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
- 0:04:23
From Tasks to Goal-Driven AI Agents
Luqman presents a qualified outlook, based on what he has heard and early experimentation rather than a settled fact: AI work may shift from tightly specified tasks toward large goals pursued over extended periods through sub-agents. He describes a platform intended to provide a project-wide view, showing what has been completed, what is happening now, whether anything is blocked, and general usage visibility without exposing private figures.
- 0:07:02
Checking Desktop and Mobile Stream Layouts
During the stream, Luqman checks whether mobile viewing uses a vertical format or remains horizontal, and wonders whether the vertical layout is intended for Shorts.
- 0:08:37
A Searchable Archive and Summaries for Long Streams
Responding to a viewer who wants to understand the work without watching long streams, Luqman describes a plan to transcribe and index streams so terms can be searched with timestamped results. He also proposes using AI to extract useful material into shorter summaries, and says the channel was started in March 2026 to work publicly and transparently; the page ends before his explanation of the work is complete.
- 0:10:20
From AI-Assisted Building to Go-to-Market
A brief framing of the full product journey, from starting an AI-assisted build through entering the market and promoting what was created.
- 0:11:35
Coding-Tool Usage Limits and Agent Workflow Design
After an audience member reports quickly exhausting a usage limit while building and testing modules, the speaker offers the uncertain opinion that AI services may become cheaper in the future. He argues that agents and careful work decomposition may save time and usage depending on the quality of the PRDs and task breakdown.
- 0:15:00
Does the Project Actually Need RAG?
Responding to a question about completing a RAG prototype, the speaker says a precise recommendation requires knowing its purpose and details. The first step is to define what the system needs from AI and verify that RAG is genuinely necessary, since the perceived need for it may be mistaken.
- 0:20:25
Choosing a Coding Tool Based on Usage Consumption
The speaker said that, in his experience, Claude consumed usage quickly, so he planned to rely on Codex temporarily. This was a personal impression, not a systematic comparison.
- 0:21:58
Ideas for Organizing Sessions in the Interface
The excerpt briefly brainstorms placing sessions on the left, distinguishing missions from quick sessions, and trying related elements in a top panel. The final interface design is not clear from the captions.
- 0:22:34
Suggested Resources for Exploring RAG Systems
A participant suggested reviewing NVIDIA NeMo Retriever, NVIDIA AI-Q Research Agent resources, and RAG blueprints. The speaker said he had not yet explored RAG deeply and would keep these resources on his radar.
- 0:24:24
Building a RAG Library for Interrelated Sources
A participant described a RAG system that collects a large body of books, organizes them into suitable sections, models relationships among them, and incorporates specialist input when tuning its logic and calculations. He reported—without an independent evaluation in the captions—that the approach produced highly accurate results comparable to specialist expertise.
- 0:29:58
Tools and Practices for Testing Web Applications
The speaker suggested Playwright for testing, Sentry for debugging and error tracking, and repeated end-to-end tests to check production behavior. He also mentioned a script that clicks application controls, but the explanation ends mid-sentence.
- 0:30:22
Agent-based checks do not prove scalability
Agents can trace interface flows and verify where controls lead as part of assessing production readiness, but the speaker cautions that this does not solve scalability or bugs that emerge with more users.
- 0:36:43
Software engineering value beyond writing code
The speaker acknowledges that AI coding tools can create understandable unease, but advises engineers to delegate some code writing and apply their training to deeper problems and engineering judgment.
- 0:37:40
Signing and distributing macOS applications
The speaker tentatively says that distributing a downloadable macOS app may require signing and an annual Apple subscription costing about $99–$100. This reflects the discussion at recording time, not a current verification of Apple’s policy.
- 0:38:01
Visualizing agent swarms and continuously watching a codebase
The speaker presents a planned design that changes a session’s representation when several agents run together, and proposes recurring “watchers” for tasks such as pull-request audits, secrets, and permissions, alongside surfacing stuck agents and possible causes. These are presented as planned concepts and diagrams, not completed results.
- 0:44:22
Handing work between coding agents
The session discusses moving work between coding tools when usage limits arise and preparing a handoff prompt so brainstorming can continue in Codex without losing context.
- 0:45:20
Reducing Codex latency by splitting tasks
A participant offers the practical observation that dividing work into smaller tasks improved their experience of Codex latency, without presenting benchmark measurements or a controlled comparison.
- 0:46:52
Letting an orchestrating agent order the topics
The speaker wants Astra to understand every topic, write build prompts, initiate sessions, and lead the ordering of work as long as everything is covered. This describes a desired workflow, not a demonstrated implementation.
- 0:50:04
Checking whether background music may be streamed
The speaker checks whether background music is audible and says he believes he is not allowed to play it, without specifying the applicable license or policy.
- 0:51:42
Mac Targeting and Framework Choice
The speaker asks whether the app targets Mac only and suggests that using Electron or Tauri may be easier.
Got a question about something in this stream?
Ask in the group