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 explores Luqman’s concept for a visual command center that organizes AI-agent work through a timeline and milestone-driven task board, with planned OpenClaw integration and an Electron app that were not completed. In the practical workflow, he divides the initial skeleton into phased prompts across several terminals, separates parallel from sequential dependencies, reserves a later Codex polishing pass, and adjusts project-level permissions to reduce repeated confirmations. He also compares Antigravity’s embedded preview and direct UI-element selection with terminal agents, arguing that the former often suits frontend adjustments while multiple terminal sessions fit heavier and backend work. The session includes starting a development server, requesting visual feedback, noting alignment issues, and investigating Electron updates that did not appear; the requested evaluation and update fix remain unresolved in the checked records. It concludes with monitoring a Git-triggered deployment and opening the site to confirm the updated version appeared, without claiming that the broader command-center project was finished.
14 moments from the stream
- 0:01:46
From an Agent Timeline to a Visual Command Center
Luqman explains that the agent timeline/swimlane concept evolved into a visual command center for tracking work and agents. He expects it to require an Electron app with planned OpenClaw integration; these are design intentions, not completed implementation.
- 0:03:34
Building a Skeleton Through Parallel Work Phases
Luqman prepares the project file and multiple terminals to work through phases in parallel, with a later polishing phase assigned to Codex. The session goal is to establish a clean, understandable, functional skeleton first, then review the content and open questions once the system is clearer, while acknowledging it may not be completed that day.
- 0:12:04
Configuring project-level permissions for coding agents
A discussion of configuring project settings to reduce repeated confirmation requests for terminal commands, while keeping the global setting unchanged and applying a project-level override.
- 0:15:03
Inspecting the app structure and noting UI issues
An attempt to locate and inspect the app structure in the project folder, followed by two screen-level visual observations involving element placement and line alignment, while noting that the app launches. Some UI wording is unclear in the captions.
- 0:17:27
Splitting project implementation into parallel phases
A five-prompt workflow is outlined: establish the first prompt as the foundation, run several later phases in parallel, then perform a final polishing phase with Codex. The speaker also plans to restart agent sessions after updating permissions to avoid repeated confirmation prompts.
- 0:20:04
Visual Preview Inside the IDE
A brief comparison between letting Claude visit a website through Playwright MCP and Antigravity's ability to display the site inside the IDE, allowing the developer to expect to see changes while working.
- 0:21:59
Coordinating Phased Tasks Across Multiple Terminals
Several terminals are prepared for different work phases, with an additional terminal assigned to Codex, followed by a dependency check to distinguish tasks that can run together from those that must run sequentially.
- 0:24:46
Previewing a Project and Requesting Visual Feedback in Antigravity
The project is prepared in Antigravity, its development server is started with npm, and the site is opened in the embedded browser, where changes are expected—but not shown to be verified—to appear live. The IDE agent is then asked to inspect the site visually and provide feedback and a score; the page ends before any result appears.
- 0:32:59
Direct UI Element Selection vs. Terminal Agents
The speaker argues that selecting a webpage element and sending it directly into chat makes UX adjustments easier, while preferring terminal agents and multiple terminal sessions for heavier work, especially backend tasks.
- 0:36:33
Troubleshooting Updates That Do Not Appear in an Electron App
During the experiment, an update does not appear in the Electron app. The speaker notes uncertainty about where the change was applied and tries closing the window and reopening the content; the captions do not confirm a resolution.
- 0:39:10
Concept for an Agent Command Center
The speaker outlines a proposed large agent command center combining a swimlane-style project timeline, a board showing tasks occurring throughout the day, and connected agents that can communicate with one another. Live-editing behavior was still awaiting evaluation.
- 0:48:53
Comparing AI IDE UX and Parallel Work
The speaker says editors such as Antigravity, Cursor, Visual Studio, and VS Code offer a stronger interface experience, while another workflow may be better for moving between projects and running multiple agents in parallel. The choice depends on the work, with Antigravity favored for frontend tasks.
- 0:50:49
A Milestone-Driven Task Board for Agents
The speaker outlines a Command Center that gives agents an end-to-end understanding of a project, then populates a task board from weekly milestones and the work required to reach them. Tasks are assigned across multiple agents to expose the app’s structure, flow, and core functionality.
- 0:54:30
Verifying a Git Push and Deployment Update
The speaker follows up on a failed or unavailable push, checks whether the deployment service has detected the Git update, and monitors its progress. Once the update completes, he opens the site to confirm that the new version appears.
Got a question about something in this stream?
Ask in the group