Angelina YangFounder and Host, TwoSetAI · AI builder for visibility and discoverability

Why You Don't Need That Many Engineers Anymore? | Simon Du, Chief Scientist @Apodex

13 Jul 2026 · 1 min

Watch on YouTube

1:03

Simon Du argues that LLMs are more predictable than humans, making them easier to coordinate into autonomous agent networks that can tackle hard scientific problems. Building powerful agent systems doesn't require training massive models; existing APIs are enough. Self-evolving loops could accelerate research by an order of magnitude.

Chapters

Large language model is still less random compared to humans. Large language model listens to you, like follows your instruction. The humans are not. So it's still easier to organize 100 agents than 100 people. We are not just building a chatbot which can have a nice conversation with you. What we really want to do is to solve very hard problems, scientific problem, economic problem, something that can create a lot of value. Usually it is like an unsolved problem. So we are trying to use our heavy-duty solver to solve the hardest problem in developing larger models. They are, you know, creating a loop. They are self-evolving to become a stronger and stronger model. AI can accelerate this work 10x or 100x. You don't need that many people to do the same kind of work. If you have some open source project, especially agent, it's very hard to get resource to train a large model. But for writing agent framework, you only need APIs. You can still build a very comprehensive, strong agent framework just using API.