Dear Reader,

This week I am exploring a slightly different format for this newsletter. My goals are (1) to make it more personal, and (2) to be about more than just EffortlessAcademic. Let me know what you think and which parts you like by replying!

In the last few weeks, I have been absorbed in two deep paper revisions and experimenting with how I can use my notes, my code/data and AI to help me design experiments, code, and review the literature to interpret results. I am trying to build an AI system that combines these, can tell me what questions my data can answer, and even code and run the tests semi-autonomously. Here is a short unedited screencast I shared with the AI agents course founding cohort:

(The workflow is partially based on this underappreciated article from the EA archive).

So far, this has led to interesting insights:

1. The newer AI models are capable of holding a conversation for days, and it's vastly more powerful than having a series of topical AI chats, because you can keep discussing intermittent decisions and ideas and loop back to them.
2. My biggest struggle is to rein AI in to do less. It is far too eager, creating too many ideas, and confusing me. The only solution is to rely heavily on personalisation settings (or an Agents.md file if using agents).
3. Try dropping a plot from your/someone's paper (plus some explanation if not obvious), and ask AI to "interpret the findings and contrast them to existing literature using internet search". Maybe it is just that I haven't read enough, but almost every time I do this, the agent finds papers that make me think: "How did I not hear about that!?" This is gold for crafting a good discussion section and also works particularly well in the context of long conversations I just mentioned. (Results are much better if using the Consensus MCP).

Now let's dive into the weekly tutorial, some interesting content online and the free webinar I am giving.

Let me know if you like this new format by replying!

Ilya

Weekly Tutorial: The Best AI Tools for Lit Reviews in 2026

There are now dozens of AI tools promising to make literature reviews faster. Some help you find papers. Some help you screen a large set of studies. Others let you compare PDFs, extract data, or check whether a claim has support.

But most tools have a significant overlap. It is not trivial to decide which ones are worth their money and fit your way of doing research.

This week’s tutorial is a practical review of AI tools for literature review in 2026. It connects the use cases for each tool into a workflow, which you can adapt and modify to your needs. It presents a part of the lit review course, which covers most other available academic tools to enhance your research.

Three articles from outside EA

​Why I don’t use LLMs in writing my books​

Stephen Heard, an evolutionary ecologist, explains why he does not use LLMs to write his books (but other materials). His argument is that he wants his books to (1) exist, (2) be good, and (3) be written by him. This makes writing a craft (or even art) and self-expression. I think this is a useful distinction and food for thought to define your own stance on the use of AI in writing.

🌲 My take: If writing comes easily to you and you genuinely enjoy it, it can be an art form. But, in my experience, much writing for publication is functional, optimised for skimming, and increasingly mined by AI anyway. In that case, I would be functional, too. The distinction is like choosing between a handwritten note or a WhatsApp message. They can carry the same information, but they are perceived differently by the reader. Importantly, this reader has to be human to appreciate it. So use the tools based on audience and function of the content rather than the tool's capabilities to accomplish the task. (Heard’s website is also a great resource for writing advice).

​The perils of AI that knows you too well​

A thoughtful article in The Scholarly Kitchen warns that personalised AI can become an intellectual echo chamber. As an assistant learns your preferences, previous conclusions, and ways of framing questions it may increasingly agree with your assumptions, select supporting evidence, and downplay uncomfortable alternatives.

🌲 My take: AI memory is, technically speaking, an auto-generated document summarising some of your conversations. The problem is that AI chooses or interprets what goes there, and as with an LLM wiki, this leads to errors. Memory is a feature of user-centred software like the AI chat in the browser, but not with Claude Code or Codex. For research, I switched to working almost exclusively with agents in Obsidian, where AI sees my notes but doesn't form automated memories. This gives you extremely personalised replies but no AI-generated and accumulated bias.

​Nature Careers' webinar on AI in PhD programmes (Sep 23rd)

AI is changing PhD programmes: PhD students are already using AI for literature search, coding, presentations, and paper summaries, while many worry that this could weaken the research, thinking, and writing skills a doctorate is supposed to develop. So the question is which tools to use, how to use them ethically, how to retain critical thinking skills, and which parts of research should remain deliberately human.

👤 Who is it for: Relevant for supervisors to decide how to best train strong researchers and likely students curious about what tools are being used.

Sep 30th: Free lit review webinar with EA & Litmaps

In two weeks, I am running a free webinar with Litmaps on building a practical literature review workflow using Litmaps, Zotero, and Obsidian.

We will use Litmaps to explore the citation network and identify relevant research, organise the results in Zotero, then capture and connect the important ideas in Obsidian.

The webinar is aimed at those of you who have not yet adopted the Effortless Academic information management method and want to see what it is all about. Over 800 students and academics have already signed up. Join us!

Date: Sep 30th (or Oct. 1st based on location).

Want to leave Effortless Academic forever: Unsubscribe from everything. I am sorry to see you go, and farewell.