Update + Checkpoint
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 developing Burnish, a Shopify application intended to measure and improve merchants’ visibility in AI answers through content and automation. Practical work covered the dashboard, settings, post-install onboarding, V1 information architecture, and surfacing autopilot activity while keeping proposed changes subject to merchant review. The UI/UX phase was described as complete for this stage, but full debugging and validation remained ahead; proposed next steps included testing the journey from fresh data, defining an objective readiness rubric, and verifying features incrementally because defects become harder to diagnose as dependencies accumulate. In audience discussions, Luqman offered qualified views on Codex, Claude, Remotion, multi-model review, and coding benchmarks, stressing that agent scores and benchmarks do not replace human validation and that model and workflow choices should balance quality, cost, and usage constraints. The stream also explored finding product ideas in real everyday problems, combining SEO with GEO, positioning Burnish within Shopify, and designing dashboards around a clear next action. Questions about AI-crawler permissions, parts of the settings behavior, and the tradeoff between deep automation and fast responses remained unresolved or in need of further verification.
64 moments from the stream
- 1:04:06
Using a second model to find bugs
A discussion of trying another model for a second perspective, emphasizing that the primary challenge is finding bugs in the first place, not merely carrying out the debugging.
- 1:06:52
Forking an AI conversation by workstream
A proposal to fork the conversation for a focused settings review while keeping the original thread on the dashboard and carrying the crawler-access finding into the new branch.
- 1:09:43
When multiple model opinions help
An audience-question response supporting consultation of multiple models to improve quality, especially for high-need cases or a promising new model, while acknowledging that this is not always done in practice.
- 1:16:45
Timing a final review with another model
A suggestion to finish the frontend first, then run an additional review or debugging pass with another model once it becomes available, even after the main issues appear resolved.
- 1:23:04
Consistent Usage-Limit Enforcement in Scheduling UIs
A review of a competitor’s scheduling interface identifies a possible validation inconsistency: dragging blocks dates outside the allowance, while direct date entry may bypass it. The broader lesson is to enforce the same limits across every input path.
- 1:30:32
A Unified Autopilot Toggle
The discussion addresses ambiguity between separate storefront and blog autopilots, leaning toward one control that enables autopilot across the product while allowing individual parts to be disabled from their dedicated pages.
- 1:32:12
Simplifying V1 Settings Architecture
The discussion reviews ambiguous settings tabs, proposes removing the agency section from the V1 surface, and considers consolidating overlapping areas—especially the brand and manifesto tabs—after checking the rest of the settings structure.
- 1:37:36
Showing Autopilot Activity on the Dashboard
The idea is to make the dashboard feel active by surfacing autopilot activity and linking to its detail page. Progress depends on using the correct latest application and code version.
- 1:48:48
Refining the Working Panel and Scores
A Burnish interface review suggests showing the latest activities in the working panel. The readiness score is set, while the visibility score’s placement or state remains unclear in the captions.
- 1:51:09
Consolidating Items Under Brand
The speaker proposes combining two unclear profile-related items under the Brand section; the captions do not clearly identify the items.
- 1:52:50
Product Page Implementation Checkpoint
A brief implementation checkpoint says the Autopilot page is complete, with the dashboard and settings pages remaining next.
- 1:53:06
Opening of an AI Capability Discussion
An opinion-based comparison begins about reaching AGI sooner and the relative strength of some models, but the model names and the comparison are incomplete and unclear on this page.
- 1:53:23
Model Benchmarks Versus Practical Experience
The speaker discusses AI-model rankings across several benchmarks but questions parts of the comparison because they did not match his practical experience, while awaiting a later GPT release.
- 1:56:10
Execution Handoff and Hands-On Debugging
He adds planning notes, keeps account and pricing on separate pages, and prepares a plan file for a fresh session to carry through to completion. He also stresses that the complex debugging phase will require hands-on, section-by-section work.
- 2:04:06
An AI Session as an Application-Testing Orchestrator
The speaker defines an AI session that acts as an assistant and orchestrator: guiding him through key pages and the user journey, helping distinguish gaps from genuine bugs, recording findings, and assigning fixes to separate sessions. The workflow may reset development data when needed to test the first-time-user experience from scratch.
- 2:10:26
From UI Completion to Deployment and Fresh-State Testing
He outlines a proposed sequence after settings and dashboard work finishes: merge the pull request, deploy on Railway, then open an assistant session that explains and records problems while other sessions address them incrementally. Testing begins from a fresh data state to review the experience end to end.
- 2:13:32
Configuring an Assistant for Design and QA Review
He describes configuring an assistant to follow the work, take notes, and help identify bugs before fixes begin. He also discusses resetting the database to test as a new user, difficulty working across multiple pages in the Shopify environment, and server slowness compared with the expected production application.
- 2:17:49
Auditing an Outdated Website and Revising Pricing Strategy
He audits the website, identifying parts of the dashboard, pricing section, and listed capabilities that need updating while other sections can remain. He says revised pricing should allow an appropriate margin and be reconsidered after observing market response, with the broader website rebuild deferred.
- 2:22:41
Finish the Core Engine Before Optimizing the Frontend
He reflects that optimizing the scan interface too early was a mistake because changes to the underlying application can repeatedly force frontend changes. His lesson is to finish the application and core engine before investing heavily in frontend optimization.
- 2:25:57
Reading Coding-Model Benchmarks Cautiously
The speaker said he personally looks at SWE-bench when comparing coding-model releases, but did not assume that a newer version would necessarily be better. He also acknowledged uncertainty about some benchmark categories, so the comparison remained an expectation rather than a test he had performed.
Got a question about something in this stream?
Ask in the group