Building Unfair Advantages
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 the ongoing development of Burnish as a GEO application in the Shopify ecosystem. A central workflow uses Claude in Chrome to study competitor products and designs, then passes findings and review feedback to Codex for implementation and debugging, with human inspection still required. Luqman also discusses preparing the go-to-market foundation before spending on ads: analytics and attribution, landing-page, headline, and pricing experiments, lead qualification, and a unified metrics dashboard. He makes clear that many forecasts and experiments were not yet complete. Practical work exposes late-stage difficulties, including a mobile-performance regression after some changes, followed by a clearly measured improvement after a later deployment; whether that gain justified the design compromises remained unresolved. The work also covers deployment and Git-branch friction, pull-request checks, accessibility reviews, and Shopify requirements. Some deployments succeeded, while other technical questions remained open. Audience discussions broaden into multi-agent coordination and usage limits, learning SEO, GEO, and development, the growing importance of marketing and distribution as AI makes building easier, and choosing a familiar market. A substantial secondary thread covers content and audience strategy: batching short-form content or temporarily pausing it to focus on building, building in public, weighing Arabic against North American audiences, YouTube monetization and sponsorships, and livestreaming as both a focus aid and a possible distraction, alongside rest and burnout prevention. The stream documents an unfinished process of development, research, and market preparation rather than a completed launch, with several decisions still requiring measurement and review.
151 moments from the stream
- 0:02:57
Analyzing Competitor Designs with Claude and Chrome
The speaker explains that Claude’s Chrome integration can interact with the browser and may speed up competitor analysis. The proposed workflow focuses on examining each competitor’s design and checking interface technologies, such as distinguishing Polaris Web Components from other Polaris options.
- 0:08:41
An Unresolved Claude Issue in Incognito Mode
During the practical attempt, the speaker confirms that the Chrome extension is allowed in Incognito mode, but Claude does not detect its presence as expected. The issue remains unresolved in this excerpt.
- 0:11:09
Why Start a GEO Product in the Shopify Ecosystem?
The speaker explains his rationale for choosing Shopify as a focused entry point instead of competing immediately with broad GEO platforms: deliver real value to merchants, establish a position in a defined ecosystem, and consider an independent platform later if the approach succeeds. He emphasizes avoiding fragmentation across platforms while presenting the competitive opportunity as his assessment, not a settled market fact.
- 0:15:45
Using a Coding Agent to Study Competitors’ Methods and Design
The speaker describes using Claude to analyze several Shopify GEO competitors, collect the findings in a comparison document, and then examine a competitor’s design system. The described review covered visible methodologies and technologies such as GraphQL and indications about backend and database structure, while also exposing inconsistent practical limitations around the tool’s access to an incognito browser window.
- 0:18:53
Preparing Analytics and Attribution Before Expanding Experiments
The speaker outlines the work plan: prioritize the application design and get the backend running, while making slower progress on GTM and measurement. He mentions adding Google Analytics and PostHog, using UTM tags to identify session sources such as ads and organic traffic, and considering tests of different website versions as a future step rather than a completed experiment.
- 0:23:26
Turning Competitor UI Analysis into an Implementation Plan
A workflow uses a browser agent to inspect competitors’ interface components, capture screenshots, and document design details before turning the findings into an implementation plan for Codex. The initial output is not expected to be perfect and remains subject to later human review.
- 0:26:01
Testing Three Homepage Hero Variants
A discussion of preparing three header or homepage-hero variants and rotating them to see which converts better. The stated view is that testing should be ready early when ads are expected to bring traffic, with each click treated seriously.
- 0:29:18
Comparing Personal Judgment with AI Recommendations
The recommendations matched three options the speaker was already leaning toward, but he noted that much AI output is not understandable to him. The discussion supports comparing AI suggestions with personal judgment while retaining skepticism and review.
- 0:31:47
Batching Short-Form Content into a Publishing Pipeline
The speaker considers preparing about two weeks of short videos in advance so publishing can continue without constant daily production.
- 0:32:16
Self-Hosting for Data-Residency Constraints
An audience participant describes converting an old laptop into a server for Supabase because, according to the discussion, the application and user data could not be hosted outside Algeria.
- 0:32:43
Learning and Progress Without Guaranteed Success
An opinion that success cannot be guaranteed, while the work can still produce learning and forward progress even when it does not deliver an immediate result.
- 0:35:40
Submitting the Core Version While Continuing Development
A scope-control approach: finish the app’s core to an acceptable level, submit it to Shopify for review, and continue building improvements and supporting services while awaiting approval, then deploy the updates afterward.
- 0:36:54
Setting Usage Limits for AI Agents
Advice to monitor an autonomous agent and understand what it is doing because it may keep working for an extended period, while setting a usage cap to prevent uncontrolled spending.
- 0:39:15
Coordinating Multiple Agents with a Kanban Board
An account of coordinating many agents through a Kanban board, with agents investigating blocked tasks and attempting to resolve the cause before continuing. The speaker regarded Kanban as essential for organizing this style of work.
- 0:41:13
Coordinating Multiple Agents with Kanban Boards
A discussion of managing multiple projects without overlap, assigning a responsible agent to each project, and whether the system goes beyond a task board by understanding project context automatically without extra setup such as MCP or custom commands.
- 0:42:03
Reviewing Competitor Design with Screenshots
An agent navigates and captures screenshots from multiple sites, not to repeat competitor research, but to gather references for improving the application's design before continuing backend work.
- 0:43:07
Testing Multiple Landing-Page Variants with Ads
An explanation of running several landing-page variants through different ads and comparing their performance to identify the strongest version.
- 0:44:18
Monitoring Agent Conversations and Progress
An explanation of viewing conversations between a supervising agent and specialist agents, intervening when needed, tracking progress, and keeping work on task through a visual mission-control-style interface.
- 0:46:30
Dynamic Headers and Conversion-Rate Measurement
A description of an experiment that automatically shows different headers to website visitors, then collects conversion data to determine which header performs best and derive a useful signal from it.
- 0:49:08
Showing Task Dependencies Across Multiple Boards
A brief observation that the agent is examining work dependencies in an environment with multiple boards.
Got a question about something in this stream?
Ask in the group