"Agentic workflow" is what's next for AI, says Andrew Ng. Agentic meaning the AI acts as an "agent". Basically, ChatGPT gives you one-off answers. You ask it to write something for you, or write some code, and you can have follow-up conversation, but each time it has to generate a new response.
In the "agentic" workflow Andrew Ng envisions, the AI agent has a work item that it works on in an iterative manner, interacting with you at each step. If you ask a human programmer to write some code, they never blast out the whole thing right off the bat. They write some code and then iterate on it until they get it right. By changing language AIs into full-fledged agents, they will be able to engage in this practice themselves. An AI agent tasked with writing code can run and test its code, it can write its own unit tests, it can engage in self-reflection, and so on.
The next step after that is mult-agent collaboration. In this case you could give it a high level task, and an agent can do high-level planning, and other can search on HuggingFace for an AI model appropriate for the task, and another can write the code and so on.
What's next for AI agentic workflows ft. Andrew Ng of AI Fund - Sequoia Capital