Applied AI Seminar ยท 2nd Edition

Agentic AI for Research

Practical AI usage for Physicists

LLMs ยท Chat Interfaces ยท IDE Assistants ยท Agentic Workflows ยท Claude Code ยท Installation ยท Project Management ยท Skills

๐Ÿ“… June 16, 2026 ๐Ÿ•™ 09:15โ€“12:00 ๐Ÿ“ Margrethe Bohr Salen, Niels Bohr Building

๐ŸŽฅ Join Online via Zoom

cern.zoom.us/j/67768992649

Meeting ID: 677 6899 2649

Tutorial โ†—

About This Edition

๐Ÿ“ Draft: the programme below is being finalised and will be updated before the event.

Following the first Vibe Coding seminar, this second edition turns to agentic AI for research: using AI systems that can plan, write, run, and debug code autonomously to accelerate research workflows. Rather than writing every line by hand, you collaborate with agents that operate across your editor, terminal, and repository.

We revisit what has changed in the fast-moving landscape, share best practices for working with coding agents, and show live demonstrations โ€” from quick chat workflows to autonomous agents tackling multi-file scientific projects. We also look critically at where these tools excel and where they fall short.

Whether you are curious, skeptical, or already using these tools daily, this seminar gives a grounded, evidence-based look at the current state of agentic AI for research.

How to Join

๐Ÿ™‚ I'm just curious

Simply show up โ€” no preparation needed. Sit back and watch the demos.

๐Ÿš€ I want to try it myself

Show up with a Claude account (or a credit card to set one up on the spot). Students will be on hand to help you through the installation and setup.

Programme

A hands-on session for physicists. No hype โ€” just evidence and live demonstrations. (Draft schedule, subject to change.)

09:15โ€“09:30
Live Demo: Agentic Research in Action

An agent tackles a real research computation end to end โ€” a concrete example of what these tools can do before we open the hood. (Inar) ยท ๐Ÿ“Š Slides โ†—

09:30โ€“10:00
Intro: How LLMs Work โ€” and What They Cost

Caching, prefill vs. generation pricing, context size, and reasoning โ€” the mechanics that shape how you should work with agents. (Inar) ยท ๐Ÿ“Š Slides โ†—

10:00โ€“10:15
Setting Up Claude Code

Install, authenticate, and run your first "thing" โ€” follow along in the tutorial.

10:15โ€“10:40
Break

Coffee, plus hands-on setup help from the team.

10:40โ€“11:10
Agentic Research Patterns

The whole toolbox in half an hour โ€” plan/auto/effort modes, subagents, context & compaction, skills & MCP, feedback loops, evaluation & artifacts โ€” demonstrated live on one simple running example. (Daniel)

11:10โ€“11:55
Multi-Claude: Three Demos in Parallel

Putting the patterns to work โ€” driving three independent agents at once on three different research demos, live. (Daniel)

11:55โ€“12:00
Outro

Where this is going, and how to keep going after today.

The Four Surfaces

1. Chat Interfaces

Claude, ChatGPT, Gemini โ€” copy-paste workflows. Good for exploration, quick questions, and one-off scripts. Breaks down for complex projects.

2. IDE Copilots

GitHub Copilot, Cursor autocomplete โ€” inline suggestions as you type. Reduces boilerplate but requires constant attention.

3. Terminal Agents

Claude Code, Cursor Agent, Aider โ€” delegate entire tasks. Think of it like working with a capable research assistant.

4. Project-Level Automation

GitHub Issues + PRs with AI. Create issues, let agents implement, review PRs. Scales to large codebases.

Who Should Attend

  • Researchers comfortable with AI who want to upgrade their workflows
  • Researchers who use AI occasionally who want to see what is possible
  • Researchers who are skeptical about AI and want to see the arguments for and against

Organizers

Inar Timiryasov

Assistant Professor

Works on foundation models for particle physics (IceCube, LHC). Published papers on efficient language model training (BabyLlama). Used LLMs for coding before ChatGPT existed. Diehard Claude Code user.

Daniel Murnane

Postdoc, ATLAS ML Coordinator

Lead ML efforts at ATLAS (CERN LHC experiment), 3 years of vibe coding fanaticism, user of LLMs for writing code, managing projects, writing grants, organizing entire life. Creator of chATLAS AI assistant for ATLAS collaboration.

