Fixing Broken Code
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
This portion centers on developing a Shopify app and preparing it for submission, with a brand “manifesto” presented as an optional profile that can guide content personalization across the product. The discussion covers questionnaire design, terminology and approval actions, manifesto visibility, and locked research states, while pushing back on feature creep in favor of finishing backend wiring, debugging, mobile review, and compliance work. The build and final testing were not completed: upload and draft actions were tested without a documented final result, while unexpected localization behavior and a batch-update bug were identified. The speaker stresses human visual QA after each milestone because automation alone had allowed bugs to accumulate. This portion also explores coding-agent workflows, including session handoffs, multi-agent orchestration, unresolved sequential-versus-parallel tradeoffs, environment contention and wait time, and why large refactors should not be assumed to finish autonomously. Audience questions prompt cautious opinions on B2B versus B2C pricing, acquisition and conversion economics, two-sided marketplace growth, advertising-platform feasibility, and researching profitability before building. Additional discussion touches on marketing automation, learning through practice, agent-tool choices, and trying different AI models, while keeping economic claims, model-capability inferences, and interpretations of AI-policy reports explicitly uncertain or opinion-based.
65 moments from the stream
- 2:15:35
Using AI to Research Acquisition Economics
The speaker recommends using a capable AI model for research, citing OpenAI, Gemini, and Claude as examples. He emphasizes defining the objective, understanding the profit margin, and asking the right questions.
- 2:21:16
What a Coding Agent Can Complete in a Large Refactor
The speaker expects the agent to complete only a substantial portion of a full-page refactor, leaving wiring, polish, and backend work for manual completion. A prior similar task reportedly remained unfinished despite consuming usage limits, underscoring that large automated tasks should not be assumed to finish end to end.
- 2:22:36
Assessing a Two-Sided Marketplace’s Profitability Before Building
Responding to an app idea that depends on barbers and customers, the speaker recommends understanding its profitability structure and margins before building. He then checks his understanding of the proposed model: first attract many customers for free, then sell the app to barbers, while noting that the app has little value without barbers.
- 2:24:08
When an Agent Session Blocks Another Debugging Task
The speaker explains that he cannot continue another debugging task while the agent session is running: the debugging work requires starting a server, while the current session is using a server and database. This illustrates a practical constraint when coordinating concurrent development work.
- 2:31:47
Improving the Pace of Short-Form Storytelling
Feedback says the concept, analogy, and visual direction are strong, but the story moves too slowly for short-form viewing; the suggestions are to shorten it, tighten the pacing, and account for the mobile experience.
- 2:37:07
Avoiding Overcorrection in an AI Coding Workflow
A discussion of a coding session that chose sequential execution, while broader debugging waits until the database, server, and installation can be reset and the experience tested from the beginning. The speaker prefers letting the agent follow its inferred plan when the direction remains acceptable instead of overcorrecting it.
- 2:40:53
Sequential vs. Parallel Agent Execution
A general comparison of running agent tasks sequentially or in parallel, highlighting a possible tradeoff between token use and speed. The recommendation is attributed to someone the speaker watched, while the referenced model name is unclear in the captions.
- 2:41:22
Balancing Agent Output Quality and Execution Speed
A suggestion to run work sequentially, or deploy multiple agents when necessary, to improve output quality while recognizing that execution may become slower.
- 2:44:32
Managing a Blocking Dependency in a Coding Task
A coding task cannot advance to its next phase until the current work finishes because it precedes a server and database reset, making tool wait time a significant workflow constraint even in a faster mode.
- 2:46:34
A Two-Sided Marketplace Growth Strategy
The speaker proposes a two-phase marketplace launch: first build a user base through free usage, then offer businesses value through access to that audience and improved visibility. He recommends estimating user-acquisition costs before advertising and consulting local metrics such as cost per install, CPM, and cost per click, while stressing that the figures discussed were rough guesses requiring validation.
- 2:50:06
Handling a Coding Execution Mode That Is Not Working
A practical note that a particular execution mode did not behave as expected even though normal processing continued, leading to a temporary decision to keep the task sequential and await completion while considering the next steps.
- 2:52:26
Planning Before Fanning Work Out to Agents
The speaker describes giving the tool an implementation plan rather than a requirements document, then observing it perform sequential preparation before fanning the task out to multiple agents. He presents this as an inference from the tool’s behavior, not a confirmed account of its internals.
- 2:57:50
How an Orchestrator Manages Multiple Coding Sessions
The speaker demonstrates a workflow in which an orchestrator receives documents and handoff context from an earlier session. It prepares the task, assigns each agent a scope, responsibility, and goal, then waits for their reports, giving every agent an independent context window instead of requiring manual coordination across sessions.
- 3:06:20
When Warp Fits an Agentic Coding Workflow
Answering an audience question, the speaker says Warp remains a strong option for people who want to use different models or agents and multitask. In his view, desktop agent applications have reduced the need to coordinate parallel work manually; he recommends using an agent’s native interface when available and Warp when a model lacks a suitable platform.
- 3:17:42
A Marketing Strategy Without Outbound Outreach
A review of an app strategy that drops paid outbound outreach in favor of brand consistency and email-marketing automation, with programmatic SEO as a possible later addition. The discussion stresses that brand building needs dedicated planning.
- 3:20:16
Testing Income Claims Against Time and Effort
Advice to evaluate an opportunity by calculating the work needed to reach a hypothetical daily or monthly return, using published information and AI tools to estimate hours and requirements. The central caution is that unusually easy-looking income claims may conceal costs or conditions.
- 3:23:30
Assessing a Niche Advertising Platform
A discussion of both the appeal and difficulty of the advertising-platform model: a niche platform depends on having a suitable audience or network and matching advertisers’ target market. It may require substantial resources, though the speaker explicitly treats that estimate as uncertain and recommends researching the requirements first.
- 3:27:57
Avoiding Distraction Across Multiple Projects
A comparison between building several apps, which can broaden experience, and concentrating on one app until it proves its value. The expressed preference is to focus after studying the market and profit structure, while allowing a project to be abandoned if sufficient time and effort reveal poor prospects.
- 3:36:00
Human Visual QA After Each Development Milestone
The speaker explains that running automated milestones for weeks allowed bugs to accumulate and become entangled. Audit agents could catch some issues, but could not replace a person visually inspecting the application after every milestone; this lesson would inform a future refactor of the workflow system.
- 3:39:04
Reshaping a Go-to-Market Strategy Around Automation
The speaker discusses reducing manual launch operations through paid-ad automation, brand and review monitoring, email marketing automation, and occasional programmatic SEO. The proposed GTM structure is simplified, with larger paid-ad activity deferred until after the application launches.
Got a question about something in this stream?
Ask in the group