1
Define your starting point
Users define what they’re working on and how familiar they are. This sets the scope for what comes next.
Turning a static sandbox into a learning playgroundGEN AI for Obsidian
Enhancing learnability in a tool built for thought.
Obsidian is a powerful note-taking toolbuilt around linked thinking.
People don’t just store notes in it, they build 
Networks of ideasLinked notes can grow into a personal knowledge graph, making relationships between ideas easier to see and revisit.. Over time, these notes form a personal knowledge graph: a 
Second brainA trusted system outside your head for storing and organising ideas, so you can build on them without relying on memory alone., attracting users who want to create intricate knowledge systems.
1.1Challenge
Introducing AI features in Obsidian…
The tool currently has no native, built-in AI features in its core workflow.
The Verge interviewAug, 2025View
…that are meaningful.
respecting privacy, user control, and local-first values.
We had one week to answer:
How should AI behave in a tool made for thinking?
The learning curve is steep paralyzing.
Obsidian is powerful & beloved by power users, but intimidating to new ones. Across the board users felt stuck at the starting line.
User reviews & articles
2.1Insights
The pattern was clear, new users were:
#1Overwhelmed by features they didn't understand yet
#2Unclear where to start learning
#3Frustrated by documentation that assumed too much knowledge
2.2Constraints
Principles set in stone.
Obsidian has their guiding principles set in stone. Any AI feature we would come up with had to respect their philosophy.
Given our limited timeframe, this ultimately helped us stay focused and make more deliberate decisions.
Diving in deeper...🎯
3.1Thinking with constraints
We started with listing down the obvious areas of intervention:
Add AI, but where?
- redesign onboarding?
- add tooltips !
- Guided walkthroughs...
The problem
All of these go against Obsidian’s manifesto: intrusive onboarding limits user control, mandatory walkthroughs reduce flexibility, and cloud-synced progress compromises privacy.
3.2The Sandbox: Buried Potential
I went back to the interface, as a fairly new user myself I started exploring and just clicking around which led to discovering the “Sandbox vault.”
Discovering sandbox
It sits behind the “Get Help” button, easy to miss. When you access it, you get pages of documentation about how to use different features; backlinks, templates, graph view—all in text form.
The promise
It’s called “a space to play around and experiment with various features”
The reality
A text repository with backlinks. No interaction or hands-on practice.
The gap between promise and reality was obvious. Obsidian calls it a playground, but users end up getting a manual.
3.3The Pivot
The Sandbox already respects Obsidian’s core values:
What it gets right
- Separate vault means your main data stays untouched
- Nothing saves, respecting privacy
- Local-first, no cloud dependency
- Already positioned as a learning space
This made us narrow our focus from broad onboarding redesign to this specific feature.
Lights, camera, AI ction
We split the Sandbox into two interconnected features that transform passive documentation into active learning:
Playground
Learn with intention
Choose what you want to learn; Playground generates structured lessons inside Obsidian so you can understand features step by step.
Experiment
Practice with purpose
An interactive practice space that creates contextual challenges on demand. Regenerate challenges, rate difficulty, and refine your skills through repetition.
The redesign keeps humans in control with AI curating and generating scenarios and users doing the actual learning and practice.
Using AI to design AI
5.1Research and Synthesis
Here’s how we uncovered and validated the problem:
The research-to-insight funnel
We mixed human judgment with intentional AI use. Some of our guiding principles were:
5.2Information Architecture
Before designing screens, we defined the GenAI logic:
Collaborative synthesis
Mapping input → analysis → output directly shaped the architecture, hierarchy, and interaction model of both experiences.
5.3POC Screens
Entry redesign
A clearer starting point
BeforeAfterReplaced experience-based branching with Playground as a structured path that moves users from exploration to practice.
Playground
Learning, structured by you
2
Select and generate
Users choose features and output formats. AI generates focused learning modules based on those selections.
Experiment
Turn learning into practice
1
Move from lesson to challenge
After completing a module in Playground, users can choose to enter Experiment. AI generates contextual challenges based on what they just learned.
2
Generate, reflect, refine
Users can regenerate challenges at varying difficulty levels, reveal solutions, and rate difficulty.
Validating the redesign with new users.
We tested with two new Obsidian users to validate whether our redesign improved discoverability and learning clarity.
User testingDive into the detailsMethod: Think-aloud protocol comparing the original Sandbox interface with our redesigned version.
After observing how participants interacted with both versions, we analysed patterns in task completion, engagement, and learning behaviour.
6.1Impact
Task completion increased
Users completed learning tasks faster with structured paths vs. searching through documentation and external sources.
Reduced abandonment risk
In the original interface, both participants considered stopping the task. With structured guidance, neither expressed abandonment intent.
Increased learning intent & confidence
Users indicated preference for Playground as their primary learning entry point and reported higher confidence navigating Obsidian.
6.2Next Steps
Based on feedback, we identified refinements for the next iteration:
Challenge previews
Show example challenges before generation so users know what to expect.
Progress indicators
Add visual feedback showing which tools they've learned and practiced.
Add category icons
Visual grouping would help users distinguish between feature types in Playground.
Surface Sandbox more prominently
Users questioned why such a helpful feature was buried and hidden. Make Sandbox stand out as the primary help space rather than tucking it behind “Get Help.”
Lessons from Building with AI
Intent Shapes Intelligence.
AI is only as strong as the direction it’s given. This project pushed me to define intent clearly; in prompts, in flows, and in the overall experience. Better inputs led to better outcomes.
It’s all connected.
Who doesn’t love a good Charlie Day moment? Jokes apart, introducing AI into Obsidian while staying true to its philosophy was a tricky balance to strike. It made me realise how every little decision creates a ripple effect, shaping how people think, work, and relate to their tools.