# AI Writing Code and Conducting Automated Scientific Experiments AI can now write its own code and perform scientific experiments. - **Source:** [Eric Jang](https://evjang.com) - **Published:** 2026-02-04 - **Topic:** [Data & AI](https://brifing.si/c/data-ai) - **Original article:** - **On Brifing:** https://brifing.si/s/d17a6a6c-9561-42ca-bdd9-6cebcdb46605 ## The AI Revolution Since 2022, advancements in AI technologies, particularly the emergence of ChatGPT, have marked a turning point. Machines can now not only write code but also solve complex mathematical problems and automate cyberattacks. This transformation is reshaping the global economy. ## Coding Capability Modern coding agents are fundamentally changing traditional programming methods. Tools like Claude Code not only write infrastructure code but also propose hypotheses, draw conclusions, and plan next experiments. This is revolutionizing the nature of software engineering. ## Automated Experiment System Tools like the /experiment command automate scientific research processes. They create folders, write Python files, and analyze results autonomously. This saves researchers significant time and boosts efficiency. ## Hyperparameter Optimization AI can automatically optimize machine learning model hyperparameters. Using techniques like maximal update parameterization, it finds the best settings to enhance model performance, far more efficiently than traditional methods. ## Sequential Experiments AI agents can conduct a series of sequential experiments, analyzing results and planning next steps. They evaluate findings after each experiment to generate new hypotheses, accelerating the scientific discovery process. ## Unconstrained Search Space Unlike traditional hyperparameter optimization systems, modern coding agents can modify the code itself. This removes constraints on the search space and enables more creative solutions. AI can assess result consistency and generate theories. ## Automated Scientists Coding agents, combined with computer use, are evolving into automated scientists. Beyond software engineering, they become general-purpose thinking machines capable of solving short digital problems, shaping the future of research. ## Mathematical Proofs AI can prove unsolved mathematical problems without human intervention. It generates new proofs for complex topics like Erdos problems, heralding a new era in mathematical research. ## Self-Improvement AI agents can speed up their own CUDA kernels to run faster. This capability leads to self-improving systems but also raises ethical and security concerns that must be carefully addressed. ## Debugging and Problem Solving AI excels at debugging and problem-solving, enabling persistent pursuit of goals. Coding REPL agents are rapidly adopted due to their relentless determination and effective search capabilities. ## Golden Age of Computer Science All computer science problems are becoming solvable. Significant progress is being made in areas once considered challenging, like Go, protein folding, and media generation, potentially marking the start of a new golden age. ## Vision of the Future AI's ability to write code and conduct experiments offers immense potential for humanity. However, ethical and security considerations must be carefully managed. In the future, scientists and engineers collaborating with AI could achieve even greater breakthroughs. ## Sources covering this story - https://evjang.com/2026/02/04/rocks.html — --- _Summary written by Brifing from the original reporting. Copyright in the original article belongs to its publisher; read the full piece at the link above._