Troels Petersen

Head of Subatomic Physics Section

Long-time AI/ML advocate, particle physicist with ATLAS and IceCube.

Johann Ioannou-Nikolaides

PhD Fellow, Experimental Particle Physics

PhD Fellow in Experimental Particle Physics at the Niels Bohr Institute.

Setup (Optional)

If you want to follow along during the demos:

GitHub Copilot (free for academics)

Sign up for GitHub Education with your university email to get free Copilot access.

Claude Code

curl -fsSL https://claude.ai/install.sh | bash

Requires Claude Pro subscription ($20/month) or Claude credentials.

No prior setup required to attend โ€” you can just watch the demos.

Useful Resources

Essential tools and resources for AI-assisted programming. Most of these are free for academics!

GitHub Education FREE

Get free access to GitHub Copilot, GitHub Pro, and many other developer tools with your academic email.

What you get:

  • GitHub Copilot (normally $10/month)
  • GitHub Pro features
  • Access to GitHub Codespaces
  • Developer tool discounts

How to apply: Visit education.github.com and sign up with your university email address. Approval usually takes 1-2 days.

Visual Studio Code FREE

Microsoft's popular code editor with extensive extension support.

Key features:

  • Lightweight and fast
  • GitHub Copilot integration
  • Rich extension marketplace
  • Excellent Python support

Download: code.visualstudio.com

Cursor

AI-first code editor built on VSCode with advanced autocomplete and chat features.

Key features:

  • Multi-line AI autocomplete
  • Codebase-aware chat (Cmd+K)
  • Natural language edits
  • Compatible with VSCode extensions

Pricing: Free tier available, Pro plan $20/month

Download: cursor.sh

Claude Code SUBSCRIPTION

Anthropic's official terminal-based AI coding assistant. Autonomous agent that can edit files, run commands, and complete complex tasks.

Key features:

  • Works directly in your terminal
  • Can edit multiple files and run commands
  • Codebase exploration and analysis
  • Git integration (commits, PRs)

Installation:

curl -fsSL https://claude.ai/install.sh | bash

Requirements: Requires subscription to Claude Pro ($20/month) or Claude credentials. See code.claude.com/docs for details.

uv FREE

Ultra-fast Python package installer and resolver written in Rust. A modern replacement for pip and pip-tools.

Key features:

  • 10-100x faster than pip
  • Drop-in replacement: use uv pip install
  • Better dependency resolution
  • Works with existing requirements.txt

Installation:

curl -LsSf https://astral.sh/uv/install.sh | sh

Learn more: github.com/astral-sh/uv

Vibe Coding

AI-Assisted Programming for Physicists

๐Ÿ“… January 26, 2026 ๐Ÿ•™ 10:00โ€“12:15 ๐Ÿ“ Margrethe Bohr Salen, Niels Bohr Building

๐ŸŽฅ Join Online via Zoom

ucph-ku.zoom.us/j/63222289709

Meeting ID: 632 2228 9709  |  Passcode: 460019

About Vibe Coding

Vibe coding is the new way to use Large Language Models (LLMs) to work on software, projects, and even scientific research itself. Rather than writing every line of code manually, you collaborate with AI systems that can generate, edit, debug, and refactor code based on natural language instructions.

In this seminar, we explore what vibe coding is, best practices for working with AI coding assistants, the tools available across different workflows, and critically โ€” when these tools excel and when they fall short. You'll see real demonstrations of AI-assisted coding in action, from simple autocomplete to autonomous agents that can tackle complex multi-file projects.

Whether you're curious, skeptical, or already using these tools, this seminar provides a grounded, evidence-based look at the current state of AI-assisted programming and how it can fit into research workflows.

๐Ÿ“Š Presentation Slides

View slides on Google Slides โ†’

What You'll Learn

A hands-on introduction to AI coding tools, designed for physicists. No hype โ€” just evidence and practical demonstrations.

10:00โ€“10:25
An Introduction to LLMs in January 2026
10:25โ€“10:40
The Chat Surface

Claude, ChatGPT, Gemini โ€” copy-paste workflows and their use cases.

