We've added AGENTS.md files to every Frontend Mentor challenge

AI coding tools are everywhere now. If you're learning to code in 2026, you're almost certainly using GitHub Copilot, Claude, ChatGPT, Cursor, or something similar. And for good reason. These tools can be genuinely powerful learning companions when used well.

The question we've been thinking about is: how do we help our community get the most out of AI tools while working through our challenges? We wrote about this in our AI coding assistants for beginners guide, and the response from the community made it clear that you want practical support here. You want to use AI, and you want to make sure you're learning as effectively as possible while doing so.

So we've built that support directly into the challenges. Every Frontend Mentor challenge now ships with an AGENTS.md file in the starter code. If you use Claude, there's also a CLAUDE.md file that points to the same instructions. These files tell AI tools how to behave when helping you with a challenge, and they're tailored to each difficulty level.

What are AGENTS.md and CLAUDE.md files?

AI coding tools like Claude, Cursor, and GitHub Copilot look for instruction files in project directories to understand how they should behave. AGENTS.md is a widely supported convention, and CLAUDE.md is the Claude-specific equivalent.

By including these files in our starter code, we've set up your AI assistant to be a better learning partner from the moment you open a challenge. Think of them as a brief for a tutor. They explain who you are, what level you're at, and how the AI can help you learn most effectively.

The core principle across every difficulty level: AI should guide your thinking, not replace it. The files are designed so that AI tools help you work through problems step by step, building real understanding along the way.

How the guidance changes by difficulty level

Everyone's at a different stage of their learning journey, so the AI instructions are calibrated to match. Here's how the approach shifts across our five difficulty levels.

Newbie

The AI acts as a patient, encouraging mentor: someone who remembers what it felt like to see code for the first time. It breaks everything into the smallest possible steps, uses real-world analogies (the box model is like a gift box with bubble wrap padding), and gives multiple hints before revealing an approach. If you ask for a complete solution, it'll help you break the problem into something you can tackle piece by piece. The focus is on building confidence and making coding feel approachable.

Junior

Here the AI shifts to being a supportive guide that helps you connect the dots between what you know and what you're building. It introduces debugging techniques, encourages you to use browser DevTools, and starts explaining the "why" behind concepts rather than just the "what." You'll get fewer hints before more direct guidance, and the AI will help you develop the problem-solving instincts that carry forward into every future project.

Intermediate

At this point, you're treated as a capable developer building professional skills. The AI becomes an experienced colleague who discusses trade-offs, presents multiple approaches, and lets you make decisions. Instead of one "right answer," you'll get options with pros and cons. It'll push you to think about maintainability, code architecture, and industry standards: the things that separate working code from professional code.

Advanced

The AI steps into a senior colleague role. It challenges your thinking, plays devil's advocate, and explores edge cases you might have missed. It won't over-explain, but it also won't assume you know everything. If you're making an architectural decision, expect probing questions about long-term implications. If you share code for feedback, expect honest critique alongside acknowledgment of what's working well.

Guru

At the highest level, the AI is a peer collaborator. It debates approaches, brings contrarian viewpoints, and explores problems with genuine curiosity. It'll reference specs, discuss experimental browser features, and engage with the kind of nuance that advanced work demands. If it disagrees with your approach, it'll say so, and back it up with reasoning. If it doesn't have a confident answer, it'll say that too.

Why we're excited about this

Nobody has fully figured out the best way to learn with AI yet. The industry is still working it out, and that's an exciting space to be in. We see an opportunity to make AI tools work harder for our community's learning: not just by answering questions, but by asking them, encouraging experimentation, and adapting to where each person is in their journey.

A few things these files consistently encourage across all levels:

• Learning through guided discovery. The AI is instructed to lead with questions and hints, helping you arrive at solutions through your own reasoning.
• Building debugging skills. From the Junior level onward, the AI introduces tools and techniques for diagnosing problems, a skill that pays dividends across your entire career.
• Understanding trade-offs. As difficulty increases, the AI presents options and lets you decide, which mirrors how real development teams work.
• Connecting you to resources. Every level points to relevant learning materials including MDN, CSS-Tricks, our own articles and learning paths, and our Discord community.

How to get the most out of this

If you're already using AI tools with our challenges, you don't need to change anything. The files are already included in the starter code, and tools that support AGENTS.md or CLAUDE.md will pick them up.

A few tips to make this work well for you:

Be honest about your level. The instructions are calibrated to your challenge's difficulty. If you're working on a Newbie challenge, the AI will be more patient and explanatory. If you're on an Advanced challenge, it'll treat you like a peer. Pick challenges that match where you actually are, and the AI guidance will be most useful.

Lean into the process. The best learning happens when you're working through something that feels just out of reach. These files are designed to keep AI in a supporting role during those moments, helping you push through rather than jumping ahead.

Pair this with good prompting habits. If you haven't read our AI coding assistants for beginners article, it's worth your time. It covers the Ask-Don't-Copy principle, stage-based guidance for your first 18 months, and templates for writing better prompts.

This is just the beginning

We'll be iterating on these files based on how things go and what we hear from you. If you have thoughts, suggestions, or experiences to share, come find us in our Discord community. We'd love to hear how you're getting on with them.

The goal hasn't changed: build real skills by building real projects. Now your AI tools are set up to help you do exactly that.