Building Grok Bot | 5 Executives, 16 Employees
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 experiment to build a Grok Bot agent workforce with a clear hierarchy, role charters, cloud environments, permissions, handoffs, recurring meetings, contribution metrics, and human review before publishing. Luqman prioritizes marketing and organic traffic over further feature work, discussing programmatic SEO, stream repurposing, social channels, and deliverable email outreach while warning about random content changes, sensitive integrations, and broad agent access. Audience questions prompt comparisons among Codex, Claude, and Grok Bot; he stresses that selection should follow the task, efficiency, and cost, and that his views are early impressions rather than systematic findings. Other discussions cover engaged communities, freelancing and practical AI learning, possible labor-market effects, and AI-assisted music. The impact of the marketing agents, the value of the organizational design, usage-limit consumption, and several automation workflows remain unresolved; smoke testing was not completed and the overall system was not shown to be successful.
139 moments from the stream
- 2:30:50
Reviewing Restrictions on Adding Agent Tools
While reviewing plugins, the speaker questions why the rules are cautious about adding more tools and asks for a specific explanation of restrictions involving Shopify, Gmail, and Supabase. The discussion reflects a practical tension between expanding agent capabilities and following tool-integration rules, without showing the final decision.
- 2:32:06
AI Agents for Ads and Placement Analysis
Answering a question about agents for Meta ads and advertising generally, the speaker said he had tried them and considered them effective. He explained that a tool-connected agent can quickly read analytics and help compare ad placements, presenting AI-assisted advertising as a future direction; this reflects his stated experience and opinion, not an independent test.
- 2:34:21
Risks of Agent Commands in Sensitive Systems
The speaker relayed a warning about connecting agents to sensitive systems: a single bad command caused by model hallucination could seriously damage the system. Supabase was mentioned as an example, but the captions do not establish that an actual incident occurred.
- 2:35:05
What to Learn Before Using AI for Advertising
The speaker advised a beginner to focus on advertising metric terminology, what each metric means, and clearly communicating the campaign goal and budget to the chosen tool. In his view, AI can help build campaigns on platforms such as Meta or Google, rather than making deep targeting specialization the starting point.
- 2:39:16
Using MCP with Coding Tools
In a brief answer about tool choices, the speaker said he knew of MCP support with Claude and believed Codex could also be used. The Codex statement was tentative and was not accompanied by a detailed demonstration or test.
- 2:41:36
Creating Music with AI and Python
The speaker discussed an AI-assisted music project, saying it required substantial time and effort rather than being a quick track generation, and that Python was part of the creation process. Some tool and workflow details are unclear in the captions.
- 2:42:06
Generating Hijaz-Mode Music with Code Assistance
The speaker describes using Python through Claude to prepare musical notes associated with the Hijaz mode, then passing them to Suno while specifying the intended mode.
- 2:43:35
Ensuring Agents Understand the Organizational Hierarchy
The speaker discusses making sure every agent understands its role, department, and supervising layer instead of assuming that organizational knowledge held by management is automatically shared by all agents.
- 2:49:01
Direct Messaging and Communication Routines Between Agents
The discussion covers allowing executive agents to message other agents directly, using group chats and recurring routines for coordination, and then verifying that agents already understand their reporting relationships.
- 2:51:59
Creating Music Variations and Stitching Them into One Track
The speaker requests several variations in the same musical style, with some improvisation, and asks for them to be stitched into one track without requiring a fixed duration, while referencing the Hijaz mode.
- 2:54:12
Prioritizing Traffic and Data Before More Features
The speaker says the available product should not currently chase more features; the priority is marketing, traffic, and usage because resulting data can reveal which improvements and features may help retention. This is presented as a current-stage strategy, not a claim that the product is complete.
- 2:55:42
Evaluating a Programmatic SEO Agent and Its Publishing Workflow
The speaker describes evaluating a programmatic SEO agent by checking its understanding of the job and available tools. Programmatic SEO is framed as producing keyword-related pages to improve visibility over time, followed by a discussion of automating article publishing and updates through a CMS such as Sanity or an alternative workflow.
- 2:58:49
Customer Acquisition as a Shared SaaS Challenge
Responding to a discussion about difficulty attracting customers, the speaker argues that customer acquisition is a common challenge for SaaS builders with and without marketing backgrounds, and presents the effort to address it as ongoing rather than resolved.
- 3:03:48
Economic Growth Alongside Job Loss: An Unfamiliar Scenario
Luqman describes a highly competitive, saturated market around the start of 2027, then cautiously relays expert predictions that AI could rapidly expand economic output while reducing employment. He stresses that the social consequences of such a scenario remain uncertain.
- 3:06:07
How Might Governments Respond to AI-Driven Labor Changes?
Luqman considers how governments may need to respond if economic growth and unemployment rise together, mentioning relief for citizens only as a possible example rather than expert policy advice. He warns against denying AI’s impact or resisting change and urges people to start learning now while acknowledging uncertainty about the coming years.
- 3:08:21
Gradual Model Updates and Availability
Answering a question about a new model, Luqman says it was not yet available to him and notes, based on what he believed or had heard, that the broader update might be rolling out gradually.
- 3:09:41
Clear Vision as the Foundation of a Strong AI Agent
Luqman answers a question about improving AI-agent skills by arguing that tools and technical skills are not the whole foundation: the builder needs a strong vision and must guide the agent so it reflects that vision.
- 3:10:36
Choosing a Model for Practical Efficiency
Luqman explains that when two models offer similar quality, he prefers the one that delivers greater workflow efficiency, since resource consumption that obstructs work can matter more than small quality differences. He frames this as a practical, provisional judgment amid continuing model updates.
- 3:11:40
Naming Grok Bot Employees While Keeping Roles Visible
In a brief side task, Luqman proposes giving each Grok Bot employee or agent a first name followed by its role label, adding personality while keeping the function clear and avoiding full names.
- 3:12:54
Rapid Model Releases and the Cost of Distraction
Luqman comments on intense competition between companies and the steady arrival of new models, describing how difficult it is to stay focused amid that pace and his continuing desire to avoid distraction.
Got a question about something in this stream?
Ask in the group