Meet the Swarm
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 wide-ranging, practical stream about building AI products in public. Its strongest thread is the shift from short tasks to long-running “missions” pursued by agent swarms, supported by a control center and living roadmap that expose progress, quality, and blockers. Luqman discusses requirements documents and system design as foundations for greater agent autonomy, splitting work between Codex, Claude, or specialized models, and measuring real cost per completed task rather than advertised token prices. These are presented as concepts, early experiments, and personal judgments—not settled results. Practical activity includes reviewing an interface and identifying a scrolling bug, requesting an interactive mock to clarify the design direction, and beginning an application walkthrough by granting access to a sample project and requesting a quick capability test. The stream also discusses web testing and continuous monitoring, with the explicit caution that agent-driven UI checks do not prove scalability. Audience questions broaden the stream to RAG and learning resources, context compaction, usage limits and tool handoffs, handwriting OCR, reviewed data-entry automation, and disclosed AI assistants for social engagement. It also covers app frameworks and community platforms, product validation, free-plan pricing, distribution and sales, SEO, ads, direct outreach, freelancing, and uncertain effects of AI on jobs and SaaS. Open questions remain around Mission Control’s interface architecture, subagent concurrency, the value and cost of bridging models, and whether several proposed integrations and workflows will work in practice. The supported conclusion is that documentation quality, low-friction onboarding, human oversight, realistic usage measurement, and distribution matter at least as much as model capability; many demonstrated products and workflows remained exploratory or unfinished.
124 moments from the stream
- 4:05:54
Challenges in Automating Data Entry from Handwriting
The discussion distinguishes computer-printed data, described as relatively easier to extract, from handwriting, whose difficulty varies substantially with each person’s legibility. It also notes that the absence of a suitable integration mechanism in the work system can create another obstacle to automation.
- 4:07:19
Gemini for Handwritten-Note Recognition
The speaker argues that modern Gemini vision capabilities have made text extraction from handwritten material—even older or difficult handwriting—far more practical, so handwriting is no longer a major OCR obstacle in his view.
- 4:11:02
Choosing a Platform for an Arabic-Speaking Community
The speaker explains choosing Telegram for the community because it is more widely used locally than Discord, while WhatsApp felt too personal for this purpose.
- 4:11:36
Automating Speech-to-Form Entry with Human Review
An audience participant describes a workflow that captures spoken content into a form and reviews the values afterward. The speaker endorses integrating AI into the workflow; private performance figures and account details are excluded.
- 4:14:52
A Control Center for Long-Running AI-Agent Missions
The speaker outlines a system that turns project documentation—such as system designs and roadmaps—into “missions” larger than ordinary tasks. An agent may work on one for days while the interface helps review its work, quality, status, and whether it is stuck in a loop; missions are treated similarly to milestones.
- 4:16:47
Payment Barriers When Trying Models
An audience participant says a model they wanted to try required an upfront payment on first use, so it was unsuitable and they chose an alternative. The model names are unclear in the captions and are therefore not attributed to specific products.
- 4:17:02
Learning from Corrections and Choosing a Model
A discussion of a system that learns from repeated corrections so its errors may decline over time, while noting that the model comparison was already several months old and a newer model might handle the task well.
- 4:17:45
Learning AI Project Building in Public
The stream is described as a transparent space for building and taking AI projects toward market, letting viewers observe the work and form their own judgments instead of receiving a short packaged comparison or sales-oriented lesson.
- 4:20:23
Monetizing an Open-Source Tool Through Virality
An early strategy for keeping a useful tool open source on GitHub while potentially offering paid features. The speaker argues that stars, virality, and perhaps a competition could create monetization opportunities if the tool is genuinely strong for vibe coders and software engineers, while stressing that the strategy is undeveloped and major returns are uncertain.
- 4:22:41
When Product Need Comes Before Monetization
The speaker distinguishes between a product he personally needs, which can justify building before deeply defining monetization, and one he does not need, which would require stronger market validation and a clearer reach and revenue plan. He also expresses uncertainty about the idea's fit for an Arab audience.
- 4:24:33
An AI Agent for Social Replies
A concept for an agent that handles direct messages and comments on Instagram and TikTok, with explicit disclosure to users that it is an AI-powered assistant.
- 4:25:17
Choosing a Meaningful Prize for a Tech Competition
An early discussion of a competition prize: cash would be the easiest attention-grabbing option, but the speaker finds it less meaningful and wants to consider an AI-related prize with practical value. No decision has been made.
- 4:26:06
Repurposing Streams and Reviewing Video Quality
A proposed agent repurposes stream recordings into shorter YouTube videos and thumbnails. Reviewing its first result, the speaker criticizes the video and thumbnail quality and the lack of an introduction, then asks for a detailed dissection; the request continues beyond the page.
- 4:32:09
Monetization Research and a Non-Chat AI Interface
The speaker discusses researching monetization models and tools resembling an early-stage idea, while envisioning an AI product that is not centered on chat, a CLI, a terminal, or an IDE and may not require direct prompting. The technical and commercial details remain preliminary and incomplete.
- 4:35:10
Start Before Finding the Perfect Idea
The speaker argues that continually chasing the perfect idea can prevent someone from starting. He feels more confident when projects overlap and reinforce one another, and suggests that building reusable coding tools may support new work rather than necessarily wasting time.
- 4:37:07
One Focused Project or a Portfolio of AI Experiments?
The speaker questions the traditional advice to focus on one software project, arguing that agents may enable several experiments to be launched before concentrating on those that gain traction. A counterview favors quality over quantity, while uncertainty remains over whether AI speed truly closes the quality gap and operational cost remains a practical constraint.
- 4:39:00
From Loop Engineering to Goal-Driven Agent Swarms
The speaker describes loop engineering as a phase that began in late 2025 and, in his view, ended in March, followed by a shift from assigning fixed loops to assigning goals. In his account, agents work through separate repository branches, continue when a partial loop fails, and showed coordination and exchange he had not explicitly requested; some tool names and implementation details are unclear in the captions.
- 4:42:23
Codex vs. Claude: Capability and Usage Limits
Responding to a general question about whether a paid subscription is worthwhile, the speaker recommends trying Codex while maintaining that Claude is better overall in his opinion. He says Codex can handle most of the same work, but finds both services’ usage limits restrictive and considers reset timing important to practical efficiency. This is personal opinion, not a systematic test.
- 4:43:45
Where Should the Task-Planning Interface Live?
While reviewing a plan containing milestones, tasks, and some human-owned tasks, the speaker raises an architectural question about where the mock interface should live: separate from mission control or integrated into it, and how it relates to another overview interface. The page provides no final answer, and some plan counts are too unclear to publish confidently.
- 4:47:56
Making Roadmap Tracking More Visual
The feedback calls for a simpler, more user-friendly interface with progress indicators and visual elements that make the roadmap’s current state understandable at a glance.
Got a question about something in this stream?
Ask in the group