The head of Nvidia on the All-In podcast shared his views on where the industry is heading technically and how the development process will change:

⚙️ AI as a new operating system
Huang defines modern AI not as a model, service, or library, but as a full-fledged personal computer. It is literally a new OS that can manage resources, handle scheduling, has I/O subsystems and APIs for running multiple applications.

🔐 Agentic software and security
The industry is moving toward systems that can independently execute code, access databases, and communicate with the outside world. Huang emphasizes: when you have such agentic software, security becomes paramount. The architecture must provide hardware and software guarantees for privacy protection and strict process isolation when the agent accesses sensitive information.

📈 Exponential growth in computing
The evolution of AI requires a multiplicative increase in computing power at each stage:
▪️ The transition from basic generation to reasoning models required roughly a 100x increase in compute resources.
▪️ The transition from reasoning models to full-fledged agentic systems will require another 100x in computing power on top.

👨‍💻 New programming paradigm
The very process of creating software is fundamentally changing. According to Huang, writing code by hand is becoming a thing of the past. In the new reality, an engineer's tasks will boil down to the following:
▫️ Designing system architecture.
▫️ Writing precise specifications and ideas.
▫️ Organizing "teams" of AI agents and routing tasks between them.
▫️ Defining evaluation metrics — strict formalization of what constitutes a successful task completion for a given system and what is an error.

🤑 And most importantly
Huang claims that tokens are an investment in productivity. And if an engineer with a $500k salary does not spend AT LEAST $250k on tokens — that's a red flag.
"Who benefits?" — you may ask?

We're ready to bet half our salary that in some time we'll be spending half our salary on AI tools.