Chill Session Q&A - TTFD?
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 an unfinished attempt to build a multi-agent system for research, content production, and traffic acquisition, including role design, coordination, and visual observability. Forgotten routines and agent-to-agent conversations produced unexpected usage and made the system difficult to understand, so the speaker paused it, proposed mapping and simplifying its relationships, checking retry safeguards, and evaluating quality, reliability, and token use with measured results. Practical work included a pipeline for turning long streams into captioned vertical clips, an experiment in which Codex operated Claude through the computer interface, and visual previews and activity timelines intended to improve review. Cost efficiency, operational stability, and whether some visualizations accurately represented real activity remained unresolved. Audience questions broadened the discussion to defining agents like measurable job roles, choosing local versus cloud infrastructure and tools, and connecting error monitoring to an agent while retaining human review before merging. On the business side, the speaker stressed that AI makes building easier but does not solve selling: demand validation, traffic acquisition, conversion data, and pricing evidence should take priority over premature polish and excessive analytics. The stream also covered ethical competitor research, career skills and professional presence, limits of voice AI, subscription models and APIs, and cautious, unsettled discussion of superintelligence, employment, and privacy. The library additionally includes audience-led side discussions about caffeine timing and sleep, an experience of quitting vaping, awareness of nutritional support, work and stream schedules, and a debate over road development and the protection of Cairo’s historic neighborhoods; the health and lifestyle material was presented as personal experience or opinion rather than medical guidance. Its clearest theme is to favor simpler, observable systems and measurable outcomes over complexity and first impressions; the stream does not establish that the overall system succeeded or became economically viable.
106 moments from the stream
- 5:30:11
Using Multiple Agents in Codex
Answering an audience question, the speaker explains that a newcomer generally does not need to build agents manually in Codex: they can define tasks and let the system select suitable agents for the work. He adds that creating custom agents remains possible when needed.
- 5:32:00
Public Surveillance Versus Crime Reduction
The speaker discusses the argument that pervasive cameras and monitoring everyone could reduce crime, and strongly objects to that trade-off. He cites a commonly made China–US comparison, while making clear that his description of China is based on reports from acquaintances rather than direct verification.
- 5:33:21
Adding External Skills to Codex
The speaker explains that Codex can use built-in and external skills, suggesting that users search skill libraries and provide a skill link. He recommends first asking Codex to explain the skill and assess whether the project actually needs it.
- 5:34:20
A Brief Opinion on Surveillance in China
The speaker expresses his belief that surveillance in China is extensive and precise, while acknowledging that he does not know how much it affects people's everyday lives.
- 5:35:41
Reducing Token Use in Multi-Agent Workflows
The speaker suspects that repeated handoffs and agents revising one another's output drive up token use. He discusses setting a stopping threshold, but prefers to get the full workflow producing quality results first, then measure and reduce consumption afterward.
- 5:38:16
Comparing AI Tools with Real-World Use Cases
The speaker praises a video that compares AI tools across a set of real-world use cases, reporting time and cost per task before giving an explicitly personal preference. He considers the supporting resources valuable while noting that the final judgment is subjective.
Got a question about something in this stream?
Ask in the group