Back to Warp.dev? 2x Claude + Codex
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 indexed portion of the recording centers on coding-agent orchestration. Luqman describes using Codex to supervise Claude sessions through computer use and scheduled checks, allocate tasks, hand off context, and review work. Audience discussions raise workflow visibility, lost details, and trust in reviews, particularly for scientific code; reliability and token efficiency remain unresolved. He explores a goal-driven workspace combining agent teams, memory, roadmaps, and an interface that adapts to each project. Practical work includes reviewing system architecture, extending plans beyond MVP, and requesting landing-page and hoopoe-animation revisions. The concept and revisions remain in development, alongside discussion of open sourcing; completed implementation and fixes for computer-performance problems are not established. Audience questions cover getting started with AI, programming fundamentals, hardware requirements, security, and developing skills. Other discussions examine task cost versus token pricing and model speed without a measured comparison. On product strategy, Luqman emphasizes lasting usefulness, discovering problems for a focused audience, and planning marketing and distribution. He argues that easier building does not guarantee sales and that marketing outcomes remain uncertain. Secondary topics include agents that read and reply to Facebook and TikTok comments while identifying themselves as AI. Luqman reports workflow problems, local slowness, and agents stopping when the application or computer closes, without documenting fixes. Participants also discuss voluntary creator support through tips and paid messages, affiliate marketing, sponsorships, personal branding, and professional visibility through building in public; the question of how personal branding relates to technical competence remains unresolved. Discussion also covers AI-company IPOs and profit-and-loss disclosure. At recording time, Luqman offers unverified forecasts about IPO timing and funding improving pricing and efficiency, while audience contributions suggest funding could support data centers and research. He also discusses his trust in expert forecasts and a warning attributed to Geoffrey Hinton about superhuman-AI risks, presented as a risk forecast rather than an established outcome.
160 moments from the stream
- 3:05:26
A question about MCP and agent selection of SaaS APIs
An audience member asks about building an MCP integration for SaaS APIs, then describes difficulty with an agent understanding the user's question and choosing the appropriate API, particularly the search endpoint. The speaker considers MCP creation straightforward in principle but requests more context; no specific technical solution is provided.
- 3:11:28
Polling audience preferences for stream timing
The speaker recalls an earlier poll in which the audience preferred evening streams, although he continued streaming in the morning at the time. He explains that the current evening stream responds to that audience preference.
- 3:12:02
Clarifying project continuity after a naming vote
Asked whether the roadmap belongs to a new project, the speaker clarifies that it is the same project under a name selected through a community poll. He says he had preferred other options but also likes the chosen name.
- 3:16:36
ComfyUI and hardware requirements
An introductory discussion of ComfyUI, including a correction that it is open source and discussion of GPU memory needs. It includes a participant’s report of loud computer noise and a tentative hardware comparison; the speaker acknowledges limited hardware expertise.
- 3:18:29
Reviewing system architecture beyond the task list
A request to explain the architecture and intended outcomes behind implementation milestones. The proposed design covers voice-led interaction, central control, isolated execution, durable state and memory, independent verification, continued operation with owner control, and capabilities for building and use.
- 3:21:59
Mac preference and iOS game-development tools
The speaker expresses a personal preference for working on Mac while remaining comfortable using Windows. A question follows about whether iOS game development requires a Mac, and Xcode is identified without a definitive technical answer about the requirements.
- 3:24:15
From an MVP to a full product roadmap
The review discusses a gap between a foundation for controlled execution and consistently turning broad ambition into useful results. It calls for a broader roadmap distinguishing the MVP, version one, and the full product, supported by deeper study and an additional review rather than presented as completed work.
- 3:25:15
An Anthropic reference on harness design for long-running development
The speaker recommends an article on harness design for long-running application development as a reference for Anthropic’s perspective. He considers it important despite describing it as older, and says it was published during Ramadan of the recording year; its contents are not explained in detail here.
- 3:28:10
Usage-limit reset timing and optional resets
Discussion of how knowing usage-limit reset timing could affect usage intensity, followed by a suggestion to let users accept or defer resets. The speaker speculates that mandatory resets may save the provider money.
- 3:28:36
Extending a roadmap from MVP through V1 and beyond
Review of a roadmap that stops at MVP, with a request to extend it through V1 and distinguish milestones for MVP, the first release, and subsequent development. The request recurs later in the pages.
- 3:33:52
Developing a 3D character for a brand
A conditional proposal to hire a 3D artist to develop a character if the idea succeeds, with potential trademark use. This is a future plan, with no implementation documented.
- 3:35:31
Prompting an assistant for precise alignment and measurements
A prompt asks the assistant to apply geometric expertise, maintain alignment, and check measurements carefully. No resulting implementation is shown in the text.
- 3:36:13
AI company IPOs and the potential effects of funding
An explanation of IPOs and profit-and-loss disclosure, alongside a personal prediction of better pricing and efficiency. An audience contribution suggests funding may support data centers and research. At recording time, the speaker cites October 26 for an Anthropic filing, mid-November for its debut, and 2027 for an OpenAI IPO. These dates are unverified here, and he acknowledges his predictions may be wrong.
- 3:42:05
Comparing token output speeds and exploring Fast mode
Review and discussion of token output rates for different models, including a question about an official metric. The discussion moves to a claimed Fast mode speed increase and its availability, without providing a measured performance comparison.
- 3:43:29
An input bug duplicating characters after a Claude update
The speaker reports an input bug after a Claude update: characters appear twice and pressing space produces a dot. He says an earlier debugging attempt did not resolve it and proposes restarting, with no outcome documented.
- 3:44:03
Attempting Fast mode and discussing its additional cost
An attempt to try Claude Fast mode and determine its speed increase. From an interface message, the speaker concludes it requires payment beyond the subscription and associates it with API usage. API exclusivity remains his interpretation, and no speed measurement is obtained.
- 3:45:03
Comparing model speeds within comparable tiers
The speaker argues for comparing speed between models in comparable tiers, rather than making broad speed claims without considering which models are being compared.
- 3:45:37
Task cost depends on more than token pricing
Explains that price per million tokens alone can mislead because models consume different amounts per task. Reviews average output-token usage while explicitly allowing that the statistics may have been misread.
- 3:48:18
Extending Paperclip to address tool and skill problems
An audience question asks about combining Codex and Claude with agents that select skills and MCP tools. The speaker recommends extending open-source Paperclip and fixing its integrations; no completed fix is presented.
- 3:49:43
From prompting to goal-driven agent teams
Discusses a proposed system that understands a goal and assembles agents to build, operate, and maintain it, with memory and token efficiency. The speaker sees a progression through prompting, loops, and graphs, and speculates about simpler goal-based interaction; implementation remains uncertain.
Got a question about something in this stream?
Ask in the group