Building AI Apps Live from Cairo 🔴
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 an evolving project command-center concept that connects a swimlane-and-milestone map with a task board executable by AI agents. Luqman discusses giving each task sufficient context, files, dependencies, and verification criteria, potentially assembling a dynamic prompt after a fresh codebase scan, while experimentally visualizing agent roles, collaboration, and activity. He stresses that the interface, the relationship among the Agent Hub, task board, and swimlanes, task and timeline accuracy, and the mechanics of shared memory and communication still require verification and iteration; this is not a completed result. Other useful discussions cover a research-based checklist for Shopify app submission, balancing Arabic and English, the unit economics of AI services in Egypt as the speaker’s opinion, and audience questions about Claude Code agents and cybersecurity. CTFs, deep training, and judicious AI use are suggested as practical starting points for security learning. The speaker also advocates learning AI app development through direct building and mentions deferred weekly and daily planning that was not completed in the recording.
25 moments from the stream
- 1:16:50
Why Building the Command Center Requires Iteration
The speaker expects that the result will not be completed in one attempt, comparing it with the command center build, which he estimated during the stream took about six prompts. He therefore anticipates additional rounds of review and adjustment before it is finished.
- 1:19:23
Connecting the Task Board to Swimlanes and Verifying Tasks
The speaker argues that the swimlane-based task board is more important than the Agent Hub interface itself. He questions how subtasks were generated from the project manifesto and stresses carefully verifying that the resulting tasks are correct rather than accepting them automatically.
- 1:21:50
Connecting the Task Board to the Swimlane and Milestones
A discussion of the task-board logic connected to the workflow: milestones with subtasks appear on the board, the swimlane conveys where the work is headed, and the task board shows what is being worked on. It covers work spanning multiple milestones, the need for a reliable week-based timeline, and reviewing subtasks with multiple AI tools because much of the board depends on their structure. The workflow connection is also noted as one-way, though the speaker says this is not very important.
- 1:31:16
Turning a Task Board into an Agent Command Center
The speaker explores a still-in-brainstorming design for running tasks directly from a board, with cards moving through ready, in-progress, review, and done states. He stresses that assignment alone is insufficient: execution needs assembled layers of project context, relevant files, dependencies, and system instructions, while responsibility for prompt generation still needs further thought. He also compares the idea with a simpler earlier prototype, without claiming that the new system has been implemented or tested.
- 1:40:50
Dynamic Prompts from Task Context and a Fresh Codebase Scan
The speaker explores preparing each subtask at execution time instead of writing many prompts in advance that may become outdated. The proposed system gathers milestone context and relevant changes, then runs an explorer agent to read the task description and inspect the context and codebase before assembling the final prompt. He presents it as a powerful but complex early understanding, not a verified implementation.
Got a question about something in this stream?
Ask in the group