What Happens If This Works?
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
Luqman explores an early open-source tool concept that starts with user goals and coordinates agent teams to build and operate projects, potentially through voice. Discussion covers agent harnesses, memory, knowledge and context graphs, RAG, and custom development versus learning from existing tools such as Hermes. Feasibility and the eventual interface remain unresolved. Practical work includes repeated reviews of a bilingual teaser page, clearer messaging, bird animation, and audience polls to choose a name, without a final naming decision. Domains are discussed separately, with availability checks reflecting the time of recording. Blogging and SEO agent work includes requests to review articles while keeping daily publishing paused. A comment-reply system is described as retaining human approval. The discussions do not establish completion of these improvements. Audience questions examine token limits versus quality and when agents should retry or seek help. The discussions also examine Luqman’s requests for clearer roadmaps and verification before approval, practical evaluation of Codex and Claude, model-intermediary privacy, sustained use as demand validation before monetization, branding, marketing, and web-service pricing. Luqman reports that a Shopify app competitor was acquired within the two weeks preceding the stated recording date, October 16, 2026. He interprets this as a possible demand signal; no confirmed acquisition price is provided. The discussion includes a simplified explanation of IPOs and financial disclosure, using a hypothetical Anthropic listing as an example. Predictions about future interfaces and AI pricing after IPOs are explicitly uncertain, while some privacy and performance questions remain unverified. Luqman encourages proactive self-directed learning and asking questions, mentioning courses as a structured option without recommending a specific platform. An audience contribution describes a site being polished that collects tech and AI news and features skills, tools, and open-source apps. Discussion touches on making the project open source and monetization, without enough clarity to establish a revenue model. The discussion also covers disk space repeatedly filling after cleanup, keeping local recordings for higher-quality short clips, and ongoing background work on livestream transcription. No storage diagnosis or fix is established, and the transcription implementation is unexplained.
115 moments from the stream
- 2:13:42
Context graphs and harness engineering: planning and constraints
The speaker asks to explore the technical stack and harness strategies, recalling an earlier discussion of context graphs, knowledge graphs, and RAG. In his view, excessive constraints can reduce AI output quality, and harness engineering requires experience. He also proposes initially free access and adding tools later.
- 2:20:14
Clarifying project scope and the need for a roadmap
After expressing confusion, the speaker requests a concise, plain-language explanation of the work, its goal, and the project scope understood so far. He later argues that even a small project needs a substantive roadmap.
- 2:23:02
A visual roadmap to track project direction
The speaker argues that a concrete visual roadmap, such as a swimlane diagram, helps track project direction and avoid losing focus, especially when working on several projects concurrently.
- 2:25:56
Clarifying a tool’s goal before planning its first version
The speaker proposes clarifying the goal while leaving creative latitude, then using probing questions and comparing and combining planning approaches to develop a first-version roadmap. The intended scope includes maintenance and sustainable operations alongside launch readiness; the captions do not establish that the plan was implemented.
- 2:33:16
Requesting shared browser context in Codex
Asks the assistant to access Codex’s in-app browser and see the same page without screenshots. The captions do not establish whether access succeeded.
- 2:37:13
Clarifying the vision before architecture review
Proposes using a grill-me session to clarify the vision, then having AI sessions review goals and repository architecture notes to determine how to build it. This is a proposed workflow; no completed review outcome is shown.
- 2:38:20
Goal-driven AI with fewer clarification loops
Luqman outlines a speculative 2027 vision in which AI continuously pursues broad goals, such as improving conversion, while reducing unnecessary exchanges and asking only consequential questions. He connects harness engineering with a lower learning curve and inferring standard software-planning needs. He considers moving beyond prompting and conventional interfaces, while acknowledging uncertain user acceptance.
- 2:43:10
A vision for goal-driven AI software development
The speaker proposes a tool that handles project foundations, including an entity relationship diagram, from a goal instead of detailed instructions. He also explores moving beyond IDEs, graphical interfaces and command lines, mentioning voice as a possibility while leaving the replacement interface unresolved.
- 2:50:40
Disk space repeatedly fills after cleanup
The speaker reports recurring low disk space that returns quickly after cleanup and asks to address its cause. This passage establishes neither a diagnosis nor a fix.
- 2:53:04
Accepting an invitation and returning to the dashboard
A brief confirmation of signing in, accepting an invitation and returning to a dashboard. The captions do not clearly identify the service.
- 2:55:29
From development environments to goal-driven, voice-controlled agents
The speaker discusses the shift from traditional development environments toward chat and agent interfaces, referencing Cursor in 2025 and tools such as Claude and Codex. He imagines defining goals and delegating work through voice, with less typing and less reliance on terminals, presenting this as a future vision.
- 2:56:47
Choosing a domain extension and its alternatives
Answering an audience question, the speaker says his preferred domain extension is not mandatory and mentions alternatives and potentially costly options. The extension names are unclear in the captions, so his exact recommendation cannot be identified.
- 3:03:39
Future agent interfaces and the role of IDEs
Luqman expects chat to become less central without disappearing. The discussion considers voice interaction and why IDEs may remain useful for developer control; these are speculative views.
- 3:06:41
Token limits and usage estimates before execution
Responding to a question about agents consuming many tokens, Luqman proposes usage limits and a harness that estimates consumption beforehand. This is a proposed approach, not a demonstrated implementation.
- 3:07:56
An intent-based interface for developers too
Luqman wants a tool that lets users explain their intentions and reduces the work involved while allowing developers to go deeper. He argues that the idea should also serve experienced developers.
- 3:09:25
Build a custom harness or reuse an open-source solution?
Luqman explains a harness as giving a model arms and legs. In his view, modifying the capabilities supplied by tools such as Codex and Claude can reduce quality without sufficient expertise. He discusses investing time in a custom harness versus reusing an open-source solution.
- 3:12:40
Model access intermediaries and data retention
An audience question about accessing a model with a provider key leads to discussion of intermediaries such as OpenRouter and zero data retention (ZDR). Luqman expresses uncertainty about data routing; the passage does not establish providers’ retention or content-access policies.
- 3:14:09
The burden of supervising and managing assistants
The speaker argues that initial enthusiasm for assistants can give way to burdensome supervision and management, and mentions a system that creates and activates them.
- 3:15:00
Choosing an Arabic name: meaning, associations, and brevity
The discussion weighs Arabic name candidates through meaning, personal associations, and brevity. One candidate is explained as successive groups of birds according to the cited interpretation. The speaker expresses a personal negative association with that word and later favors a shorter alternative while the poll remains open; no final choice is established.
- 3:16:49
Domain availability and brokerage comparisons
The speaker explains that changing registrars does not make an already registered domain available and suggests comparing brokerage services when trying to acquire it. An alternative extension is also checked and described as available at the time of the discussion.
Got a question about something in this stream?
Ask in the group