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
- 4:49:12
Choosing AI Models by Task Type
The speaker describes grouping work into larger missions or milestones and using different models for building, ideation, and research, favoring higher-capability models for exploratory thinking. This is his workflow preference, not a validated benchmark.
- 4:51:56
A Disclosed AI Assistant for Audience Comments
The discussion explores an AI assistant that automatically replies to comments while clearly identifying itself as an assistant. It highlights that comment replies should first review and understand the original content, with human oversight remaining important to catch misunderstood questions or context; direct messages may sometimes operate more independently.
- 4:53:35
Using a Browser AI Interface with Projects
The discussion mentions a recently released tool—its name is unclear in the captions—that connects a browser-based AI interface to project files, presenting it as one practical attempt to make use of available usage allowances.
- 4:59:46
Separating a Project from Mission Control
Recapping the stream’s main decision, the speaker said an unclearly named project would be separated from Mission Control for now because he sees Mission Control as a strong project worth developing independently.
- 5:03:29
Technical Products Still Need Distribution and Sales
The speaker sees potential in products serving the AI market, but stresses that they still require traffic and that years of engineering experience do not guarantee sales; in his view, reaching customers and selling the product are decisive.
- 5:05:30
Replacing a Free Trial with a Free Plan
The speaker describes changing the pricing model by removing the free trial and offering a free plan while keeping selected features paid. He frames this as an attempt to improve monetization, with traffic acquisition remaining the central challenge.
- 5:07:14
Measuring Cold Outreach Beyond Social Posting
The speaker discusses cold outreach as a possible source of targeted traffic after concluding that repetitive social posts may have limited impact and may dilute a platform’s presence. He proposes evaluating outreach through open, click, and reply tracking, reviewing why recipients were targeted, and comparing results with industry benchmarks. The projections discussed are estimates, not proven outcomes.
- 5:10:45
Considering Higgsfield for AI Video Content
An audience prompt suggests trying AI-generated video content with Higgsfield rather than relying only on images. The speaker confirms using the tool while expressing a personally critical opinion of its practices.
- 5:24:09
Correcting Voice Dictation Mid-Sentence in Wispr Flow
An explanation of a feature attributed to Wispr Flow: the speaker can revise an earlier statement while dictating, and the task text is updated instead of requiring the whole phrase to be repeated. The comparison with other tools is presented as the speaker’s opinion.
- 5:27:18
Balancing SEO, Ads, and Outreach for Traffic
A brief discussion of SEO being slow when faster traffic is needed, with ads or direct outreach presented as quicker options while still recommending a gradual start with SEO and GEO.
- 5:31:58
An Opinion on Choosing Between Bootstrapping and Fundraising
A personal view that bootstrapping may suit some founders, while experienced builders might target clear demand, demonstrate limited early traction, and then approach investors for an early round. It is presented as an opinion, not a universal rule or funding guarantee.
- 5:33:16
Running Hugging Face Models Locally and Hardware Limits
A participant pointed to models that can be downloaded from Hugging Face and hosted locally, while the speaker said he had not tried downloading them because his computer was already near its limits. No specific model was clearly identified, and the speaker reported no firsthand download or test.
- 5:39:59
Planning Goals in a Fast-Changing Market
Advice to set clear goals and plan how to reach them, alongside the speaker’s view that sustained work matters because the market changes quickly.
- 5:41:08
When Product Hunt May Fit Better Than Shopify
The speaker argues that Product Hunt may suit launching a standalone product or app outside a closed environment, while Shopify does not fit the case under discussion. This is a contextual opinion, not a general verdict on either platform.
- 5:47:03
Exploring a Tool Integration in a Claude Code Workflow
A question is framed for a new Claude session to explore integrating an unspecified tool or project at two levels: the regular workflow in the Claude Code setup and a new mission-control system under development. The captions describe a proposed investigation, not a completed integration.
- 5:50:58
Codex Autonomy and Application Architecture Questions
The speaker asks the agent to proceed in Chrome without requesting permission and says that, in his experience, Codex is more autonomous in some computer-use tasks. He then raises open questions about whether the interface is an Electron app and whether it is separate from the mission-control system.
- 5:53:36
Preparing for Fasting by Gradually Reducing Caffeine and Nicotine
The speaker believes caffeine and nicotine can make fasting more difficult and suggests reducing them two weeks to a month before Ramadan. The exchange also mentions personal experiences of headaches lasting through the first five days and, in another stated experience, about two days; the captions do not clearly mark the speaker changes.
- 5:55:48
Inspecting Codex Agent-Swarm Behavior
The speaker discusses agent swarms in Codex and then inspects a running mission. During the inspection, the speaker considers the observed behavior incorrect, though the captions do not clearly establish his reason or the agents’ exact state.
- 5:59:32
AI as Personal Support During Smoking Cessation
The speaker recounts a personal experience in which an AI assistant explained nicotine-withdrawal symptoms, checked in periodically, and offered reasoning and encouragement during cravings. This is his attributed experience, not medical advice or evidence of clinical effectiveness.
- 6:11:29
Are There Limits on Subagent Concurrency?
The speaker and an audience participant discuss an observed failure to launch as many subagents as expected, asking whether a default cap or configuration setting is responsible. They compare this with their experience in Claude Code without establishing the actual limit or the cause of the slowdown.
Got a question about something in this stream?
Ask in the group