Inbound vs Outbound 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 a widening gap between building products with AI—which is becoming easier—and the harder work of distribution, customer acquisition, and sales. Luqman reassesses a heavy reliance on inbound content and considers a broader mix of outbound outreach, cold email, advertising, partnerships, and SEO, with the practical principle of testing channels and concentrating on those that show real conversion signals. Practical work includes reconsidering Burnish’s pricing by replacing a time-limited trial with a free plan, investigating a reported HTTP 500 error and an observability gap, and simplifying AI-agent workflow diagrams; the records do not establish that these efforts were completed or successful. The discussion also explores multi-agent systems organized around shared goals, context, memory, and explicit follow-up, with a preference for starting small instead of introducing early complexity. In comparisons of Codex, Claude, and AI providers more broadly, Luqman emphasizes balancing quality, speed, price, usage limits, and completed workload, while repeatedly qualifying his views as personal and treating mixed reviews and benchmarks as insufficient for declaring an absolute winner. Audience questions broaden the stream to SaaS development, AI-engineering portfolios, content automation and scheduling, visibility in AI-generated search answers, team size, personal branding, transparent affiliate marketing, and AI support for study or business management. Major unresolved questions include the best acquisition channel, the commercial value of content, the right agent architecture, and the changing choice between tools and subscription plans.
133 moments from the stream
- 5:32:49
The Difficulty of Traffic and Sales
The speaker describes traffic acquisition and selling as among the hardest stages of the work, noting that progress requires time and learning from mistakes.
- 5:34:03
Model Cost and Usage Constraints
The speaker says the high cost of an upcoming model could make it impractical to use, and mentions focusing on two other options whose names are unclear in the captions.
- 5:42:32
Build an Agent System Around Goals and Quality, Not Agent Count
The speaker describes agent-system engineering as arranging AI agents to communicate and collaborate toward a shared objective, with both overarching goals and department-specific goals. Based on his experience, he recommends starting with a small number of agents and expanding gradually, since beginning with excessive complexity can cause workflows and tasks to become tangled. He also explains that a GitHub repository can serve as shared working memory connected to the workflow graph.
- 5:45:05
Coordinating AI Agents Around a Shared Goal
A conceptual explanation of a system that records agents’ actions, timing, and usage in structured files while giving them shared context and an environment outside the main project. The speaker argues that directing an agent network toward a time-bound goal could support areas such as marketing, traffic, and sales.
- 5:46:50
How an AI Agent Could Support a Physical Business
In response to a question about an automotive service and parts business, the speaker says he does not have a confirmed specialized use for its physical operations, but suggests using an AI agent as a general management assistant after providing sufficient business context and needs.
- 5:47:49
From Loop Engineering to Graph Engineering
The speaker contrasts an earlier approach—assigning an agent a specific task at a fixed time—with a newer approach that gives multiple agents a goal and lets them interact. He describes the former command-center board as a useful workflow map but, in his opinion, already dated by the rapid pace of AI development.
- 5:51:41
Design Questions for a Content-System Prototype
While requesting a clear weekly table of what a prototype would create and publish, the discussion raises key content-agent design questions: whether it researches topics itself, how it adapts tone and language for each format, and whether it produces video voice-overs. The speaker also says he simplified the earlier design from several agents to one, which he considers an improvement.
- 5:54:35
A Weekly Live Community Discussion
A tentative proposal to hold a weekly live session, if the community grows, where members could join a voice-and-video discussion through a community platform.
- 5:57:17
Model Configuration and Clear Deliverables
While reviewing a task configuration, the speaker reverses a proposed model change, keeps the current model at medium effort, and emphasizes making the required deliverables clear.
- 6:00:51
Rethinking Free Trials and Pricing
The speaker expresses enthusiasm about removing a free trial and changing pricing, while critically reconsidering the confidence behind an earlier pricing decision; no validated outcome is reported.
- 6:01:25
Using Codex and Claude Together for Coding
The speaker recommends trying Codex and, if feasible, using it alongside Claude. This is presented as personal advice rather than a demonstrated benchmark.
- 6:03:46
Choosing a Stream Time from Community Preferences
The speaker weighs morning streams, when he feels more energetic, against evenings, which most community poll participants preferred, highlighting the tradeoff between presenter energy and the preference expressed by most poll participants.
- 6:04:42
Designing a Useful, Engaging Competition
The speaker is seriously considering a competition intended to be enjoyable, attention-grabbing, and genuinely useful; its subject and prize had not yet been decided.
Got a question about something in this stream?
Ask in the group