Story perspectives

AI-Driven Coding Workflows Could Revolutionize Scientific Research

3/24/2026

Summary

Story summary

Siddharth Mishra-Sharma discusses applying multi-day agentic coding workflows in scientific computing, enabling AI agents to autonomously tackle tasks such as a cosmological Boltzmann solver.
The workflow uses a progress file, a test oracle for validation, and Git for coordination.
Agent-driven development can significantly shorten complex tasks, compressing months into days.
The solver is not production-ready, but the example shows AI's potential to accelerate scientific research.

1 of 1

Story summary

Siddharth Mishra-Sharma discusses applying multi-day agentic coding workflows in scientific computing, enabling AI agents to autonomously tackle tasks such as a cosmological Boltzmann solver.
The workflow uses a progress file, a test oracle for validation, and Git for coordination.
Agent-driven development can significantly shorten complex tasks, compressing months into days.
The solver is not production-ready, but the example shows AI's potential to accelerate scientific research.