define AI Engineer (AI column)
I watched with great interest. speech Greg Brockman at AI Engineer World's Fair 2025. Greg is one of the co-founders and president of OpenAI, who once started at Harvard, then moved to MIT and then dropped out to join Stripe as his fifth employee. K 2013 He became the first technical director of Stripe, and 2015 He left Stripe to co-found OpenAI with Sam Altman and Ilya Sutskever.
The presentation was filled with interesting ideas, including the following: 1. Philosophy of Learning and Independent Development Greg believes that self-learning is the key to success. He believes that such independent learning can lead to exponential growth, and he drew this conclusion from his experience. (He took a three-year course in mathematics in eighth grade.). By the way, I completely agree with this - I always studied what I was a fan, even if I did not know how I would use it later. 2. Transition from Mathematics to Programming At first, Greg planned to become a mathematician, but then he discovered the magic of programming after creating a simple table-sorting widget. The trick is that he decided to move from theoretical mathematics to practical programming, where not three people will read the theorem, but millions will benefit from the software he created. 3. Engineering and Research Partnership So this is a really cool idea: the differences in perceptions of system constraints between engineers and researchers created friction. Conventionally, an engineer thinks about abstractions and APIs for the interaction of different parts with each other. And the researcher tries to think about the quality of the end2end, so he always tries to get into everything. Therefore, it is important to combine engineering skills and research ideas. (In OpenAI, this was done.). The solution was technical modesty—the ability of engineers to listen, understand context, and then make changes to architecture. 4. Scaling Laws and Basic Research It's also an interesting idea that basic research has become important again. After a period of simple scaling of transformers, OpenAI is now operating at a level where algorithms have once again become a critical constraint on progress. 5. Vibe Coding and the Future of Development Vibe coding is an old term about creating software through natural language using AI. Greg sees this not only as an empowering mechanism, but also as a future where agents work in the cloud, connecting to different systems even when your laptop is closed. 6. Transformation of existing applications It is also a cool idea that we not only have the creation of new software, but already have a lot of software, which will have a deep modernization. (Moving from one language to another, updating libraries.). Separately, Greg noted that in order to maximize the impact of AI, it is often necessary to change the approach to how we make systems. Conventionally, we used to make them as convenient as a person, and now we need to take into account the convenience of LLM models.
There were some interesting predictions. 1. The Future of AI and Economics Greg predicts a world where AI-driven economies will not just replace humans with machines, but rather a multitude of specialized agents using different models with different cost and performance compromises. 2. On scaling and infrastructure When asked by Jensen Huang about the future of AI infrastructure, Greg said that the main limitations for GPTs are:6 Computation, data, algorithms, energy and money are in that order of priority. 3. The role of engineers in the age of AI Greg stated emphatically, “AI agents do not replace software engineers. They greatly enhance their individual capabilities. 4. On AI Security and Iterative Deployment Greg believes the only way to ensure security is to continue deploying more powerful models as they are developed and learn from each deployment. That's what OpenAI does.
#AI #Engineering #Management #Leadership #ML #Software