TOFU Acquisition Strategy
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 documents wide-ranging practical work on Burnish, a GEO application for Shopify merchants intended to measure store and product visibility in AI answers and recommend catalog improvements. The session focused on rebuilding and simplifying onboarding, polishing the scan page, report-wait experience, and mobile UX, and reviewing plan selection, Shopify billing, and data-processing consent. In parallel, it developed a Meta/Facebook acquisition funnel built around a free report and signals such as scan started, scan completed, and lead, with a recurring view that the mid-funnel and user journey should be improved before ads run. The work also covered clean-state testing and troubleshooting 500 errors, scan failures, deployments, Turnstile, tracking events, Stripe redirection, and Supabase, Cloudflare, Railway, and GitHub integrations; several causes and outcomes remained unresolved, and end-to-end verification was not shown as complete. Performance testing found that reducing sampling depth did not deliver the expected speedup in one attempt, leading to discussion of progressive results and a clearer waiting experience. Audience questions and broader discussions covered home-hosted databases, building Shopify apps through vibe coding, learning programming with AI, GEO versus SEO and AEO, dropshippers as a possible audience, and personal comparisons of Codex, Claude, and agent workflows. The indexed topics also cover separate discussions of Shopify-app market economics and saturation, D2C and Egyptian market strategy, cash flow and AI-tool spending, internet and OBS troubleshooting, digital-access constraints, digital-nomad visas, and asynchronous communication. The supported conclusion is that clear value, low-friction onboarding, reliable funnel behavior, and sound measurement should precede advertising scale; final product readiness, several bug causes, and the best landing-page and consent structure remained open.
183 moments from the stream
- 3:06:21
Fixing Onboarding Bugs and Removing Questions the App Can Answer
An onboarding review notes that workflow and monitoring stages appear stuck or are automatically marked complete, while the continue-to-trial button is disabled. It also criticizes a step asking merchants for product category and priority information that the application should infer itself, proposing that bugs be fixed first and the unnecessary step removed.
- 3:08:48
Moving a Cold Audience Through the Acquisition Funnel
The discussion argues that a cold audience should be warmed before being asked to qualify: the path moves from problem awareness and discovering the product, to understanding how it works, and then to qualification and closing. The stated view is that the funnel should be improved before ad spend begins.
- 3:09:29
Evaluating Ads by Expected Customer Value, Not Lead Cost Alone
The segment discusses Facebook ads and a claimed Claude integration for campaign management. It explains that results should not be judged only by lead cost, but also by pricing, retention and churn benchmarks, conversion likelihood, and customer lifetime value, while explicitly noting that such projections contain substantial speculation.
- 3:11:46
Starting at the Middle of the Funnel Instead of Awareness
The speaker argues that ad results are not immediate and that trying to break even from day one can undermine growth. The proposed approach pairs investment with a strong CRM and funnel, initially optimizing for middle-of-funnel actions such as leads and requests rather than page views or brand awareness.
- 3:13:55
Handing End-to-End Funnel Verification to Codex
The speaker defines Codex's task as continuing with end-to-end funnel verification, a smoke test, and possible design adjustments, while postponing further pixel-firing optimization until the full flow is confirmed.
- 3:15:49
Verifying the Meta Domain During Funnel Review
The speaker notices that the Meta domain had not yet been verified, asks whether to proceed, and then confirms completion before checking whether any other Facebook-side setup is needed while end-to-end funnel verification continues.
- 3:16:21
Cash Flow Before Brand Awareness
In a brief discussion, the speakers agree with the view that awareness-building follows the establishment of cash flow, describing liquidity as a core business priority.
- 3:18:26
Simplifying Ad Tests on a Limited Budget
The speaker recommends avoiding the dilution of a limited budget across too many landing pages, tests, and ads. The plan is one landing page and one funnel supported by several creatives across a small set of themes and formats such as video, single images, and carousels; whether to produce them on-stream remains undecided.
- 3:23:42
Why Creative Angles Now Matter More Than Laser Targeting
The speakers discuss how mastering the buyer persona can reduce waste and make performance increasingly dependent on creative angles. In the speaker's view, Facebook has changed substantially since 2019: its AI-driven algorithm is now better at finding likely customers, favoring broad targeting while creatives become the main competitive lever.
- 3:27:03
Choosing a Market for Direct-to-Consumer Commerce
The speaker argues that the US may offer stronger conversion prospects for direct-to-consumer commerce, while Egypt presents logistics challenges. He says a strong supply-chain position could make an Egypt-focused business viable, but stresses that he lacks enough data to offer reliable expectations.
- 3:31:44
Structuring Change Verification and Pull Requests with Codex
While waiting for end-to-end workflow verification, the speaker reviews a change reported as implemented and pushed, then asks Codex to create a pull request with a title and description once work is ready to merge or deploy. He also requests proactive, explicit next-step guidance so he can begin verification.
- 3:39:17
Codex Limitations During Pull Requests and Verification
The speaker discusses a problem encountered when having Codex create GitHub pull requests and expresses uncertainty about Codex's access capabilities compared with Claude. He then returns to the scan page to verify the work, but it does not load by the end of the excerpt; the captions do not establish the cause.
- 3:41:55
Market Culture as a Competitive Edge in Egypt
The speaker argues that understanding local buyer habits can give startups an edge over international brands, citing cash on delivery and limited familiarity with subscriptions and automatic payments in Egypt. He presents thoughtful market education as part of that advantage.
- 3:48:16
Codex vs. Claude: Coding Benchmarks and Task Fit
The speaker argues that Claude is stronger for general coding, citing benchmark results he says were published, while allowing that Codex may excel in backend work, debugging, or complex codebases. This is his interpretation of benchmarks and experience, not a test conducted in these captions.
- 3:49:48
Proactivity and Tool Access in Coding Agents
The speaker describes difficulties with Codex’s proactivity and access to tools such as GitHub, deployment platforms, and command-line interfaces, contrasting them with a smoother experience using Claude. This remains an environment-specific personal assessment; the captions do not establish the cause or resolution.
- 3:58:12
How Context Compaction Affects Long Agent Tasks
The speaker explains that a smaller context window, in his assessment, can force an agent to invoke auto-compaction repeatedly during a large task. He describes this as handing a summary to a fresh context so work can continue, and argues that repeated compaction may slow execution and reduce session freshness compared with a larger context window.
- 4:01:20
GitHub Conflict and a Localhost Failure
During live coding work, two problems were observed: a GitHub conflict remained visible, and localhost on port 4321 would not open despite repeated refreshes. The speaker first asked for an explanation of the blocker, then used another assistant to diagnose it and draft corrective guidance; the page does not show the fix being applied or succeeding.
- 4:05:44
Comparing GPT and Claude for Research, Coding, and Proactivity
The speaker argues that GPT may be stronger for research and complex equation solving, while preferring Claude's coding interaction style because it seems more balanced between proactive and reactive behavior. He emphasizes wanting rational guidance without excessive proactivity.
- 4:12:22
Diagnosing Codex Sandbox and Permissions
The speaker discussed a permission that had previously worked in Codex, along with the need to relaunch it after changing its policy configuration. He suspected the issue might be Codex failing to attempt available external or cloud capabilities, citing its initial reluctance to run SQL through a Supabase integration, and noted that the bind check only curled localhost without starting a listening service.
- 4:22:01
Testing the Frontend and Backend After Deployment
After merging and deploying, the speaker tested an AI-report generation flow and encountered an error that persisted outside Codex. He said the interface still needed substantial design refinement, while the immediate priority was backend testing and debugging one issue at a time.
Got a question about something in this stream?
Ask in the group