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
- 1:41:35
Evaluating Agent Tools by Results and Cost
Luqman discusses comparing tools such as Codex, Claude, and Grok Bot by work quality and cost, suggesting a limited practical test before choosing. He also speculates—rather than reports—that Codex may release a bot product.
- 1:42:54
Dedicated Email for AI Agents
Agent Mail is presented as a way to give agents dedicated email addresses, potentially allowing them to be invited into platforms and services without using the owner's personal email. The benefit is framed as an idea he is eager to test, not a proven result.
- 1:44:19
Inter-Agent Communication and Personal Assistant Routines
Luqman describes a setup where a user can talk with agents and let them communicate with one another, and suggests that mobile access could make Grok Bot useful as a personal assistant. One example is a daily routine that gathers news from multiple sources and returns a summary, while he stresses that the setup is still experimental.
- 1:46:30
Agent Stand-ups, Performance Metrics, and Human Review
Luqman imagines agents holding a short daily stand-up to summarize completed work, plans, and blockers, plus a longer periodic review of departmental performance and traffic sources so effort can focus on stronger channels. He presents this as an unproven vision and stresses personally reviewing content published under his name until trust in the system grows.
- 1:47:41
Connecting Error Detection to Fixing Agents
Luqman explains an idea for using a Sentry integration so that a detected software bug can be sent to coding agents such as Codex or Claude for a possible fix. He says he already has a similar workflow elsewhere, so this is presented as an option rather than a documented successful test.
- 1:52:23
Testing Direct Communication Between AI Agents
The speaker explores whether two agents can communicate directly without him first creating a group chat, arguing that this could make multi-agent coordination more like communication inside a company. His reaction suggests the attempt appeared to work at that moment, without documenting broader validation.
- 1:57:19
Reviewing Each Agent Like an Employee
The speaker proposes examining each agent individually as if conducting an employee interview: reviewing its contract and system prompt, defining a contribution or return metric, and adding missing tools according to its role, such as for a programmatic SEO agent.
- 1:59:58
Improving Agents Through Accountability and Contribution Metrics
Rather than punishing an agent for a mistake, the speaker describes updating its contract or instructions to reduce repetition of the problem. Over time, he proposes measuring each role’s contribution and deciding whether to improve, replace, or discontinue it, while stressing that the company analogy remains experimental and may or may not produce results.
- 2:04:46
Provisioning Computers and Access for AI Agents
The speaker discusses giving each AI agent a virtual or cloud computer, potentially assigning it an email address, and reviewing plugins across the whole workforce instead of repeating the process agent by agent. Access should be planned by role, such as granting analytics access to the responsible agent.
- 2:06:29
Choosing Workflows for Agent Automation
The speaker argues that the practical value is not limited to coding and is strongest when there is a workflow suited to automation. He uses processing and publishing stream content as an example of work agents could automate, while noting possible uses in both professional and personal life.
- 2:08:24
Designing an Organization for a Team of Agents
The speaker describes starting with a clear vision and turning it into a company-like organization chart: an executive layer oversees departments, with several agents under each lead. He stresses that sound structure matters even when the team is small.
- 2:10:15
Potential Use Cases, with Uncertainty Preserved
The speaker says the clearest use case he currently sees is marketing and traffic generation. He suggests that building with the system may also be possible, but explicitly notes that he has not seen a demonstrated example and does not yet know how it works.
- 2:10:51
Options for Live Audience Participation
The speaker considers bringing participants into a live stream through Google Meet or using a Telegram live feature for a group conversation resembling a video call.
- 2:11:52
Question About the Usage-Limit Reset Cycle
A participant asks whether the usage limit resets every five hours or weekly, and the response says it is weekly. The excerpt does not establish the limit type or its other conditions.
- 2:12:47
Telegram or Discord for a Working Community?
The speaker sees Telegram as a stronger home for an engaged community than relying on YouTube subscribers, suggesting voice sessions where members work together while also considering Discord a suitable option.
- 2:15:59
Choose Grok Bot by Task, Not as a Direct Replacement
Answering whether to switch from Claude agents, the speaker says Grok Bot, Claude, and Codex serve different working patterns. In his assessment at the time of recording, Grok Bot competed more closely with OpenClaw and Hermes than directly with Claude or Codex agents; he also said that users had moved from Hermes and expected other tools to enter this pattern. In his view, Grok Bot fits continuously automated work or a digital “employee” assigned a specific goal and routine, so the choice should follow the task. He stresses that this is an early assessment that may change with more experience.
- 2:20:37
A Dedicated Virtual Machine for Each Grok Bot Agent
Asked what distinguishes Grok Bot for automation, the speaker highlights that each employee-like agent has its own virtual machine, effectively a dedicated cloud computer. The excerpt ends before the practical demonstration is completed.
- 2:22:03
Why Benchmarks Alone May Not Settle Model Comparisons
The speaker notes conflicting opinions about the strength of forthcoming models, says comparisons may be relative, and argues that the lack of a shared benchmark makes judgments harder; no firm conclusion is offered.
- 2:22:37
Grok Bot Agents: Hierarchy, Persistent Runtime, and Teachable Skills
The speaker describes an interface with agents grouped under executive supervisors and says each agent has a virtual environment that can keep running after the user’s computer is shut down. He also describes plugins, skills learned from a recorded task, and mobile access to agents. He cautions against directly comparing these agents with Codex or Claude agents because the products serve different goals, positioning Grok Bot more broadly around automation.
- 2:25:41
The Prediction of a One-Person Billion-Dollar Company
The speaker relays a prediction attributed to someone referred to as “Sam” that a company run entirely by one person could reach a billion-dollar valuation. He is uncertain about the date, first suggesting 2023 and then correcting himself to 2024.
Got a question about something in this stream?
Ask in the group