Dear Reader,

Yesterday I showed you a video demonstrating how powerful AI agents are when you set them up well (click here if you missed it) and give them (restricted) access to your notes and papers.

Today, let’s talk about exactly how this system is built. It has five components. Implementing them correctly is like making a wheel round, absolutely essential for its frictionless function. Let me explain in this video from the new AI Agents for Academics course:

The 5 components needed to use AI agents for academic work:

1. Your Obsidian vault & PDF collection: This is the home for everything you manage, containing your notes, PDFs, templates, and projects. The AI works directly inside it, reading and writing the same files you do. Every other component exists to make this vault more useful.
→ See the tutorial for setting up Obsidian
→ If you want to go all in, my flagship Information Management Course is about Obsidian for academics. You can get a bundle discount if you get both courses.

2. The AI plugin (Claudian or Gemini Scribe): The plugin is the bridge that brings a large language model into your vault and lets it read and edit your notes. You only need one: Claudian if you work with Claude or OpenAI, Gemini Scribe if you prefer Google’s models. For Claudian, you will need to install Claude Code or Codex first and log in using your subscription.
→ See the Tutorial for Gemini Scribe​
→Install Tutorial for Codex and for Claude

3. The AGENTS.md file: This is a single note that tells the AI how your vault is organised and how you want it to behave. Every AI conversation you have loads this file by default, so it knows where to find things and how you want them handled. Whenever your structure changes, you update it. A solid agents file covers four things:

• Structure: your folders, tags, properties, templates, and active projects, so the AI finds the right note instead of re-reading your whole vault.
• Behaviour: how the AI addresses you and, importantly, an editing-safety rule so it checks with you before making heavy-handed changes.
• Policies: your house rules, such as always marking AI-generated content, always reading the actual PDF rather than guessing from the title, and how to name new files.
• Capabilities: the skills and MCP servers available, so the AI knows which tool to reach for (for example, use Zotero when you ask it to import a paper).

In the course, I provide a drag-and-drop solution that uses AI that analyses your vault, asks you a couple of questions, and writes a perfect agents file.

4. Skills (Optional): A skill is a small file that tells the AI to do one repetitive task exactly the way you want, ready to reuse on demand. A skill can analyse 20 papers based on a criteria table, creating a preliminary lit review matrix, for example. It is very easy to create skills. So whenever a task repeats, you capture it once and never do it by hand again.
→ See this tutorial on building skills
→ I have a set of skills, I’ve built here​

5. MCP servers (Optional): These are connectors that let the AI reach software and websites outside your vault. Two matter most for academics: Zotero pulls your papers, annotations and tags straight into your notes, and Consensus, an AI literature search engine, returns papers for any research question and explains why each paper is relevant.
→ See the Zotero MCP Tutorial to get an idea of how this works

The video above is one of 23 modules in the Effortless AI Agents for Academics course. This course demystifies the process of setting up your AI agents safely and reliably(Part 1) and expands your imagination on how to use them (Part 2) for your academic work.

This second part is the real challenge, as the vast majority of examples you find online about AI agents is either related to creating "content" or organising your daily meeting notes and "running your life" (which I find quite pointless).

In the course, you can learn how to use agents to actually get academic work done: Synthesising claims, screening papers, searching for literature, connecting ideas & develop deeper thinking. And the founding cohort (next 3 days) will give us the opportunity to share and learn from one another.

Want a step-by-step walkthrough for setting it up?

Setting up efficient AI agents was a little technical, even though in retrospect it can be done quickly. There are no clear standards yet and things shift quickly. I talked to an engineer from Consensus, and he told me, "It's like the Wild West out there. We also don't know what we're doing."

The course will give you simple step-by-step instructions to set everything up, and as things change and evolve, there will be updates.

Folks joining in the next 3 days are automatically part of the founding cohort, which I will give a lot more hands-on help from me. We will also discuss and share how to use AI agents ethically, safely and effectively in academia - something I think nobody yet fully can imagine.

In my last email, I described the exact contents of the course (check it out here, if you missed it). But there are also a few materials included:

Included Materials in the Course

This course contains numerous skills and scripts you can download and use immediately:

1. Setup script for personalising your AI and defining clear policies.
2. A custodian skill to organise or restructure your notes and keep your vault healthy
3. A scoring skill to screen abstracts from Zotero in bulk based on a criteria table (for systematic lit reviews)
4. A connector skill (template) to cross-reference new papers with your existing notes and sources and spark new ideas.
5. A Zotero importer skill to pull papers directly from Zotero, analyse their full-text PDFs, and integrate them into your network of notes.

But my philosophy is that AI is highly flexible, so I don't want to force you into how I do things. For that reason, I provide setup scripts and questionnaires, rather than finished skills for most materials. This lets you use the principles I have developed and easily adapt them to your academic domain.

Updates

AI is evolving so quickly that this course will get updates. Even in the ~6 weeks I was working on the course, I already had to re-record two videos. In the near future, as new use cases become available, the course will also be expanded - Free of charge.

Don't miss the founding cohort

If you are serious about learning AI for your academic work, you need a peer group that works on the same challenges. This is why everyone who joins the course in the next few days is automatically included in the founding cohort space on Circle. Here you will:

1. Get exclusive help from me and other members
2. Get a bonus lecture on requested topics (academic writing, maybe?)
3. Learn from others and share your achievements
4. Have some homework to guide you through the course
5. A supportive community where no question is "too dumb".

Time left to secure a spot in the founding cohort:

​

Wishing you a wonderful weekend,

Ilya, The Effortless Academic

Before creating the course, I surveyed 73 academics and asked them: "What do you want to accomplish with AI?" Tomorrow, we will look at what academics want from AI, find some solutions, and answer a few remaining questions about the course.

Previous parts of this series​
Part 1: My PhD Defence: 6 days to rebut 30 pages of critique​
Part 2: What can AI agents do and how they helped me ace my PhD defence (Video)

This email is part of a 4-day series on AI agents. If you don't want to know about it, unsubscribe from this series.​
Want to leave forever: Unsubscribe from everything (never hear from me again)