10:40โ€“10:50
The IDE Surface: Editing and Autocomplete

GitHub Copilot and Cursor autocomplete โ€” inline suggestions and productivity.

10:50โ€“11:00
Break
11:00โ€“11:20
The IDE Surface: Agents

Cursor agents and agentic workflows in the IDE.

11:20โ€“11:45
The Terminal Surface: Claude Code

Terminal-based autonomous agents for complex tasks.

11:45โ€“12:15
The GitHub Surface

GitHub Issues, PRs, and project-level AI automation.

The Four Surfaces

1. Chat Interfaces

Claude, ChatGPT, Gemini โ€” copy-paste workflows. Good for exploration, quick questions, and one-off scripts. Breaks down for complex projects.

2. IDE Copilots

GitHub Copilot, Cursor autocomplete โ€” inline suggestions as you type. Reduces boilerplate but requires constant attention.

3. Terminal Agents

Claude Code, Cursor Agent, Aider โ€” delegate entire tasks. Think of it like working with a capable research assistant.

4. Project-Level Automation

GitHub Issues + PRs with AI. Create issues, let agents implement, review PRs. Scales to large codebases.

Who Should Attend

  • Researchers comfortable with AI who want to upgrade their workflows
  • Researchers who use AI occasionally who want to see what is possible
  • Researchers who are skeptical about AI and want to see the arguments for and against

Organizers

Inar Timiryasov

Assistant Professor

Works on foundation models for particle physics (IceCube, LHC). Published papers on efficient language model training (BabyLlama). Used LLMs for coding before ChatGPT existed. Diehard Claude Code user.

Daniel Murnane

Postdoc, ATLAS ML Coordinator

Lead ML efforts at ATLAS (CERN LHC experiment), 3 years of vibe coding fanaticism, user of LLMs for writing code, managing projects, writing grants, organizing entire life. Creator of chATLAS AI assistant for ATLAS collaboration.

Troels Petersen

Head of Subatomic Physics Section

Long-time AI/ML advocate, particle physicist with ATLAS and IceCube.

Setup (Optional)

If you want to follow along during the demos:

GitHub Copilot (free for academics)

Sign up for GitHub Education with your university email to get free Copilot access.

Claude Code

curl -fsSL https://claude.ai/install.sh | bash

Requires Claude Pro subscription ($20/month) or Claude credentials.

No prior setup required to attend โ€” you can just watch the demos.

Useful Resources

Essential tools and resources for AI-assisted programming. Most of these are free for academics!

GitHub Education FREE

Get free access to GitHub Copilot, GitHub Pro, and many other developer tools with your academic email.

What you get:

  • GitHub Copilot (normally $10/month)
  • GitHub Pro features
  • Access to GitHub Codespaces
  • Developer tool discounts

How to apply: Visit education.github.com and sign up with your university email address. Approval usually takes 1-2 days.

Visual Studio Code FREE

Microsoft's popular code editor with extensive extension support.

Key features:

  • Lightweight and fast
  • GitHub Copilot integration
  • Rich extension marketplace
  • Excellent Python support

Download: code.visualstudio.com

Cursor

AI-first code editor built on VSCode with advanced autocomplete and chat features.

Key features:

  • Multi-line AI autocomplete
  • Codebase-aware chat (Cmd+K)
  • Natural language edits
  • Compatible with VSCode extensions

Pricing: Free tier available, Pro plan $20/month

Download: cursor.sh

Claude Code SUBSCRIPTION

Anthropic's official terminal-based AI coding assistant. Autonomous agent that can edit files, run commands, and complete complex tasks.

Key features:

  • Works directly in your terminal
  • Can edit multiple files and run commands
  • Codebase exploration and analysis
  • Git integration (commits, PRs)

Installation:

curl -fsSL https://claude.ai/install.sh | bash

Requirements: Requires subscription to Claude Pro ($20/month) or Claude credentials. See code.claude.com/docs for details.

uv FREE

Ultra-fast Python package installer and resolver written in Rust. A modern replacement for pip and pip-tools.

Key features:

  • 10-100x faster than pip
  • Drop-in replacement: use uv pip install
  • Better dependency resolution
  • Works with existing requirements.txt

Installation:

curl -LsSf https://astral.sh/uv/install.sh | sh

Learn more: github.com/astral-sh/uv