Ads, Data & AI Marketing Agents
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 follows the connected work of building and marketing a Shopify app: testing it in a real store environment, seeking additional perspectives and user feedback, monitoring speed and bugs, and making the product’s value quickly understandable. Much of the discussion concerns advertising readiness and measurement—reviewing creatives, correcting social-page branding, consolidating platform metrics into a data pipeline, and choosing a conversion event that balances signal volume with proximity to the real goal, such as installation or completed onboarding. Luqman argues that creative intuition alone is unreliable and that testing, market response, and data matter more, while stressing that trials need time to mature and that the campaigns had not yet demonstrated success. The group also discusses Meta’s shift toward broad targeting, “learning limited” campaigns, lead-capture funnels, and attribution, leaving questions such as the best optimization event, the value of lookalike audiences and affiliate marketing, and parts of tracking and implementation unresolved. The stream also presents experiments and proposals involving Claude, Codex, MCP, and specialized agents for campaign setup, ad review, cross-session strategy handoffs, content production, SEO, and social publishing. Luqman emphasizes human review because AI recommendations can become inconsistent, and describes designing an agent that challenges assumptions instead of automatically agreeing. Audience questions cover possible Arabic–English differences in token use and coding quality, the limits of using AI for advertising without marketing fundamentals, and how to choose a SaaS idea or learning specialization. Broader discussions address market and competitor research, unit economics, the difficulty of customer acquisition, and the value of starting with a focused market or specialty. They also include explicitly uncertain forecasts about autonomous agents, robotics, optional work, and large one-person companies, followed by personal—not universal—views on partnerships, accountability, disagreement, and the social effects of economic pressure.
87 moments from the stream
- 0:56:54
Trying to Create a Meta Ads Campaign with Claude and MCP
The speaker describes a new experiment in which he prompts Claude to handle ad-campaign setup through MCP, then reports during the work that the campaign was created. This reflects his live account of the experiment, not independent verification of the result.
- 0:57:10
Choosing an Optimizable Conversion Event in Meta Ads
The speaker says that, according to his understanding at the time of recording, campaign optimization requires roughly 50 conversions in under seven days. Because an install event may be difficult to reach at that volume, he proposes choosing an earlier but still meaningful event, with a fuller explanation deferred.
- 0:58:55
Risks of Unofficial Ad-Account Automation
Responding to concerns about bans, the speaker tentatively distinguishes official or contracted AI integrations from unofficial third-party automation, suggesting the latter may expose an account to enforcement. This is his interpretation, not confirmation of a specific Meta policy.
- 1:00:14
Using Claude with Human Review and Choosing a Prompt Language
The speaker says Claude has made much of the work easier, while stressing that its output still needs human review. He also relays Claude’s answer that English prompts may be slightly cheaper than Arabic prompts; this is presented as a reported answer, not an independently verified test.
- 1:04:05
Turning Project Context into an English Claude Prompt
An audience member describes using a tool to understand a project, produce an English prompt, and then execute it with Claude. The speaker notes that others use a similar workflow.
- 1:07:16
Holding Back Ad Guidance Until Value Is Verified
Answering a question about ad setup, the speaker said that, at the time of the recording, the campaigns had not yet been published. He therefore did not want to prescribe specific steps before practically verifying that the approach worked and provided value.
- 1:10:46
Generating Ad Components and Understanding Competition
A system is given several ads within a defined strategy to generate headlines and ad copy. The discussion notes that being able to build a comparable tool is not enough by itself to compete, because competitors may also have formidable marketing teams.
- 1:19:18
Running Advertising Tasks Through Chrome MCP with Human Review
The speaker directs a tool to open a browser through Chrome MCP and navigate to an ad accounts library, noting that these steps were previously manual. He emphasizes reviewing the tool’s work rather than accepting it without inspection.
- 1:20:29
Navigating to the Ad Library Page
A brief reference to navigating to an ad-library page, with videos mentioned but the intended action left unclear.
- 1:24:41
Forking an AI Conversation to Research Conversion Decisions
The speaker forks the conversation to ask whether it previously covered conversion optimization, reaching 50 conversions per week, and the dilemma involving cost per install and the choice of target conversion. He explains that branching allows several directions to be pursued from the same context.
- 1:29:43
Specialized Marketing Agents for Organic Traffic
An early concept for several marketing agents specialized in organic traffic, including agents for posting on LinkedIn and X and another for research. It is presented as an idea under consideration, not a completed or tested system.
- 1:32:41
Why Broad Targeting Replaced Laser Targeting on Meta
The speaker explains that, in his view, Meta targeting has shifted from narrow demographic and interest filters toward broad targeting, where the algorithm identifies likely customers from many signals and data points. He emphasizes creative assets and sufficient conversion events as learning inputs, then describes using a customer list to create a lookalike audience.
- 1:35:58
Choosing a Proxy Conversion Without Losing the Real Goal
The speaker discusses the difficulty of optimizing a campaign directly for an install and considers an easier event, such as a page visit or demo-store view, as an intermediate signal. He cautions that optimizing for a proxy event may attract the wrong behavior even if Meta can find people likely to perform it; the final objective remains the install.
- 1:38:10
Apollo as a Lead Source and the Limits of Data Collection
Responding to an audience suggestion, the speaker describes Apollo as a useful source of lead data and considers whether its records could be collected automatically. The discussion leaves the method’s permissibility and ethics unresolved and does not establish that it was tested or successful.
- 1:40:33
Should Meta Campaigns Accept “Learning Limited”?
The speaker challenges the idea that campaigns can simply remain “learning limited,” arguing that conversion volume may affect stability and scalability. He also asks, without reaching a conclusion, whether lookalike audiences built from externally sourced leads could improve the campaign’s starting point.
- 1:43:41
Comparing Conversion Funnels and Tying Spend to Performance
The speaker compares a lower-friction funnel for capturing interest and email addresses with a heavier conversion funnel, supported by a limited nurture sequence intended to encourage installation without excessive messaging. He proposes comparing their performance and directing spend according to observed acquisition or install cost, while acknowledging that market learning takes time and outcomes are uncertain.
- 1:49:24
Designing a Marketing Agent That Challenges Assumptions
The speaker explains that he instructed an AI marketing agent to disagree directly and clearly when it sees a marketing decision differently, rather than defaulting to agreement. He then evaluates whether the agent actually fulfilled that role during the “learning limited” discussion.
- 1:59:01
Navigating Facebook Privacy and Business Settings
While attempting to configure Facebook, the speaker criticizes the privacy-settings interface and considers an alternative path that begins inside Business Settings to locate user assignment. The captions do not record a completed resolution.
- 2:00:35
Critically reviewing Claude’s paid-ad recommendations
Luqman weighs different campaign funnels and optimization choices, observing that Claude’s spending-allocation advice became inconsistent when challenged with further objections. He argues that paid-ad decisions are not straightforward and that AI recommendations require careful human review because ad spend is an investment.
- 2:03:32
Preparing social pages before launching ads
While reviewing campaign readiness, Luqman flags outdated Facebook and Instagram page logos that should be fixed before the ads launch, highlighting brand consistency as part of a launch checklist.
Got a question about something in this stream?
Ask in the group