RIP. NEW MISSION. CC V2.0
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 redesigning a largely autonomous “Command Center” that routes coding tasks by size and dependency, delegates research, exploration, and implementation to agents, and introduces task- and milestone-level audits before human review. Luqman demonstrates unfinished practical work across Codex, Claude, and terminal workflows, weighing output quality against token consumption and usage limits while encountering real friction involving Supabase, Docker, MCP, and stalled tasks; the workflow is therefore not presented as a completed success. The discussion broadens to unit testing, lightweight feature specifications, competitor research before building, and the need to market products rather than stopping after implementation. He also outlines a Shopify tool concept for measuring and improving product visibility in AI answers through GEO, which he described at the time of the 2026 recording as an emerging field and forecast to grow, particularly toward late 2026, while leaving market viability, implementation complexity, and the GEO–AEO distinction unresolved. Drawing on experience that began with dropshipping in 2017, Luqman compares dropshipping with DTC and white labeling, e-commerce with app economics, and B2B with experimental B2C directions; he also offers opinions on SaaS margins and recurring revenue, reasons for choosing the Shopify ecosystem, and app marketing through defined audiences, acquisition channels, and a mix of organic and paid approaches, without presenting these views as guarantees or settled outcomes. Secondary discussions cover automated short clips and the risk of lost context, recorded streams as a work archive and community-learning channel, programmatic marketing videos with Remotion, thumbnail experiments, and copyright constraints. Audience questions add useful discussions about learning software development through practical AI-assisted projects, verifying the security and provenance of agent skills, estimating time, managing dependencies, model pricing and compute efficiency, and scheduling work around session limits. The qualified conclusion is that automation may reduce manual coordination but does not remove the need for oversight, debugging, and validation; Luqman argues that strong planning and context matter more than minimizing token use alone.
155 moments from the stream
- 1:06:32
Dispatching Agent Tasks Remotely
The speaker described dispatch as a possible way to direct an agent from a phone to work on a particular milestone or task, presenting it as a remote-control option before moving his work to Codex.
- 1:08:20
Simplifying the Workspace to One Terminal with Codex
The speaker said task-board automation let him work through a single terminal, leading him to consider a desktop application and Codex for a simpler setup. He remained comfortable with his existing workflow and framed the Codex trial mainly as an attempt to handle usage constraints better, not as proof that one tool was superior.
- 1:10:32
Making Workflow Commands Compatible with Codex and Publishing Them on GitHub
A discussion of custom commands that work in another environment but not yet in Codex, with a request to investigate the incompatibility. The speaker says the command-center tool has evolved into a more autonomous workflow and considers publishing it on GitHub instead of only sharing an instruction file for others to rebuild.
- 1:13:18
Batching Small Tasks While Respecting Blockers and Dependencies
A new Codex session is tried with a command that batches available small tasks from a to-do list. The explanation notes that blocked tasks are not run automatically; their prerequisite larger task must be completed first.
- 1:16:01
Balancing Coding-Model Usage and Cost
The speaker argues that heavy usage may make some options impractical and suggests considering Codex or lower-cost models through OpenRouter. This is general practical advice, not a disclosure of personal account usage or spending.
- 1:16:54
A Qualified View on Using Codex for Coding Work
The speaker cites opinions that Codex may outperform certain alternatives and argues that it remains a practical option capable of substantial coding work. This is presented as a personal, untested assessment rather than a verified benchmark.
- 1:18:14
A Limited Observation on GLM 5.1 Persistence and Expanding Changes
The speaker describes limited experience with GLM 5.1: when given an error screenshot, it keeps working toward a solution before returning. He considers that persistence a double-edged sword because it may solve the issue but can broaden the changes and affect other backend areas, whereas he generally prefers the smallest viable fix. He explicitly says his experience with the model is shallow.
- 1:20:23
Organizing Large Tasks with AI Agents
An explanation of an automated workflow that prepares large tasks, spawns agents for research and exploration, and then executes the work incrementally. The speaker says the task classification and plan were AI-generated, with quality assurance reserved for a later stage.
- 1:24:51
Balancing Token Efficiency and Output Quality
The speaker argues that reducing token use is not an end in itself: it may save tokens while weakening quality. He suggests checking whether agents, plugins, and skills genuinely contribute to result quality, while avoiding overlap and unnecessary agent deployment.
- 1:28:33
How Token and Rate Limits Affect Time Estimates
Answering a question about time estimation, the speaker explains that workflow speed is not determined by task-list size alone: token availability and rate or session limits can become the main bottlenecks. Consequently, estimates may change substantially when those constraints change.
- 1:33:33
From a Master Manifesto to an Actionable Roadmap
A workflow starts with a document describing the business mission, vision, domains, and future stages, then gives it to an AI model to develop a roadmap organized by phases, complexity, and labels. The roadmap becomes part of a command center containing tasks, milestones, and notes, with the qualification that large planning documents take time and may need the work divided.
- 1:37:11
Scheduling Automation with Polling and Session-Time Limits
The discussion explores running an automated task on a schedule, including repeated polling requests until a response allows the task to be retrieved and executed. It also considers trying a scheduled command in the command center, while session timeout remains an unresolved constraint and no successful test is claimed.
- 1:38:49
Designing an Adaptable, Shareable Command Center
A command center does not require one fixed structure; it can be shaped around project needs and lessons from prior use. The speaker also states an intention to document the implementation and later make it available on GitHub for others, without claiming that publication has already happened.
- 1:41:46
Experience with Shopify’s Merchant Side and App Side
The speaker says he has worked on the Shopify merchant side since 2017 and that working from the app side in recent days is relatively new to him. Some wording around the comparison is unclear in the captions.
- 1:43:24
Updating the Command Center Workflow
The speaker describes a substantial Command Center update and a new workflow for it, with plans to make it public on GitHub later. He also mentions extensive changes to another section or project whose name is not fully clear in the captions.
- 1:44:19
An Incomplete Task Interacting with Supabase
The speaker notes that a coding task is taking time and attempting to access Supabase, and expects part of the Supabase setup to change; the captions do not establish the cause or a resolution.
- 1:48:11
Balancing Codex and Claude Within Usage Limits
While attempting to finish a coding task, the speaker decides to bring Claude in alongside Codex, then reflects that it makes little sense to complain about limits while leaving available capacity unused. He frames this as a workflow-management issue between tools, without establishing that the task was completed.
- 1:51:01
Comparing Codex and Claude Usage Fairly
The speaker described an informal attempt to compare usage limits and token consumption on tasks of similar complexity, while acknowledging that different tasks make the comparison approximate. In his experience Codex appeared better, but he noted that Claude completing a task on the first attempt could be more efficient than repeating it several times, and that the Supabase issue discussed might not represent typical work.
- 1:59:52
An Auditor Agent for Task and Milestone Review
The workflow moves tasks from execution to review and completion, with an auditor agent checking code against specifications and a criteria checklist. After a milestone is finished, another audit reviews it as a whole, allowing progression on a pass or returning it on a failure.
- 2:02:58
Switching Coding Assistants Around Usage Limits
The speaker argued that using two coding assistants can improve continuity: when one reaches its usage limit, work can move to the other, especially when their reset schedules differ. He presented this as a practical preference rather than a controlled test.
Got a question about something in this stream?
Ask in the group