New Project? New CC?
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 a vision for Mission Control as a unified layer for monitoring agents and sub-agents, routing work by goals and available capacity, and moving from small tasks toward longer, more autonomous builds. Luqman discusses preparing PRDs, ERDs, system design, infrastructure, and access in advance while keeping humans in the loop and adding code, security, and Playwright checks before merges; however, the design, integrations, and full automation remained exploratory, with no confirmed completion. Audience questions covered choosing tools such as Codex and Claude, context and usage management, learning Python versus AI engineering, and reverse engineering with its legal risks. Practical discussions also addressed product distribution and audience reach, email-deliverability testing, AEO for Shopify stores, interface improvement, and repository knowledge. A recurring attributed view is that focus, quality, distribution, and reviewing AI-produced work matter more than maximizing project or content volume, while several technical, legal, and experimental outcomes remained unresolved.
103 moments from the stream
- 1:35:38
Tools Used for Vibe Coding
In a brief answer about vibe-coding tools, the speaker says he uses several tools, including Codex and Claude.
- 1:35:46
Brief Burnish Update
At the recording date, the speaker said Burnish was available on Shopify but had not yet gained broad visibility, and that they were experimenting with different ways to attract traffic.
- 1:37:00
Reviewing Interface Element Proportions
After testing a skill, the speaker reviews visual changes that reduce some elements, but argues that the phone mockup has unrealistic proportions. In his opinion, the skill did not produce a major difference.
- 1:40:13
Consolidating Agent Work in One Platform
The speaker explains that a Mission Control tool is intended to track work and improve its speed and quality by displaying sub-agents and providing a central coordinator instead of requiring several separate interfaces. He prefers continuous, goal-driven work, while presenting the platform as a hoped-for benefit rather than a proven outcome.
- 1:43:41
Comparing an Agent Harness with a Mission Control Layer
The repository’s documentation and package guides—but not its code—were reviewed for overlap with the proposed concept. The attributed conclusion was that the harness sits closer to the working agent, while Mission Control is a higher layer that exposes sub-agents, planning and building states, and progress; the stated recommendation was therefore to continue the current plan rather than rebuild it on that repository.
- 1:47:06
Usage Tracking Inside an Agent Orchestrator
The proposed control room would track Claude and Codex consumption across their usage windows, including tokens, so the orchestrator can assign missions according to remaining capacity and reduce workflow interruptions.
- 1:49:59
Validating Orchestrator Wiring and Reviewing Pull Requests
The chief orchestrator was described as being tested with a set of questions to verify that it was wired correctly, alongside a request to include pull-request review and Playwright checks in the workflow. Auto-fix and merge were later requested, but the captions do not establish that they completed.
- 1:55:16
Routing Tasks by Each Model’s Capacity
In response to an audience question, the chief was described as checking the remaining usage for each service and routing work according to each model’s capacity instead of sending every task to one agent.
- 1:57:29
Requesting Feedback for macOS Software Testing
A software project was said to be available for download and testing, with the qualification that it had not yet been tried on macOS and user feedback was requested. The speaker agreed to help but had not yet identified a concrete test scenario.
- 1:59:42
Improving Email Warm-Up When Messages Reach Spam
The discussion suggested moving warm-up messages from spam to the inbox and interacting with them to help build trust with the email provider, with a proposal to conduct the warm-up among a trusted group of acquaintances.
- 2:01:36
A Dashboard for Agent Monitoring and Continuous Work Allocation
The speaker outlines a platform vision that tracks sub-agents, time, usage, and project status while a central orchestrator allocates work using several factors to reduce stoppages. A shared workspace could also show agent activity and code changes, though the speaker says detailed diffs are not personally useful to him.
- 2:05:03
Removing Blockers and Moving from Prompt Engineering to Goal Engineering
The speaker argues that long-running agent work requires preparing genuine blockers in advance, such as required APIs and secrets. He distinguishes these from permission prompts that may be bypassed, while allowing the system to decide non-critical questions. He frames this as “goal engineering”: define a clear outcome, ask the agent to drive it to completion, and review its decisions afterward, while still reserving important questions for the user.
- 2:09:37
Sharing a Project-Building Journey Transparently
The speaker says he shares his project-building journey transparently through live streams, including setbacks, mistakes, lessons, and successes.
- 2:10:05
Balancing AI Work Review with Shipping Speed
The discussion presents the view that value no longer lies in execution alone, but also in maintaining focus and reading and reviewing AI-produced work. The speaker adds that deep review and study may slow shipping compared with rapid build-and-publish cycles, while learning programming still makes sense—especially for someone who enjoys it and wants to master it.
- 2:12:33
A Practical View of Agent Skills
The speaker says he uses only a small set of skills, including Grill Me, Impeccable, and Superpowers, and does not see a large skill collection as inherently necessary. His qualified advice is to use a skill when it is substantial and useful without artificially constraining the model’s behavior.
- 2:13:58
Automatic Context Compaction in Long Codex Chats
Responding to a question about agreements being forgotten in long chats, the speaker explains that Codex automatically compacts context to retain a summary of most session events. He does not claim this definitively explains the reported case, and says he has not personally encountered that kind of forgetting or identified a specific skill that solves it.
- 2:15:18
Improving Memory Through Repository Documentation
Advice to organize internal documentation inside the code repository, which may strengthen usable context or memory, with Obsidian mentioned as another option for organizing knowledge and memory.
- 2:17:55
Should You Start with Python or AI Engineering?
Responding to a learning question, the speaker recommends building projects and learning through practice. In his view, Python suits people who want deeper software-engineering foundations, while AI engineering can be a more direct applied path; he also offers the personal prediction that companies will increasingly need this specialty. He also notes that AI engineering was not yet precisely defined.
- 2:19:13
Managing Business Workflows with Automation Graphs and AI Agents
A general explanation of an automation-graph platform intended to manage mature projects across areas such as customer service, debugging, and marketing. It models multiple roles as AI agents or “employees” that collaborate toward a defined goal, with the qualification that the platform is still being built.
- 2:21:44
What Does It Mean to Be an AI Engineer?
The speaker explained that “AI engineer” can mean different things across companies: it may refer to building and deploying substantial products with AI, or it may require deeper knowledge of how AI works internally. He recommended demonstrating practical ability through a strong portfolio of AI-assisted work.
Got a question about something in this stream?
Ask in the group