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 documents practical work on a website interface and a project command center, alongside broader discussion of coding-agent workflows. UI reviews covered seamless animation, cards, locked states, legibility, and Telegram calls to action, emphasizing that good visual results require explicit standards, reference screenshots, and human review. Some visual effects still needed refinement, while state automation was deferred because the future flow and requirements were unclear. For the command center, the discussion covered dependencies, waitlists, Shopify requirements, and human, agent, and paired execution modes. A custom MCP server was then built to provide task context and receive status updates: basic create, edit, and delete tests worked, while Electron refresh behavior and fuller milestone and Agent Hub coverage remained unresolved. The stream also explored enriching agent instructions with current context, acceptance criteria, and references; reviewed an agency website from visual and sales-funnel perspectives; and considered turning live-build lessons into courses if audience interest is demonstrated. Livestream stability, YouTube API behavior, and Cloudflare deployment were monitored throughout; some states later appeared correctly, but several underlying causes were not conclusively established.
53 moments from the stream
- 0:53:48
Replacing early-access email capture with a Telegram group
While reviewing the remaining website sections, Luqman questioned the value of the early-access email collection section and proposed removing it in favor of an invitation to a Telegram group.
- 0:57:11
Clarifying the Telegram Community Call to Action
The early-access email capture section is removed in favor of a community section whose title and wording identify Telegram clearly and directly, alongside an intention to use more Arabic in the copy.
- 1:01:03
Comparing Website Improvements and Deploying the New Version
A comparison of the first and second website versions highlights that initial construction took less effort than later refinement, with room for continued optimization, followed by deploying the changes from GitHub and checking the updated site.
- 1:02:44
Diagnosing the Live-Stream Indicator After Cloudflare Deployment
The YouTube API-backed live indicator worked on localhost but initially failed to show the same state after deployment to Cloudflare. Possible cache and RSS-discovery explanations were discussed, and the published site later displayed the live state, though no definitive root cause was established.
- 1:06:38
Understanding Stream Latency and Monitoring OBS
Explains that a short delay while a stream travels across the internet can be normal, and suggests watching the chat, code, and OBS green status indicator during the broadcast.
- 1:13:01
A Command Center That Turns a Project Plan into Trackable Work
Describes a command-center concept for completing, launching, and later optimizing a project. It combines a swimlane, task board, and agent hub; the swimlane maps stages, tasks, dependencies, progress, and project domains. Planning documents plus structural and ecosystem descriptions were given to several language models—including Perplexity, GPT, Claude, and Gemini—to help produce the detailed plan.
- 1:16:04
Building a Waitlist Before Product Launch
Discusses marketing while the product is still being built through a pre-launch campaign that introduces the forthcoming product and invites interested users to join a waitlist for a future trial.
- 1:16:51
Integrating Product Distribution and Shopify Requirements into the Execution Plan
Discusses building a waitlist alongside product development and reviewing Shopify’s app-submission requirements from the outset rather than waiting until submission. Compliance items are attached to weekly tasks because review may return an app over billing or UI issues, while schedule estimates are treated as general guidance that depends on the operator’s experience.
- 1:24:22
Limits of Previewing an Electron App Inside Cursor
Explores whether an Electron app can run inside Cursor’s preview panel. The stated conclusion is that the embedded preview is URL-based: the renderer can be loaded temporarily for pure UI styling, but window APIs, the preload bridge, IPC, and filesystem-dependent behavior will not match the full Electron app. Practical alternatives are running the app in an external window or attaching a debugger.
- 1:26:47
Planning Interdependent Domains and Building in Public
Reviews how project domains—including the checklist, distribution, storefront work, and product operations—depend on one another and therefore need coordinated sequencing. It also presents building publicly from scratch as a way to review progress later and let viewers observe a real development workflow.
- 1:29:51
A Task Board Assessed and Executed by a Coding Agent
Explains the command center as a project board showing weeks, sibling tasks, notes, and blocked status. A blocked task depends on earlier work, and the discussion proposes having a coding agent assess dependencies and priorities, update completion states, and potentially execute tasks through a dedicated execution mode, while noting that the underlying board concepts are common in project-management tools.
- 1:34:06
Task Execution Modes: Human, Agent, and Pair
The design distinguishes human-only tasks, agent-executed tasks, and a tentative pair mode in which a person and agent hand work back and forth. It also considers a history of task attempts, errors, and operator feedback, while acknowledging that some interface purposes still need review.
- 1:36:27
Task Prompt Generation and the Automation-Control Tradeoff
The speaker describes selecting roles such as explorer, planner, or builder, optionally overriding or previewing a task-specific prompt, and launching an external work session. A multi-role pipeline can hand work from one role to another, but he considers that degree of automation excessive and prefers stronger operator control.
- 1:39:08
Agent Hub for Synchronizing Agent Work
The Agent Hub is presented as a place to review agent activity across a project and let each agent know what the others worked on. The proposed coordination is state and context synchronization rather than literal direct conversation.
- 1:41:05
Limits of Static Prompts in an Agent Command Center
The discussion compares a normal terminal session, an in-command-center multi-agent pipeline, and an interactive external session. The goal is both precise task guidance and continuous project synchronization, but a preassembled prompt only supplies a frozen snapshot, cannot fetch new context mid-task, and relies on a fragile manual reporting protocol.
- 1:45:36
MCP as a Live Context Server for the Command Center
The discussion proposes a local MCP server connecting an agent session to the command center. Instead of front-loading all context into a static prompt, the agent could request task details, acceptance criteria, and relevant principles on demand, then report status changes, blockers, and completion through tool calls reflected in the UI. This was selected as a direction to explore and implement, not presented as a completed, validated feature.
- 1:48:40
An MCP Server for Connecting Claude Code to Task Context
The discussion outlines a standalone MCP server that lets Claude Code sessions retrieve task context from a command center, including status, milestones, dependencies, constraints, and execution criteria, then perform the work and report updates back. It is presented as a simpler alternative to a multi-agent pipeline that hands work from exploration to planning and building, while retaining agent selection and optional prompt or context overrides. The page later reports that the server was built and configured, but does not show a completed validation of its results in this excerpt.
- 1:56:29
Making New Files Visible in a Warp Project
A workflow issue was raised about newly added files not appearing immediately inside a Warp project, with screenshots or files apparently requiring manual handling. A way to refresh the file view without disrupting the workflow was researched, but the captions do not establish that a final solution was applied.
- 1:57:43
Why AI Needs an Explicit Request for Alternatives
The speaker argues that an AI assistant tends to interpret requests literally and may not volunteer alternatives beyond the path it was given. After being explicitly asked to revisit the goal and find another option, it proposed an MCP server as a better fit for the described problems. This is the speaker’s observation, not an established universal rule.
- 1:58:38
An MCP Server Connecting an Agent to a Task Workflow
A custom MCP server was built and configured to connect an agent with the project’s task center or store. The described tools cover starting, completing, and blocking tasks; reading context, progress, and checklist status; and recording events in an audit trail. Testing from a new session in the project directory—starting with a read-only query—was proposed, but no test result is documented here.
Got a question about something in this stream?
Ask in the group