When it comes to AI, the conversation has shifted from “if” to “how” and “when.” At MILL5’s Medellin Tech Talk, we brought together technical leaders, Richard Crane, Jorge Armando González Pino, Camilo Jaramillo, and Luis Robles, for a candid discussion about the realities of AI adoption, trust, and the global competition shaping our technological future.
The Model Wars: What's Actually Working?
The panel opened with a revealing question: what’s your favorite AI model? The consensus was striking. While ChatGPT dominates everyday use, Claude emerged as the unanimous winner for coding tasks. The reason? Anthropic’s unique approach to training data – purchasing, scanning, and processing physical books to feed more refined language structures into their models.
As noted during the panel discussion, Claude successfully completed a single-page application in one attempt, while four other leading AI models failed the same task. This kind of real-world performance matters when you’re iterating 161 times to get an application working correctly.
Beyond Search: The Death of Google?
Perhaps the most provocative topic was the decline of traditional search engines. Multiple MILL5 panelists admitted they’ve largely abandoned Google, with one describing their journey in an essay series called “Escape from Google.” The shift isn’t universal – concerns remain about AI hallucinations and the “curve of exactitude” – but the trend is undeniable.
The implication for enterprise businesses? Products built on search engine infrastructure may need fundamental rethinking.
Trust, Privacy, and Nation-State Concerns
Our panel discussion didn’t shy away from uncomfortable truths. Trust in AI systems remains fragile, and as noted, “once you lose it, it’s almost impossible to gain it back.” The comparison to the MOQ library scandal, where a developer harvested user emails, served as a stark reminder of what happens when trust breaks.
More concerning: the assumption of privacy. Users are sharing relationship advice, health issues, and personal secrets with AI systems, often not realizing this data isn’t truly private. As the panel bluntly put it: “If you wouldn’t tell the FBI, don’t tell it to AI.”
The discussion also touched on geopolitical tensions, with the U.S.-China AI race dominating the landscape. While the U.S. currently holds advantages through major tech investments, the panel expressed hope for a more collaborative, democratized future.
The Job Question: Transformation, Not Elimination
Will AI take your jobs? The panelists discussed recent projections suggesting 60 million jobs will be replaced while 70 million new ones emerge. The key insight: it’s about transformation, not elimination.
Interestingly, the conversation touched on a growing trend of white-collar professionals investing in blue-collar businesses like plumbing and electrical companies as “AI insurance.” But the panel’s recommendation was clear: don’t insulate yourself from AI – master it.
The Technical Reality: AI's Kryptonite
When asked about AI’s biggest weakness, answers ranged from power consumption and infrastructure limitations to data quality concerns. The panel highlighted how AI models trained on internet data, including increasingly AI-generated content, may face a “garbage in, garbage out” problem at scale.
Water usage at massive data centers and energy requirements also emerged as infrastructure challenges that will need to be solved as AI scales.
Democratizing Access: The Path Forward
The discussion concluded with open-sourcing as the key to AI democratization. Meta’s Llama model release (even if initially unintentional) sparked community innovation in ways that proprietary models couldn’t. However, the panel acknowledged a troubling reality: only the biggest companies can afford to compete in the foundation model space, suggesting inevitable government intervention to ensure universal access.
The Bottom Line
This wasn’t a panel of AI evangelists promising utopia. It was technical leaders sharing hard-won lessons about what works, what doesn’t, and what keeps them up at night. From the 161 iterations needed for a “simple” app to the acknowledgment that even advanced AI makes junior developer mistakes, the message was clear: we’re still in the early innings.
For technical leaders, the takeaway is actionable: experiment across multiple AI platforms, understand their strengths and weaknesses, maintain healthy skepticism, and most importantly – start building expertise now.
Watch the Full Discussion
The complete panel discussion offers deeper insights into AI tool selection, coding workflows, privacy considerations, and the technical challenges ahead. Watch the full video here to hear unfiltered perspectives from practitioners in the trenches of AI adoption.
Interested in how MILL5 can help your organization navigate AI implementation? Contact us at ai@mill5.com to learn more about our approach to practical, enterprise-grade AI solutions.
