Seminar: Agentic AI Coding for Computational Scientists – Mar 30–Mar 31 (Instats)

[Leonid Chindelevitch]

Dear colleagues,

I wanted to draw your attention to an upcoming seminar I am running through InStats that may be directly relevant to your work, as I will use computational biology examples:

Agentic AI Coding for Computational Scientists
March 30–March 31 (livestreamed, with on-demand access)
Instructor: Leonid Chindelevitch

https://instats.org/seminar/agentic-ai-coding-for-computational-scie

Agentic AI tools have made it possible to write, refactor, and extend code faster than ever, but speed without understanding is a liability. Most researchers using these tools have picked up habits that work until they don't: projects that gradually become unmaintainable, results that are hard to reproduce, and bugs that take days to diagnose because neither the researcher nor the AI fully understood the codebase.

This seminar is designed to change that.

**What the course covers**

- The fundamentals of agentic AI for coding: what these systems actually do, how to direct them effectively, and where their reasoning breaks down
- Best practices in software development — and how the introduction of AI assistance changes which practices matter most
- What makes scientific software development distinct: reproducibility, long-term maintenance, domain-specific correctness, and the particular dangers of code that produces plausible-but-wrong results
- A structured framework for three distinct situations practitioners face: reviving dormant projects, extending active codebases, and starting from scratch — each with its own risks and optimal strategies

**Why this course matters**

The pitfalls of agentic AI coding are not obvious, and most resources in this space focus on what these tools can do rather than what they get wrong. This course puts significant emphasis on failure modes — the specific patterns that lead to subtle errors, technical debt, and lost reproducibility — and how to systematically avoid them.

**Hands-on throughout**

The course is built around worked examples drawn from multiple areas of computational science, including mathematics, statistics, and computational biology. Rather than toy demonstrations, they are the kinds of tasks researchers like me and you actually face, chosen to surface the issues that matter in practice.

The seminar is PhD-level and aimed at researchers who are already writing code and want to use AI tools more deliberately. It is available through Instats either directly or via your institution's membership (which may entitle you to a significant discount).

I hope to see some of you there.

Best regards,

Leonid