What’s next for AI agentic workflows by Andrew Ng (AI Ascent by Sequoia Capital)

I came across a fascinating video from Andrew Ng discussing the future of AI with a focus on agentic workflows. Unlike the current non-agentic workflows where we spoon-feed instructions to AI models, agentic workflows allow these models to act more independently, tackling tasks on their own. A great analogy mentioned by Andrew Ng is that non-agentic workflows operate as if you are writting an essay without the option to use backspace (which works remarkably well for LLMs), but what works even better is if you give them the option to contemplate on their results and provide updates (agentic workflows).

The video dives into a specific case study where an AI team used an agentic workflow to analyze data. The results were apparently mind-blowing! The agentic approach achieved significantly better outcomes compared to traditional methods. This got me really fired up about the potential of agentic workflows. Imagine AI models that can take initiative and solve problems without needing constant hand-holding!

Overall, the video left me feeling optimistic about the future of AI. Agentic workflows hold immense promise for boosting the efficiency and effectiveness of AI models. It’s important also to keep in mind that the next version of all major LLMs is focusing on improving agentic workflows and this is where we are going to see the biggest progress this year.