World's Fair 2024
Which Jobs Can Be Replaced Today: Fryderyk Wiatrowski and Peter Albert
Overview
This talk explores the potential for AI agents to replace human tasks, moving beyond simple prompting to autonomous operation. The core thesis is that agents can handle low-leverage, reactive tasks, freeing humans to focus on high-leverage, proactive activities. The discussion outlines a path from basic prompting to more complex agentic workflows, emphasizing the eventual goal of agents operating independently within broader workflows.
Who should watch
- AI engineers and builders looking to implement autonomous agents.
- Product Managers and designers interested in the future of AI-assisted workflows.
- Individuals seeking to automate repetitive or low-leverage tasks.
- Those exploring how AI can augment or replace specific job functions.
Key takeaways
- The future of AI agents lies not in constant prompting, but in autonomous operation similar to a reliable employee, requiring initial training and then independent task execution.
- Agents can be embedded into workflows by identifying high-leverage activities for humans and outsourcing low-leverage, noisy tasks to AI.
- Reactive jobs, such as customer support tasks involving well-defined rules, are prime candidates for early automation by agents.
- Proactive jobs, like those of a founder, are more complex but can still be augmented by agents handling the reactive components within them.
- Implementing agents involves creating triggers for agent reactions, defining rules for task execution, and potentially using browser-based actions for generic task completion.
- Building effective agents involves a progression from simple prompting to incorporating cognitive architectures, fine-tuning, and reinforcement learning, with each step increasing complexity and iteration time.
- Optimizing prompts involves matching model fine-tuning distributions, using structured formats like XML or JSON, and minimizing computational load per token.
- For complex tasks, splitting them into smaller, manageable pieces with explicit state tracking, planning, and verification steps can improve performance and quality.
Notable quotes
*The vision that we can keep in mind when building agents and thinking about how we can embed them into our workflows is how to distill the high leverage activities and preserve them for ourselves while outsourcing the low leverage things for agents.*
*The reactive jobs will go first. The beauty of the reactive things is that once you set the rules for the agent... you don't need to prompt the agents anymore.*
*You usually can start off with prompting and once things get more complex you add cognitive architectures you can add fine tuning and end reinforcement learning.*
Unofficial community note. Prefer the recording for nuance.