Agentic Engineering for Scientific Computing

Your newest colleague writes code faster than anyone you have ever worked with, happily takes on the boring jobs and, every now and then, makes things up with complete confidence. Welcome to agentic engineering. Researchers, analysts and quants write code almost every day, be it for simulations, forecasts or pricing models, yet many have never trained as software engineers. For them, AI coding agents are a gift that comes with a catch. In scientific work, an algorithm is either implemented correctly or it is not, and no amount of AI confidence changes that. There is little room for vibe coding. This book shows how to make the most of AI agents without losing control. It distils the invariants of a fast-moving world into rules that hold no matter which large language model (LLM) is making the headlines. Examples from tools such as GitHub Copilot and Claude Code keep the advice concrete. You will learn what LLMs, agents and harnesses really do, why planning (almost) always pays off, when to start over, how to give your agent the insider knowledge it lacks, and how tests, linters and benchmarks close the feedback loop. The book even turns the tables: your agent gets to review you. Teams will find guidance on honest AI use, code review as the new bottleneck, backfiring KPIs and whether to let an agent replace a dependency. The book pays just as much attention to the risks, including comprehension debt, automation complacency, brain rot, lost paper trails, context overload, spiralling token costs, constant distraction and leaked intellectual property. Along the way, you will find out why an agent's "leaner" rewrite can end up slower, and why "The agent said so" will satisfy neither auditors nor peer reviewers. Newcomers and seasoned agent users alike will find short chapters with self-contained sections, FAQs and example prompts. AI will not make you a better scientist or analyst by itself. Used mindfully, though, it might just give you the time back to be one.

Autorentext
Raik Becker is a scientist turned quant. A PhD in energy economics and postdoctoral research laid the scientific foundation. Since 2018, he has worked in the energy industry, currently as a lead quantitative analyst at Vattenfall. In both worlds, he has built models of all kinds in MATLAB, R and Python. He has used agentic engineering tools since the early days and still uses them on a daily basis, which has given him plenty of practical experience.