One AI Department Done. 3 to Go.
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 work-in-progress automation system that organizes AI agents like connected departments for research, coordination, content production, publishing, and analytics. Luqman explains a possible progression from scheduled tasks and role-specific instructions toward shared goals and indicators, with research handed to agents that adapt posts, articles, and short videos for each platform. Practical work includes reviewing the post-stream pipeline, trimming silence and adjusting clips, and preparing operations across LinkedIn, X, Facebook, and Instagram while retaining human review initially. Speed, quality, resource use, queues, measurement, and agent autonomy still require testing and refinement; the stream does not establish a successful end-to-end system. It also discusses distribution and building in public, comparing direct outreach with SEO, social content, and advertising, alongside the speaker’s view that understanding an audience and solving a real problem matter more than selling a generic chatbot. Audience questions broaden the discussion to changing developer roles, learning and working with AI, game development, video-editing tools, background or local agent execution, community platforms, tools, and subscriptions. Other cautious, unresolved topics include AI jobs and pricing, proposed Egyptian data centers, model and hardware performance, YouTube Shorts growth, and ideas for competitions and products; these are generally presented as opinions, proposals, or open questions rather than verified findings.
171 moments from the stream
- 3:25:28
An Unverified Observation About Token Usage
Participants discussed an impression that token consumption might have decreased, but they did not verify that any change occurred. They also questioned whether such an improvement had been officially announced or independently discussed, leaving the observation as an unsupported impression requiring evidence.
- 3:26:51
Setting the Writing Voice for Each Channel
The work focuses on reviewing system instructions to ensure that each channel has an explicit, appropriate target voice and that incoming material is rewritten in a form suited to its intended outlet.
- 3:30:26
A Multi-Agent Research and Content Distribution Workflow
The speaker outlines a workflow in which one agent performs recurring research on specified topics, then passes the findings to a coordinating role that selects subjects and distributes them to writing and channel agents. Stronger models may be reserved for decisions or visual outputs when needed, simpler models can handle writing, and the setup allows model changes and per-agent token tracking.
- 3:33:29
From Agent Routines to Goal-Directed Work
The speaker describes a progression from assigning an agent a repeated daily routine to stating a desired outcome and letting the agent devise its own tasks and workflow. Shared goals and indicators would help agents track progress and assess whether the plan is working, but this is presented as an experiment and a reported future expectation, not a demonstrated result.
- 3:40:08
Clarifying Repository Boundaries and Commit Scope
The speaker raises confusion between a living graph and another repository, asking what the actual problem is and exactly what should be committed. This page does not capture a clear answer or resolution.
- 3:41:28
Reusing a Research Brief to Avoid Duplicate Work
An instruction tells the agent to read the previous day's brief before researching, so it does not repeat work across areas including AI shopping search, development, competitor activity, and public questions.
- 3:50:04
Broadening a Content-Research Agent from a Niche to General AI News
The speaker explains that a research agent initially focused on specialized commerce, platform, and AI-related topics, then was redirected toward AI more broadly so it could support more varied content on LinkedIn and X rather than staying confined to one subject.
- 3:51:58
Managing AI Agents with Goals Rather Than Tasks
The speaker discusses activating and iteratively refining an agent-based workflow, emphasizing that giving agents clear goals can matter more than assigning only specific tasks, while expecting normal errors during development.
- 3:56:42
From a Slow Learning Curve to Faster AI Workflows
The speaker described the beginning as slow because of the learning involved, but said work can accelerate once the tools and workflow are understood.
- 4:00:06
Mobile Game Marketing and Misleading Ad Experiences
The speaker argued that game marketing is difficult, pointing to ads that present an experience very different from the downloaded game, while describing game development as an enjoyable field.
- 4:02:12
How AI Is Changing Game Development
In a discussion with a participant trying AI in game development, the speaker expected a major improvement in speed and possibly quality, especially for someone with prior experience. The participant said they had published about 13 games on Google Play.
- 4:06:15
Distribution May Be the Main Barrier to Product Success
Luqman discusses how weak revenue or a difficult market can push a builder toward another idea. In his view, solving distribution and establishing a reliable traffic source removes a major obstacle, leaving product quality as the main remaining factor. He presents building a relevant audience alongside the product as one possible approach, not a guarantee of success.
- 4:10:02
Building for an Existing Audience Starts with Understanding It
Responding to a participant who has an established audience and wants to use a new project in their content, Luqman suggests building something for that reachable audience while first studying characteristics such as age range and general audience makeup.
- 4:13:14
Organizing Large Tasks with Branches and Git Worktrees
Luqman explains a workflow that gives each feature or substantial task its own branch and uses Git worktrees for larger project scopes. After validating the code, the work is merged into the main branch; he also notes the need to clean up accumulated branches and strengthen safeguards as the project grows.
- 4:15:24
Recognition Errors and Conversation-Forking Friction
An experimental workflow is described as stopping when a word is misrecognized, prompting an attempt to fork the conversation and continue multiple tasks in parallel. A previous fork had also lost some context, and these issues were expected to improve over time.
- 4:21:54
Using AI to Generate Context-Aware Ideas
The suggestion is to give an AI model context about a person’s background, skills, goals, and ambitions, then ask it for ideas about ways to earn money based on that context, without implying guaranteed results.
- 4:22:41
Designing an Automation System as a Company of Agents
An internal work-in-progress is described as an automation system tailored to business goals. Connected agents are organized like employees, each with a goal, responsibility, and system prompt, drawing on ideas learned from other tools. Its anticipated traffic impact remained an untested expectation in this excerpt.
- 4:25:01
Python and Networking as a Start in Development
A brief answer presents learning Python and networking fundamentals as a good starting point toward becoming a junior developer.
- 4:30:01
A Practical Approach to Caption Errors
The speaker explains that manually verifying every word in a long stream is difficult and calls for a durable way to reduce errors while acknowledging that perfect accuracy may be unrealistic.
- 4:34:22
Building Agent Files from Responsibility Contracts
The discussion presents an agent’s role and responsibilities as a system prompt treated like a contract. An orchestrator can help draft it, with multiple agents stored as Markdown files in a dedicated folder or repository, without requiring deep technical expertise.
Got a question about something in this stream?
Ask in the group