3,000 developers gather to ask: What happens to software engineering when AI writes the code?
More than 3,000 software developers convened in San Francisco on Tuesday at AI Dev 26 x SF, a conference organized by DeepLearning.AI, to confront a question no one can fully answer: what will software development look like in five years?
Jonathan Heyne, COO of DeepLearning.AI, framed the core problem plainly. "The bottleneck is our imagination," he said. For decades, writing code was the constraint. Now it isn't.
Speed becomes currency
Anush Elangovan, corporate VP of AI software for AMD, highlighted the industry's accelerating pace. "Speed is the moat," he said, arguing that the ability to move quickly now separates winners from the rest.
He presented AMD's work on ROCm, the company's open software stack for optimizing AI workloads, including projects like HotSwap and a native HIP backend for llama.cpp.
Quality matters more than frontier progress
Marc Brooker, a VP and distinguished engineer at AWS, offered a contrasting view. "The opportunity for agents is limited by the defect rate," he said.
Brooker, who has written production software for 30 years, called this "the most exciting time in my career." But he cautioned that reducing errors matters more than pushing the technical frontier forward.
He pointed to AWS projects aimed at enforcing code correctness: Hydro, a Rust framework for agents and humans to write distributed protocols; Cedar, a language for writing authorizers; and Strata, an automated reasoning tool. He also emphasized spec-driven development-giving AI models clear specifications produces better results.
"Across the industry we need to have higher standards," Brooker said.
The future looks like agent orchestration
During a panel discussion, Richmond Alake from Oracle said software development will shift toward agent orchestration and agent management. He expects roles to blur, with engineers taking on product management, design, and marketing responsibilities.
Andrew Ng, founder of DeepLearning.AI, went further. Small teams of generalists overseeing AI agents represent the way forward, he argued. And instead of having AI agents write portions of code, they should write all of it.
"If I have to review the code, I become the bottleneck," Ng said. He noted that many frontier teams are trending toward 100 percent AI-generated code.
When moderator Marina Mogilko asked panelists to rate the future of software development on a scale of one to ten, Michele Catasta from Replit said ten. Joe Reis from Practical Data Media and Dan Maloney from LandingAI both said eight or higher. Richmond Alake said seven.
The consensus: the future of software development will have considerably less actual software development.
For developers looking to stay relevant, understanding how to work with AI agents and managing their output is becoming essential. AI for Software Developers covers the practices and workflows taking shape in this transition.
Your membership also unlocks:
