Meen el hodhod dah?
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
A practical stream centered on an unfinished “Mission Control” tool for visually supervising coding-agent swarms, exposing their status and consumption, and detecting stuck loops, alongside live reviews of the interface, bird characters, and animation. The discussion broadened to goal-based roadmaps, separating software-building tools from company-operations systems, and balancing custom agent harnesses against added complexity and waste. Luqman answered audience questions about Codex, Claude, AI tools, learning, fundamentals, and hosting, repeatedly framing many comparisons as personal experience rather than conclusive tests. He also argued that distribution and demand validation can be harder than technical construction, and discussed a possible open-source GitHub release with community contributions or a competition whose prize and judging details remained undecided. The interface, motion, roles, naming, and release plan were still incomplete, and neither real token savings nor practical success had yet been established.
141 moments from the stream
- 0:01:44
Managing Overlays and Chat Placement in OBS
A question was raised about preventing a popup from appearing on screen in OBS. The live scene was then shown while the chat box was repositioned centrally for easier viewing; the captions do not provide a clear solution to the popup question.
- 0:07:20
What Role Does “Al-Hodhod” Play in the New Project?
A brief, vague introduction describes “Al-Hodhod” as the element that will help determine whether the new project can be built, without explaining its identity or mechanism on this page.
- 0:11:08
Merchant AI Visibility Project and Sarmad Automation
He reviews a Shopify-based project intended to help merchants appear on AI platforms and notes that attracting traffic remains difficult. He also explains that this experience led to Sarmad, an automation system for publishing and agent-assisted work with tools such as Claude and Codex.
- 0:13:28
Hodhod for Monitoring Agent Swarms and Orchestrating Missions
He presents Hodhod as an early prototype he plans to release as open source for supervising agent swarms: seeing what each agent is doing, tracking token use and models, detecting agents stuck in loops, and either cancelling them or alerting the lead agent. He distinguishes it from managing small tasks, framing it around larger “missions” distributed by a chief orchestrator, while noting that the design is still being developed and inviting community contributions.
- 0:18:06
Building Software with AI Without a Programming Background
Answering an audience question, he says he was not previously a programmer and has no programming background, but relies on his vision and experience in other areas while using AI to help implement his ideas.
- 0:20:30
Mapping Interconnected Projects
He describes keeping a visual map of projects when a small set of core projects generates many subprojects and workstreams, reducing confusion and providing a persistent reference while working.
- 0:25:59
Crowdsourcing Ideas for a Competition
He considers organizing a competition and posting a group poll to gather ideas before deciding its format; the competition topic and mechanics remain undecided.
- 0:27:06
Specialized Agents or a Custom Harness?
He discusses the view that narrowing an agent to a specialized task may reduce errors, then explains that a custom harness can fit a particular workflow but takes time and may constrain the agent or reduce quality unless its configuration and subagents are designed carefully.
- 0:28:58
Start Small and Track Agent Work on a Calendar
He recommends testing a small scope first and expanding only after it works, while using a calendar to track agents or “employees” so concurrent work does not become tangled and progress remains visible.
- 0:31:20
Adobe Alternatives and AI-Enabled Design Workflows
The speaker recommends considering open-source alternatives or tools with direct AI integration rather than assuming Adobe is essential, while noting that computer control may let an AI agent operate Adobe applications when needed.
- 0:32:14
Codex and Claude: Usage Limits, Token Allocation, and Capabilities
Explicitly framing this as personal experience, the speaker compares Codex and Claude. He argues that headline pricing alone is insufficient because token allocation and usage-limit resets differ, and views Codex as a more integrated product with stronger computer use while considering the tools broadly similar in capability. He also suggests Claude may have a relative image and video advantage due to broader visual-content experience, without presenting a conclusive test.
- 0:39:05
Asking an Agent to Find a Deck Plan and Brief Deck Two
The speaker asks an agent to locate the deck plan and provide a quick brief of what deck two contains, then points out that the displayed interface represents a real session with active actions rather than merely a prototype. The final action names are unclear in the captions.
- 0:40:55
A Central Interface for Monitoring Agents and Workflows
An explanation of bringing Codex and Claude workflows into a platform that shows what agents are doing, brings together code, tests, and GitHub activity, and lets the operator respond when an agent is waiting for input. The interface was presented as most useful for large multi-agent tasks, while a single orchestrator may be sufficient for small ones.
- 0:47:30
Organizing Projects and Tasks on a Standing Work Page
A look at using a standing section to see projects and their smaller tasks, identify what is in focus, and shape the day’s agenda. The discussion also raised how to add items beneath a project and move them, though the captions do not establish the final outcome of that interaction.
- 0:53:07
Resetting the Direction for the Bird Characters
A new session was started to recover the project context and decide how to portray the birds with clearer individual character, while considering the available video, motion, and AI tools.
- 0:54:05
Birds as a Visual Identity for Agents
The discussion explores using birds with distinct colors and traits to represent agents instead of familiar robot or creature imagery, with the hoopoe envisioned as the lead character. The concept is framed as a compact marketing device and an identity that could expand into multiple characters.
- 0:55:41
Do AI Tools Make Multiple Projects Practical?
The speaker relays a recently heard view that challenges the traditional advice for developers to focus on one project: AI tools may make it practical to test several projects in parallel, then concentrate effort where genuine audience interest appears. It is presented as an opinion and proposed strategy, not a proven result.
- 0:58:13
Observing Agent Work and Detecting Stuck Loops
The demonstration shows an interface that brings several agents together and exposes activities such as testing, repository checks, and code writing. The speaker says its main value for agent swarms is spotting stuck agents or inefficient usage, with controls to stop agents or question the orchestrator; this is based on his account of an earlier task that ran for a long time and became trapped in a loop.
- 1:01:21
Different Tools for Building and Operating a Company
The speaker outlines a vision that separates long-term company building from ongoing operations, with each side supported by an appropriate kind of control or orchestration system.
- 1:02:08
A Visual Interface for Monitoring Long-Running AI Agents
The speaker discusses a command-center concept that visually exposes what an AI agent is doing while pursuing a goal over an extended period. He wants to share the tool with the community and invite contributions rather than limit people to using it.
Got a question about something in this stream?
Ask in the group