Optimizing Sarmad
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 Sarmad as an interconnected agent system: defining roles and delegation paths, monitoring token usage, experimenting with Telegram-based control and reporting, and reviewing the interface to improve visual separation. It also demonstrates automated content creation and cross-platform publishing; only the social workflow had been verified, while other channels still needed refinement. A major theme is that building a product is insufficient without distribution. Luqman discusses traffic acquisition, measuring content impact, competitor research, gradual ad testing, and a possible shift from Sarmad’s free trial to a free tier. He stresses that limited traffic and data prevent firm conclusions about pricing or conversion, and that cost reductions should not override privacy or healthy margins. Audience questions cover choosing tools such as Claude and Codex based on practical efficiency and usage limits, learning through real projects, AI-engineering paths, marketing, freelancing, and workstation choices. Several ideas—including open-sourcing Sarmad, proactive agent hierarchies, traffic projections, and revised pricing—remain proposals or early tests rather than completed outcomes.
116 moments from the stream
- 0:03:49
Mac Studio Workstation Specifications
Answering a setup question, the speaker shows a Mac Studio with a 42-inch display and two monitors, corrects the unified-memory figure from 32 to 36 GB, and identifies an M4 chip and a 2025 model, adding his opinion that it handles his workload very well.
- 0:04:54
An Initial Look at Sarmad’s Graphing System
The speaker introduces Sarmad as a product he built around automating a graphing system and says he may release it as open source on GitHub. He also describes a general visualization of whether organizational departments are active or idle, noting that the visual design may change.
- 0:06:17
The Challenge of Adapting Sarmad to Other Projects
The speaker discusses the potential challenge of making Sarmad activate within different projects and automatically generate a project-specific organizational chart. He frames an open-source release as conditional on the system operating automatically and beginning to attract traffic, not as a confirmed release.
- 0:07:49
Measuring Interest and Verifying Before Release
The speaker says the initial goal is to activate the tool and measure incoming traffic through a dedicated section, while emphasizing that it must be verified first because he does not want to publish something that does not work.
- 0:09:21
What Is Sarmad, and How Does It Relate to Graph Engineering?
Answering an audience question, the speaker describes Sarmad as a tool he built for himself and later thought might help others, intended to automate a graphing system. He begins comparing it with graph engineering and distinguishing that from loop engineering, but the definition is cut off and the terminology remains unresolved in this excerpt.
- 0:10:40
From Loop Automation to an Interconnected Agent System
The speaker distinguishes running an agent on a schedule for a fixed task from connecting multiple agents around a shared objective. He explains his preference for a system where each agent’s output is visible, while stressing that its value should be judged by a practical result such as traffic acquisition rather than the graph’s appearance or complexity.
- 0:13:29
Defining Agent Roles and Reporting Lines Before Delegating the Goal
The discussion considers turning the system into a multi-organization service desk with requests routed to specific people or agents and working relationships made explicit. The speaker says each agent needs a contract or job description defining its duties, objective, and organizational position, then describes a future vision in which a lead agent receives a broad goal and creates the teams and agents needed to pursue it. He attributes this vision to others he has heard and does not establish when it will become practical.
- 0:16:20
Connecting Traffic Acquisition with Pricing and Conversion Validation
The speaker explains that intense competition makes traffic acquisition a priority and that insufficient traffic prevents a reliable judgment about price fit. He suggests tracking how users move from installation and a free trial to payment; weaker conversion may indicate a pricing problem, but it is not a sound conclusion without enough traffic and data. He also notes that changing the pricing structure may require application changes.
- 0:20:55
A helper tool that chooses among supplied options
The group discussed a tool whose name is unclear and which does not operate as a standalone model. In the fragmented explanation, it receives defined options, evaluates them, and selects among them inside an existing system. Its exact nature and claims about speed or accuracy remained unverified.
- 0:25:12
Assigning tasks from team specifications
A general example suggested giving a tool a task together with team-member specifications so it can propose how to assign the work. This was an illustrative explanation, not a demonstrated test or result.
- 0:25:29
Reducing distraction from constant AI model switching
The speaker described the difficulty of continually following new AI news and model releases. He preferred keeping a small core toolset, such as Claude and Codex, while testing notable new tools selectively to reduce distraction.
- 0:28:38
Practical efficiency matters more than small quality differences
When discussing how to choose a coding assistant, the speaker argued that most users may not notice small differences in intelligence or output quality, while efficiency and usage limits are immediately visible in daily work. He therefore suggested comparing the available options in practice and choosing the one that delivers better usable capacity, while making clear this was his opinion and that one discussed option had not yet been tested enough in practice at the time of the stream.
- 0:33:51
Evaluating AI Model Efficiency Through Token Usage
The discussion notes that calling a model the best may refer to efficiency rather than intelligence alone, and proposes monitoring token consumption before judging it.
- 0:34:47
A Tentative Explanation of Reusing Linked Credentials
Based on his understanding, the speaker describes a tool that links account credentials once so they can be reused across agent workspaces without repeatedly entering an email address and password.
- 0:36:23
Planning a Large Project in Stages
The discussion says a large project should be understood and divided into stages before beginning a fresh execution session.
- 0:37:50
Building an Automated Multi-Platform Content System
He demonstrates a workflow that automatically creates posts and videos for multiple platforms, including voice-over, background music, and a call to action, then distributes them through a scheduled publishing calendar. He clarifies that only the social workflow has been verified so far; programmatic SEO, articles, and ads remain paused and need refinement before broader activation.
- 0:40:56
Measure Content Impact, Not Just Publishing Output
He argues that publishing many posts is not an outcome by itself: teams should track the traffic they generate, investigate weak results—including content quality—and use the evidence to improve.
- 0:41:43
Onboarding Sarmad Before Wider Release
While discussing wider access to the promotion system and a possible open-source release of Sarmad, he stresses the need for a solid onboarding flow that can take a repository, create an initial work structure, and let users edit the generated setup.
- 0:43:52
Open Source as a Distribution Strategy
He offers the qualified view that a strong product may use open source to build attention and adoption, then use that momentum to support later expansion. It is presented as a strategic option, not a guarantee of success.
- 0:46:05
Turning an AI-Enhanced Skill Into Income
A general answer on earning income: choose a skill that solves a problem, improve your capability with AI, and offer it online as a service, for example through freelance platforms.
Got a question about something in this stream?
Ask in the group