- AI Biz Hour
- Posts
- AI Agents, Flowith 2.0 vs. Manus Showdown & The Cultural Implications of Language Models
AI Agents, Flowith 2.0 vs. Manus Showdown & The Cultural Implications of Language Models
AI Biz Hour June 10th 2025 episode 152
AiJohn Allen
June 11, 2025
TODAY'S HIGHLIGHTS:
JanCarlos the Clarity coach shares how Flowith is "10-50 times more powerful than Manus" for document creation
Dominic Romano reveals critical findings on OpenAI embedding inconsistencies
Deep dive on cultural representation in AI models and the risk of "cultural whitewashing"
Technical discussion on agentic programming tools and workflows
SEO strategies in the age of AI and the human intuition factor
INTRODUCTION:
In today's packed AI Agent Tuesday session, our community explored the tools, techniques, and implications of building AI agents. From practical tool comparisons to philosophical discussions about AI's cultural impact, the conversation highlighted both the technical capabilities and ethical considerations shaping the future of AI business applications.
MAIN INSIGHTS:
Understanding AI's Black Box Nature
Andy shared a critical perspective on AI development:
"If you have an assumption that you're going to understand how the LLMs work and that you can't get really rolling until you understand how the black box works, I think we're going to be in for a long and torturous journey." - @andywergedal
He emphasized that LLMs are self-learning systems that organize information differently than humans do, making them fundamentally different from traditional software development:
"Development is about the output. LLMs and artificial intelligence are about the output. The method in which we get there to some degree is important. But also, to some degree, is not." - @andywergedal
AI Agent Tools Comparison: Flowith 2.0 vs. Manus
JanCarlos delivered a comprehensive breakdown comparing two popular AI tools:
"Flowith 2.0 is just 10 times, almost 50 times more powerful than Manus. Timo actually finally tested out Flowith 2.0, and he was impressed himself." - @JanCarlos
Flow With Advantages:
Superior context handling for multi-day projects
Canvas-based interface with vertical/horizontal visualization
Ability to handle online research in a structured way
Can maintain context over days or weeks
Pricing tiers from $20 to $500, with early adopter benefits
Practical Example: JanCarlos used Flowith 2.0 to prototype a book he'd been working on for two years:
"I wrote about 1,000-2,000 words per chapter across 13 chapters. I gave that plus another thousand words of context to Flowith 2.0, and it produced about 50,000 pages of text split into 3 volumes." - @JanCarlos
Manus Strengths:
Offers 300 free credits daily (equivalent to one substantial task)
Strong document creation capabilities
Effective for marketing reports and proposals
More accessible entry point for beginners
$40/month for premium tier with 4,000 credits
JanCarlos noted that while he still needs to edit about 70% of the Flowith 2.0-generated book content, it's dramatically more efficient than starting from scratch.
Prompt Engineering Strategies
JanCarlos highlighted the difference between detailed and conversational prompting:
"Detailed prompts are very good if you're looking for routine outputs, like documents or any kind of prestige rule piece of paper. But when it comes to the more creative things, or if you're trying to prototype anything or be exploratory, the story kind of prompting does work a lot better." - @JanCarlos
He explained that conversational prompting enables the LLM to gain context about what you're trying to accomplish, rather than being limited by your initial understanding of the problem.
Looking to tap into the $7 trillion government contracting market? GovBidMike helps businesses secure government contracts and grants. With important AI procurement rule changes coming in October 2024, now is the time to position your business. Mention AI Biz Hour for a 5% discount on services. Government contracts increasingly specify American-made AI technologies and interoperability requirements. Visit biddata.ai to learn how to navigate the complex world of government procurement.
SEO and AI Agent Limitations
John shared his experience exploring AI agents for SEO:
"I went into exploring AI agents for SEO about a month or two ago, and it just wasn't there, because there's a lot of nuances. You've got to be able to scrape and compare competitor sites, and then come up with content that's not readable like 100% AI." - @AiJohnAllen
He described his traditional SEO process, which involves:
Analyzing top competitors' websites
Assessing their online presence beyond websites (business listings, social media)
Gathering and analyzing information
Allowing for human intuition to arrive at an "aha moment"
John questioned whether AI could replicate this intuitive process: "I don't know how I can get AI to have that aha moment."
