- 2:52:38
Future Ideas for Agent, Marketing, and Customer-Service Automation
The speaker outlines future experiments involving agent-based automation, marketing skills, marketing and customer-service automation, and other models and tools. These are presented as exploration plans rather than tested results, with technical choices guided by the needs of the product being built.
- 2:54:59
Auditing a Capability Matrix Before Hands-On Product Testing
The speaker describes a workflow that starts with a deep audit of a product capability matrix, turns identified gaps into implementation and debugging milestones, and then uses agents to cross-reference the matrix against the application. He stresses that automated verification should be followed by hands-on use and a clear understanding of interfaces and integrations before presenting the product to customers.
- 2:57:55
An Impression of the Claude Extension Versus Playwright
The speaker shares a non-benchmark impression that direct browser control through a Claude extension may be faster and more suitable than Playwright in this context. The comparison is based on observed use and the current setup, not a controlled performance test.
- 2:58:26
When to Defer a Feature and When It Justifies Delaying Launch
The speaker considers marking a feature as “coming soon” or moving it to a later version, versus completing it in the first release when it offers meaningful differentiation. He frames the tradeoff between entering the market quickly and launching a coherent, valuable product that reduces the risk of poor early impressions, while remaining willing to defer the feature if its scope grows too far.
- 3:01:52
Integrity of Shopify App Reviews
The speaker believes Shopify has a strong system for limiting fake reviews, while cautiously suggesting that some people may still try to game it through affiliated stores. This is presented as his opinion and speculation, not proof of specific conduct.
Got a question about something in this stream?
Ask in the group