EP 590: Agents, LLMs, or Algorithms? A Playbook for Choosing AI
🎯 Summary
Podcast Episode Summary: EP 590: Agents, LLMs, or Algorithms? A Playbook for Choosing AI (32 Minutes)
This episode of the Everyday AI Show, featuring Michael Abramov (CEO of KeyMaker and KeyLabs), provides a practical framework for business leaders navigating the confusing landscape of modern AI technologies—specifically contrasting traditional algorithms, Large Language Models (LLMs), and emerging Agents/Agentic AI. The core message emphasizes experimentation over rigid categorization.
1. Focus Area
The primary focus is on AI Selection Strategy: providing a playbook for business leaders to choose the appropriate AI technology (Algorithms, LLMs, or Agents) based on specific business needs and pain points, rather than chasing buzzwords. The discussion also touches upon the current state of AI news, including talent wars, copyright rulings, and API updates.
2. Key Technical Insights
- Defining the AI Stack:
- LLMs: Described simply as a “black box” capable of answering questions.
- Agents: Defined as automated workflows involving multiple models connected by hard-coded logic (if/else statements) to perform complex tasks (e.g., a travel agent coordinating weather, bank, and flight models).
- Agentic AI: A concept where a supervisor model dynamically decides the necessary if/else logic and workflow, rather than relying on pre-coded decision trees.
- The Importance of Decision Trees: The concept of “if-else” logic is foundational to both traditional algorithms and agents, representing the conditional decision-making trees inherent in any automated process, whether manually defined or dynamically generated by an agentic supervisor model.
- Data Quality as the Foundation: Regardless of the chosen AI paradigm (algorithm, LLM, or agent), the quality, structure, and accurate labeling of training data remain the most critical factor for success. Poorly interpreted or “pseudo-data-driven” conclusions are highlighted as a significant risk.
3. Business/Investment Angle
- Prioritize Pain Points Over Terminology: Leaders should ignore the hype surrounding terms like “agent” or “LLM” and instead focus on hands-on experimentation to see which technology genuinely solves their most painful, immediate business problems.
- The “Corporate Blade” Risk: Startups building wrappers or simple applications on top of foundational models (like Perplexity wrapping GPT-4) face the risk of being quickly “bladed out” or absorbed when the foundational model providers (OpenAI, Anthropic) integrate those features directly into their core offerings.
- Talent and Innovation Balance: Successful companies must balance routine work with mandated innovation time. Leaders should encourage employees to experiment with the available toolset to solve daily problems, fostering organic innovation rather than dictating specific technology usage (LLM vs. Agent).
4. Notable Companies/People
- Michael Abramov (CEO, KeyMaker/KeyLabs): Guest expert specializing in data labeling and preparing training materials for AI models.
- Amazon Web Services (AWS): Mentioned due to the departure of a key Gen AI leader (Vasi Filiman) overseeing the Bedrock platform, highlighting the intense AI talent wars.
- Meta & Anthropic: Both recently won key copyright lawsuits regarding the use of copyrighted material for AI training, establishing precedents for “fair use” in this context.
- OpenAI: Announced updates to its API, including rolling out Deep Research mode models (GPT-4/4o Mini) and webhooks for real-time event notifications.
- Gartner: Cited for a study predicting that 40% of agentic AI projects will be canceled by 2027 due to high costs, unclear value, or inadequate risk controls.
5. Future Implications
The conversation suggests a future where the distinction between technology types blurs as LLMs gain more agentic capabilities. The ultimate trajectory beyond AGI is characterized by unknown unknowns—unimaginable technological leaps (like teleportation) that current linear projections (like building bigger air balloons) fail to predict.
6. Target Audience
This episode is highly valuable for Business Leaders, Executives, and Product Managers who are responsible for AI strategy, implementation ROI, and technology adoption within their organizations. It is also useful for AI Practitioners seeking guidance on structuring internal innovation efforts.
🏢 Companies Mentioned
đź’¬ Key Insights
"Michael, we've covered a lot in today's conversation. But as we wrap up, what is the most important piece of advice or actionable insight that you have for business leaders out there that are maybe just scratching their head when it comes to algorithms, large language models, agentic AI?"
"Now, there are some unknown unknowns. A good example of it is imagine the times when people were flying on zeppelins or on balloons... They could never imagine an airplane, right? That was an unknown unknown for them."
"I'm not saying, "Hey, come up with an LLM idea or agent idea or agentic idea or something else." I'm not trying to say, "Use this or that tool." So, I'm just saying, "Hey guys, here is a whole toolset... Just see if there is a better tool than you're using today.""
"But it's nothing that ChatGPT, like OpenAI, can do in weeks, right? And if they like the idea, they can do it and just blade-cut out all of these startups that did interesting things."
"40% of agentic AI projects will be canceled by 2027, either due to high costs, unclear business value, or inadequate risk controls."
"So this pseudo-data-driven is even worse than not being data-driven at all."