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
- 2:02:54
Clear, Goal-Based Roadmaps for AI Work
The discussion frames automation-loop failures as learning experiences and stresses that a project roadmap should be clear to both the developer and the AI. It recommends organizing work around larger missions or goals instead of long lists of granular tasks, while using tools that shorten the build process. It also outlines an interface for viewing agents, the orchestrator, and ongoing work to support project development.
- 2:07:29
Computer Resource Pressure After a Stream Interruption
After a stream interruption, the speaker associates the instability with mounting CPU, memory, and disk-space pressure and notes that the machine's workload needs to be reduced.
- 2:10:12
When to Work Locally or in the Cloud
The speaker prefers working locally when using his computer directly and turns to the cloud when away from it or when sending work from a mobile app to the computer.
- 2:18:47
A Mission-Control Concept for Building Projects and Coding-Agent Workflows
The speaker discusses a tool he has tried and found promising in some respects, while stressing that he has not yet seen real token savings. He then describes a broader mission-control system focused on building projects, contrasting it with a management system for overseeing already-built projects through automation and functions such as customer service, accounting, and marketing. He argues that harness engineering could improve the efficiency of coding-agent workflows and potentially save time.
- 2:22:28
Organizing a Separate Command Center for Each Project
An audience member described agents editing unintended files when one command center spans multiple projects. The speaker outlined an organization where each project has its own page, presenting it as a potential improvement for reducing confusion and iterating on the workflow.
- 2:23:42
Building in Public While Protecting Sensitive Details
The speaker said he may show the building process publicly without exposing sensitive documents or project details, focusing instead on tools, project management, and workflow elements that can be shared safely.
- 2:25:51
Detecting Stuck Agents and Limiting Token Waste
The speaker recounted a stuck agent continuing to work instead of pausing, asking for clarification, or reviewing its output, leading to heavy token use. He argued that exposing agent status and consumption helps reveal the problem earlier, and noted that higher effort settings may increase usage.
- 2:29:12
How Ultra Code Orchestrates Multiple Agents
The speaker characterized Ultra Code as an effort and orchestration mode rather than simply a smarter intelligence tier, allowing an orchestrator to distribute work across multiple agents. He showed an interface for tracking each agent’s status, assignment, code changes, completion, stalls, and token use, describing it as a personal tool whose practical value he was still evaluating.
- 2:34:24
Documenting partnership agreements
As a general practical opinion, the speaker stressed documenting partnership agreements in writing and with signatures, even among people who know each other well.
- 2:34:56
Judging coding tools by execution, not only vision
The speaker discussed a circulating comparison between two models, then said that liking the overall vision did not make the resulting execution acceptable. The model names are not clear enough here to assert confidently.
- 2:36:28
Preparing long-running agent sessions and detecting stalls
The speaker explained that long autonomous sessions require permissions, files, and needed services to be prepared in advance, followed by orchestration that exposes stalled agents or work drifting outside the task. He also proposed usage analysis to distinguish productive work from waste, without surfacing private usage figures.
- 2:41:00
Starting with an agent tool on a limited budget
The speaker tied subscription choice to the available budget and offered Codex as his personal recommendation for a paid option in this discussion. He also considered a basic idea of using a second chat to monitor the first, emphasizing incremental workflow improvement.
- 2:51:06
Agent monitoring dashboard versus control harness
The speaker distinguished an interface that displays agent activity with limited controls from a system that actively determines how agents work. He described moving from mostly read-only monitoring to direct control as more powerful but double-edged.
- 2:52:26
Building the shown system without n8n
Answering an audience question, the speaker said the system he was demonstrating did not require n8n to build.
- 2:52:40
Running an automation tool on a primary computer
The speaker said he did not see high risk or a need for a VPS when running the mentioned tool on a primary computer, but added that he had heard of a recent update and had not reviewed it, so the assessment remained only partially verified.
- 2:54:31
The Shift Toward Tool-Enabled Autonomous Assistants
The speaker describes a single autonomous agent equipped with tools that can monitor multiple work sources, alert the user, and suggest next steps. He sees the market moving toward digital assistants that resemble personal secretaries. Some product names are unclear in the captions, so the examples are left unattributed.
- 2:57:54
Forecasting Usage-Limit Consumption
A feature is discussed that tracks token-consumption rate and estimates when a weekly limit may be reached if usage continues at the same pace. The speaker also wants to leave room for others to build extensions on top of the tool.
- 2:59:58
A Unified Agent Dashboard with Open-Source Contributions
The speaker outlines a side tool that gathers detailed agent activity into one platform and surfaces active and completed work. He plans to present it professionally through a website and GitHub repository, and is considering open-sourcing it so the community can improve the application through contributions and possibly a competition.
- 3:03:48
Separating Company Operations from Software Building
The speaker distinguishes an automation system for company operations such as customer service and marketing from a “Mission Control” environment for building code, applications, and projects while tracking tasks and agents. The central point is to keep each tool in its appropriate role.
- 3:08:03
Competition Value, Incentives, and Judging Criteria
The speaker argues that a competition’s value is not limited to its prize: participants can gain experience, while a compelling incentive can attract attention. Entrants may have widely different experience and resources, an imbalance that remains unresolved. He proposes defining the requested work and judging code quality and the strength of each addition to the project, while noting that some criteria remain undecided.
Got a question about something in this stream?
Ask in the group