Coding Agents & Data Science

Coding agents are changing how data science work is done. But where do they create value, and where is expert judgement still essential?

You are working with data, a hypothesis, a messy notebook, a model, a report, or a pile of scripts that need to move forward. AI can write code, but the important question is no longer simply whether it can code. It is how you structure the work, give it the right context, test the results, and keep human expertise in the loop. 

We will work on:

• What coding agents are – and how they differ from chatbots and traditional copilots. 

• How agents can support exploratory analysis, data cleaning, model development, visualization, documentation, and testing. 

• How to design a useful "harness": files, tests, evals, prompts, review, and guardrails that make the agent productive and controllable. 

• Concrete examples of agentic workflows in data science. 

• Common pitfalls: hallucinated results, bad assumptions, data security, and over-automation. 

After the webinar, you will be able to:

• Assess which data science tasks are well suited for coding agents - and which are not. 

• Set up a simple agentic workflow around analysis, code, tests, and documentation. 

• Give a coding agent better context, clearer goals, and more useful checkpoints. 

• Use tests, notebooks, review, and small evals to quality-assure an agent's work. 

• Identify the most important pitfalls before they become expensive or embarrassing. 

Target audience

This webinar is for:

• data scientists 

• analysts, engineers, software developers

• AI-curious specialists, and technical leaders who work with data, code, or automation - and who want to understand how coding agents can be used practically and responsibly. 

Level

Lightly experienced to experienced. You do not need prior experience with coding agents, but a basic understanding of data analysis, code, or technical workflows will make the examples easier to connect to your own work. 

Format

Presentation with concrete cases, live demo, short reflection questions, and Q&A. The focus is practical: what already works, what to be careful about, and how to get started without making the process opaque. 

Expectations and prerequisites

No preparation is required. Familiarity with developer tooling like Python, R, SQL, Bash, Git, or similar will make the examples easier to map onto your own work, but participants do not need to install or use any specific tools during the webinar.

Q&A

There will be room for short questions along the way, and we will set aside time for Q&A in the final part of the webinar. 

Follow-up material

Slides, links to relevant tools and resources, and a short checklist for assessing whether a data science task is suitable for an agentic workflow. 

Speaker

Ulf Aslak Lai

Consultant, Data Science & AI

Ulf a data scientist with a background in physics, applied mathematics, software engineering and foundational research. He has a PhD in Complex Systems. He works as a consultant in data science and AI, building mathematical models that deliver business insight for various companies.

Would you like to learn more about the topic? IDA offers the following course:
Coding Agents & Context Engineering for Developers | IDA

This webinar is part of IDA Upgrade, IDA’s dedicated webinar week focused on continuing professional development.

The week features 16 specialised webinars in Danish and English, giving you the opportunity to strengthen your skills and gain knowledge you can apply directly in your work.

👉 View the full programme here.

Captions are AI generated