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 developing a command center that connects a project map, task board, prompts, and coding-agent sessions. It compares running agents inside Electron with a hybrid workflow that sends structured prompts to Warp or Cursor while preserving operator visibility, intervention, and task-state synchronization. Practical work covered prompt generation, Warp launching, automated paste, and interactive Claude Code sessions, but testing exposed failed prompt transfer, misleading completion states, and limits in Warp tab automation, so the end-to-end workflow was not confirmed complete. The stream also explores screenshot-, DOM-, and Playwright-assisted UI reconstruction, design-reference research, and assigning models by agent role, with some connections, mappings, and visual results still unresolved. Additional useful discussions cover content automation and SEO, validating tools before selling or paying for them, diagnosing exhausted YouTube API quota, using a project map to resist distraction, learning in public, and creating more technical content in Arabic.
43 moments from the stream
- 1:28:22
Comparing Warp launch with full execution inside Electron
Two approaches were compared: launching sessions in Warp, which was estimated to be lighter and faster to prototype but requires app switching and manual tracking and gives the app little control over failures; and running agents as child processes inside Electron, which is more involved but enables task-to-process mapping, status detection, result capture, clean separation of concurrent tasks, and automated pipeline chaining. The effort and feasibility scores were discussion estimates, not tested results.
- 1:30:57
Suitability of a Browser-Based Terminal Component
A brief assessment of a TypeScript web-terminal component: it supports browser-based command-line interfaces and UI experiences but is not a standalone downloadable application, so it may not match the needs of a broader command center.
- 1:31:26
A Command Center Connecting Tasks, Prompts, and Agent Sessions
A proposed command center that contains a project's tasks and system prompts while connecting multiple agents for automated execution. The design considers prompt assembly, cloud sessions, feeding results back into the command center, and preserving per-task attempt history instead of manually copying prompts into an external tool.
- 1:33:54
Trade-offs of Non-Interactive CLI Agent Execution
An evaluation of running an agent loop inside Electron without interactive intervention. Limitations include being unable to answer clarification or permission prompts mid-run, needing a custom output viewer, managing processes and memory, handling orphaned processes, and preventing conflicting writes when agents touch overlapping files. Strong prompts may reduce questions, but execution permissions remain the central concern.
- 1:35:39
Requesting a Practical Workflow Before Building an Execution Engine
The discussion considers whether a backend engine is needed to run a CLI while expressing reluctance to spend several days building something that may not be used. It concludes with a request for the proposed workflow to be explained end to end and demonstrated on a current task before committing to implementation.
- 1:38:05
A Full Agent Pipeline Inside Command Center
A proposed workflow would run Explorer, Planner, and Builder stages from a task card, stream their output, save each stage’s result, and pass it to the next stage. Although appealing, managing the whole process inside Command Center was judged complex and potentially excessive, so the option was set aside for the moment.
- 1:44:48
A Hybrid Task-Board Integration with Warp or Cursor
The simpler option has the task board generate a precise prompt containing task context and reporting instructions, then send or copy it into Warp or Cursor instead of hosting the agent session inside the app. The user can watch and redirect the session while the agent is instructed to update the tracker; because this depends on instruction-following, a timeout was proposed to detect missing updates.
- 1:50:10
Choosing the Warp Path and Drafting Its Implementation Spec
An API was considered as a flexible alternative but judged not cost-effective for this use, while Warp-specific automation remained somewhat uncertain. The preferred design would open a session within the existing Warp window, run a generated task prompt, and report the result and completion state back to Command Center’s swim lane, leaving the session available for interaction when work remains. A Markdown implementation spec was then requested, with the reporting protocol identified as the core minimum; the pages end while deciding whether to build the initial steps together or sequentially, not with evidence that the implementation was completed.
- 1:56:15
Using a Project Map to Resist Distraction
The speaker discusses how easily coding projects branch into new ideas without reaching marketing or monetization, and says a project map, which he describes as a swimlane, was necessary for maintaining focus.
- 1:57:39
Building and Testing an Automated Task Runner
The speaker combines several changes into one feature intended to generate a specific prompt, open a Warp session automatically, and run the task. After the build completed, he restarted the development environment and requested a temporary card, which then appeared for the integration test, but called it a game changer only if it worked; the captions do not confirm final success.
- 2:04:14
Testing Automated Task Launching in Warp
A test card was configured with agent execution, an Explorer pipeline, and a prompt override, then both “Run in Warp” controls were tried. The launcher opened Warp but did not start a new session with the intended content, prompting investigation of the dispatch mechanism and a possible conflict with an active session.
- 2:11:54
Using Clipboard Paste as a Practical Dispatch Workaround
Testing exposed a request for permission to control system events. A mechanism that opens the required tab and uses Command‑V followed by Enter was then tried; it was treated as a small trade-off for a more reliable dispatch path, and the launcher was asked to adopt it.
- 2:17:27
Retesting the Flow and Finding a Misleading Completion State
The updated flow was retested: it writes the prompt to a temporary file, opens a Warp tab in the project directory, and schedules temporary-file cleanup. Although the test card appeared complete and the flow was reported as successful, the presenter did not observe a Claude session starting, so he requested diagnosis and a reset of the card for another test. The stated goal is for “Run in Warp” to initiate a real session automatically.
- 2:23:03
Diagnosing failed automatic prompt transfer into a new Warp session
After noticing that prompt generation may have been skipped, the assembled prompt was regenerated and the Warp launch flow was tested again. The intended behavior was to open a new session and transfer the prompt automatically, but pasting and pressing Enter did not run it. Diagnosis began with the unconfirmed hypothesis that the new Warp tab was not ready for input or that the paste was going to the wrong place; this page does not establish a confirmed cause or successful fix.
- 2:28:37
From a One-Shot Command to an Interactive Claude Code Session
The print mode was found to execute one prompt non-interactively rather than open a conversational session. The proposed adjustment launches a full Claude Code session, sends the assembled prompt as its first message, and keeps it open for follow-up prompts.
- 2:33:33
Warp Tab Automation Limits and Practical Alternatives
Tasks could be launched in new tabs, but the preferred workflow was to add a terminal to the active project tab and avoid tab proliferation. According to the discussion at the time of recording, Warp’s available actions did not expose programmatic pane splitting or terminal insertion into an existing tab, so alternatives included copying the command for a manual split, keeping tasks in separate tabs, and resuming sessions across tabs.
- 2:41:11
Reviewing Model Assignments by Agent Type
After a test automatically invoked Sonnet, the agent-to-model mapping was reviewed, raising a question about why planner and builder roles were assigned Sonnet instead of the requested Opus model. The excerpt identifies the unexpected mapping but does not document a resolution.
- 2:43:06
Managing Parallel Tasks in a Command Center
He discusses a design where each task opens in a separate tab or session, allowing results to be reviewed, tools to run side by side, and multiple optimizations to proceed in parallel. Color-coding tasks is proposed to make them easier to track and hand off within the command center.
- 2:44:19
Assigning Models by Agent Role
He proposes using the best available model for the builder and mapping different models to roles such as planner and explorer, mentioning Opus for planning and Sonnet for exploration as design choices.
- 2:47:27
Launching a Prompt Through Warp and the Command Center
He walks through a prototype flow in which pipeline mode generates prompts, the user selects an agent such as Explorer, and a “Run in Warp” action—proposed to become “Run in Command Center”—opens a linked terminal session. Some details of the resulting behavior remain unresolved in the explanation.
Got a question about something in this stream?
Ask in the group