Chill Build: Automation & Traffic
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 building a controllable automation system that organizes AI agents as a graph and turns research into content, short clips, and multi-platform posts, supported by a calendar, analytics, reports, and Telegram-based coordination. Practical work included checking the publishing pipeline, correcting topic relevance, refining tone and calls to action, and dividing reasoning tasks from deterministic local scripts. Luqman repeatedly emphasized that quality, measurement, and human review should precede scaling or unattended operation. In his view, the larger challenge is no longer software construction alone but acquiring and measuring traffic through tools such as Search Console and Google Analytics, then improving the product and pricing once enough evidence exists. Audience questions covered coding tools and plans, Codex versus Claude, learning through Micro-SaaS, agent tools, e-commerce platforms, SEO, editing, and AI video generation. Other discussions covered opinions on AI safety and the software industry’s future, beginner and freelancing advice, creator-platform monetization, payment-platform choices, the risk of exposed API keys, and the ethics of AI-altered product images. Claims about unidentified models, agent security, usage limits, and tool costs were treated cautiously rather than as established facts. The automation, publishing, and traffic outcomes remained unresolved: some tests and refinements looked promising, but several workflows stalled or were still being configured, and there was not yet enough performance data to call the system successful.
131 moments from the stream
- 4:09:24
The Limits of AI in Video Editing
The speaker argues that AI tools can assist with editing, but video editing is precision work and current AI results do not match the accuracy of a human editor.
- 4:09:50
A Practical View of ChatGPT and Claude Code
The speaker says he is not dismissing Claude Code and still sees it as valuable, but in his current experience ChatGPT is more efficient and enables a greater volume of work.
- 4:10:26
Designing a Human Voice for AI-Generated Posts
The speaker describes a workflow that first studies professional context, field, and writing style, then proposes a fitting tone and previews its consistency across several sample posts without producing the final posts yet. The stated aim is natural-sounding writing that is less obviously machine-generated, while acknowledging that some people care more about the value than the text’s origin.
- 4:14:17
Measurement as a Foundation for Work
A brief point emphasizing that measurement—and clarity about what is being measured—is essential, and that working without measurement or understanding the results lacks an important foundation. Some wording immediately before this conclusion is unclear in the captions.
- 4:17:47
Article Topics and Telegram Updates
The speaker asked who supplies topics for the articles, then considered receiving Telegram updates about how things were progressing; the supplied captions do not establish the source or mechanism.
- 4:20:44
Evaluating a Generated LinkedIn Writing Voice
The speaker reviewed a generated example of how his writing voice might appear on LinkedIn and said he liked it so far.
- 4:21:25
An Opinionated GPT vs Claude Comparison for Coding and Image Tasks
Answering a question about accuracy and token use in vibe coding, the speaker said he considers GPT more efficient and less token-intensive than Claude. He also preferred it for text-to-image and image-to-image work, based on his belief that GPT has more experience and data in image generation; these are attributed opinions, not demonstrated test results.
- 4:23:36
Using Examples to Tune a Writing Voice
The speaker inspected system-prompt-based outputs showing how he might write, using examples to judge whether the proposed tone matched his writing voice.
- 4:24:59
Separating Brand Automation from Personal Posts
The discussion proposed fully automated brand publishing that can be improved over time, while personal posts would use an on-demand workflow initiated by the account owner’s idea or voice message, followed by assistance with drafting and image creation, with the option to consult a domain specialist. The key qualification was that personal opinions and voice should remain the author’s rather than being invented automatically.
- 4:29:40
Image-Generation Options and Usage Limits
A brief, incomplete comparison mentioned image creation through GPT, local AI models, and building a workflow, alongside a reference to a weekly usage limit on one plan at the time of recording. The captions do not provide enough detail to characterize that limit further.
- 4:34:14
Choosing a Local Agent Tool as a Beginner
An audience member asked about installing Hermes locally on Windows 10. The speaker said he had not personally used Hermes, although he understood its general operation, and noted experience with other alternatives. His qualified advice was that the choice depends on the work, and a simpler tool may suit a beginner when runtime-harness details are not important.
- 4:34:54
Choosing an Automation Tool for the Workflow
The speaker advises against recommending an automation tool before understanding the workflow, and discusses practical constraints such as mobile access and whether Telegram can provide an interface. One tool name is unclear in the captions, so the discussion is not assigned to a specific project.
- 4:37:49
Connecting a Codex-Assisted System to Telegram
The speaker says he asked Codex to help connect a system he had built to Telegram, describing the integration as straightforward in principle.
- 4:40:17
A Manager Agent as the Contact Point for Other Agents
The speaker outlines a design in which one manager agent receives requests, delegates them to specialized agents, and sends a periodic summary, reducing the need to manage every agent separately.
- 4:42:18
Early Impressions of Two Unidentified Model Versions
The speaker reports that version 4.7 is said to use a new base model, unlike 4.6, which is associated with an earlier version, with claimed improvements on long tasks. He stresses that real use is the deciding factor and adds a personal impression of greater speed alongside occasional hangs and agents losing track of returning results. The model identity is unclear from the captions.
- 4:42:54
A Hierarchical Structure for Managing Specialized Agents
The speaker describes a hierarchy where a manager agent is the single point of contact, knows its direct reports, and relies on each of them to manage a specialized group operating under a defined program or routine.
- 4:44:23
Verifying a Publishing Plan Instead of Fabricating Numbers
After receiving conflicting publishing counts from different sessions, the speaker asks the system to inspect the current state and derive the actual posting cadence instead of fabricating numbers, while keeping personal accounts outside the requested scope.
- 4:48:06
Distributing Posting Volume Across Multiple Platforms
Clarifies that the proposed posting volume is distributed across several personal and business platforms, rather than concentrated on one platform, where a large daily batch is not expected to perform well.
- 4:52:20
Sanity-Checking a Publishing Pipeline Before Manual Activation
Requests a sanity check of the Burnish publishing pipeline before manually activating it, then observing posts being produced, scheduled, and sent across platforms. The aim is to understand and verify the workflow before relying on unattended automation.
- 4:57:32
Comparing Payment Options for an Amazon-Related Use Case
He discussed general options including Wise, Payoneer, PayPal, Stripe, and Amazon Pay, while noting that the use case was too unclear for a precise recommendation. He favored Wise based on his own experience, said he had not tried Payoneer, and asked whether the need was payment collection or connecting an Amazon selling operation.
Got a question about something in this stream?
Ask in the group