How to Automate Software Engineering
a year ago
- #AI
- #Software Engineering
- #Automation
- AI models are improving but still assist rather than fully replace software engineers.
- Current AI lacks reliability, long context performance, and broad agentic capabilities.
- Human learning algorithms are more efficient than current AI learning methods.
- Massive compute and data are needed for AI to develop general capabilities.
- Automating software engineering requires training on human data and reinforcement learning.
- Reinforcement learning environments are currently too limited for full automation.
- AI will first change software engineering jobs before potentially replacing them.
- Full automation requires AI to handle all responsibilities of a human engineer.
- Software engineering may be one of the first and last white-collar jobs to be automated.