Cultural Representation in AI Models
Dominic Romano, speaking from Frankfurt, shared critical insights on how AI models represent (or fail to represent) non-Western cultures:
"When you bring in technology from another culture, and your citizens rely on it without representation of your own culture, it leaves cultures exposed to whitewashing—which is eradication." - @Dominic
Key Points:
70% of internet content is in English, creating inherent bias in training data
Translation tools often strip cultural context from non-English languages
Silicon Valley companies rarely participate in international AI policy discussions
The nuance of regional dialects and mixed-language communication is lost in current models
Real-World Example: Dominic shared an experience from Morocco where ChatGPT's translation between Arabic and English stripped away significant cultural context and nuance from the original Arabic. He also described villages in the Middle Atlas Mountains where people seamlessly blend Arabic, French, and Amazigh (an ancient tribal language) in conversation—something no AI model currently understands.
Debate on Solution Approaches: Andrew Altshuler and Dominic engaged in a fascinating debate about whether the solution lies in:
Model weights (Dominic's view) - The fundamental understanding must be in the model itself
Retrieval augmentation (Andrew's view) - Models can retrieve cultural information when needed
Dominic argued: "If that NLP process is trained entirely from the importation of Western culture, then how is it going to properly know how to synthesize cultural details and language from another culture?"
Technical AI Developments
Dominic Romano shared a significant technical discovery about OpenAI's embeddings:
"We detected today that OpenAI embeddings aren't consistent. We proved it mathematically by doing distance calculations on dozens of chunks from PDF files. It's pretty scary because people are ignoring that." - @Dominic
This discovery has significant implications for AI applications relying on consistent vector representations, particularly in retrieval-augmented generation systems.
The group also discussed:
N8N and Make.com alternatives (a list was shared in the Purple Pill community)
The challenges of source control when using no-code AI platforms
The importance of understanding the underlying architecture before relying on AI tools
Dominic also mentioned ongoing issues at the Defense Innovation Unit (DIU), noting "a lot of chaos there" with "unhappy people going back and forth about DIU lately."
The Power of Community Collaboration
Michael offered to help build an automation system for the AI Biz Hour using his infrastructure:
"I've got 3 V100s running, so I can run a 70 billion parameter model locally. We won't have to go out to the cloud to build our proof of concept. And I also have a 5GB pipe in and out of my house." - Michael
This exemplifies the collaborative spirit of the AI Biz Hour community, where members contribute their expertise and resources to advance shared goals.
Michael also described his approach to helping with AI Biz Hour automation:
"I have no assumption on how we're gonna solve this problem. What I'm gonna do is I'm gonna sit down and I'm gonna watch, because I don't want to use tech for tech's sake. I mean, it's cool to use the employees, but sometimes the toys aren't the solution." - Michael
Community Development & Future Plans
John Allen highlighted the growing inner community supporting AI Biz Hour:
"We have kind of an inner community that's helping to promote the AI Biz Hour, because we all believe that there's a larger audience for us. And the podcast is the first way to do that." - @AiJohnAllen
He mentioned that the AI Biz Hour has now recorded 152 episodes and is working to transform into a media company, with plans to mine previous episodes for valuable insights.
QUICK HITS:
JanCarlos is starting a private Telegram group (limited to 20 spots) for AI beginners called "AI and Clarity"
Detailed prompts work best for routine outputs, while conversational prompting works better for creative tasks
The best AI applications focus on the output quality rather than understanding how the "black box" works
Cultural context is crucial for accurate language processing, especially with mixed-language communication
Michael uses Claude 3 (Anthropic) for its writing quality, finding it "head and shoulders above the others"
Andy advocates for Pareto's Law in AI development (the square root of contributors produces most value)
RESOURCES MENTIONED:
Flowith - Advanced AI workflow platform with canvas interface
Manus - AI document creation tool offering daily free credits
George Lakoff's "Women, Fire and Dangerous Things" - Book on language and cultural categorization
MyAITutor's Multibot - Testing platform for agent concepts
N8N and Make.com alternatives list (shared in Purple Pill community)
COMING UP:
Tomorrow's AI Biz Hour - Live at 12 PM ET
Friday Inner Circle Meeting - By invitation for regular community contributors
Continued discussions on AI agents and cultural implications
OUR SPONSORS:
Gov Bid Mike
Access to $7 trillion government market opportunities
AI procurement rule changes coming October 2024
10% discount for AI Biz Hour listeners
Focus on American-made AI technologies and interoperability requirements
Virtual Synergies
$9/hour virtual assistant services
English-speaking, trained professionals
Ideal for handling emails, calls, and administrative tasks
CONNECT WITH AI BIZ HOUR:
Website: aibizhour.com
Andy: @andywergedal
John: @AiJohnAllen
Show: @aibizhour
CALL TO ACTION:
Join us tomorrow at 12 PM ET for another live session exploring AI business innovation. Subscribe to the newsletter for detailed insights and follow our hosts for daily updates on AI developments. If you're interested in JanCarlos's AI and Clarity Telegram group, check the link in his bio - only 20 spots available!
Reply