Takeaways This list isn’t comprehensive. I’m still experimenting and would love to learn from your experiments as well.

9. Reflector This final role is different. Whereas the others took as the object of inquiry a particular work — e.g., a novel or a movie — this last one takes as the object your knowledge garden itself. That is, you point the LLM to a series of notes to analyze patterns over time and suggest improvements. Example: I fed all 52 weekly posts from my humanities crash course to Claude Code, and asked it to identify the various roles in which I used AI for learning throughout the year. Its answers — with some curation from me — are the roles you just read. Suggested prompt: Here are my notes from [X weeks/months] of reading on [TOPIC]. What patterns do you notice in what I pay attention to? What do I seem to find most interesting, and what do I seem to avoid or underweight?

8. Mapper This one’s a bit more esoteric. Some people — me included — are primarily visual: diagrams and drawings aid our understanding. Concept maps can be especially helpful. I’ve built an Agent Skill to allow LLMs like Claude draw concept maps. (Download it from Github.) Example: I used this mapping skill to generate a concept map of Virginia Woolf’s To the Lighthouse. It’s not especially insightful, but more of a proof point of using LLMs in a more visual modality. Suggested prompt: (Note: install my LLMapper Skill before issuing this prompt) Generate a concept map for [WORK] centered on the question: “How does the novel’s treatment of [THEME] illuminate [BROADER QUESTION]?”

7. Analyst This role will also help you appreciate a work from a different perspective. It’s easy: you ask for the LLM to apply a specific critical lens to a reading. Common lenses include Freudian, Marxist, feminist, Girardian, etc. Example: The same week I read Freud, my son and I watched Predator, the 1980s sci fi film starring Arnold Schwarzenegger. For fun, I asked ChatGPT to analyze the film through a Freudian lens. The result was both enlightening and hilarious. Suggested prompt: Apply a [Marxist / feminist / postcolonial / Jungian] reading to [WORK]. What does this lens reveal that a neutral summary would miss?

6. Adversary Here’s a fun role: asking for an LLM to push back on your position or steelman the opposing point of view. The idea is to expand your understanding by bringing your assumptions to the surface and challenging them. Example: After watching Modern Times, I asked ChatGPT to correct my understanding of the movie as a work of Marxist propaganda. The LLM convinced me that the film is in fact more of a humanist statement than a political one. As a result of this interaction, I changed my mind on Chaplin’s work. Suggested prompt: Here are my notes on [TOPIC]. Please help me see it through the lens of someone who might be sympathetic to [OPPOSING POSITION] without fully realizing it. What could I improve? Where is my argument weakest? [paste notes]

5. Recommender This is a useful role for deepening your understanding of a subject: asking for related works that reflect similar themes. It’s also a use case where I noticed considerable improvements in LLM performance over 2025. Example: Early in 2025, I read Confucius’s Analects. Perplexity was ahead in web-backed interactions at the time, so I asked it for a list of classic Chinese movies that reflected Confucian values. It responded with five suggestions, some of which it hallucinated. But one of them, Spring in a Small Town, was a bona fide classic — and I likely wouldn’t have learned of it without an LLM. (Later in the year, other chatbots gained this ability and hallucinations dropped across the board.) Suggested prompt: I just finished [WORK]. Recommend three films that explore similar themes or ideas. Prioritize films with strong critical reputations — I’d rather have one great recommendation than five mediocre ones.

4. Orienter This role is something of an inversion of the validator. Instead of asking for feedback on your notes after reading a text, here you ask the AI for guidance before reading. You’re looking for framing, historical context, high level outlines, etc. — ideally, without spoilers. Example: Before reading Nietzsche’s Beyond Good and Evil and Tolstoy’s The Death of Ivan Illych, I uploaded both books to NotebookLM, which created a podcast for me that explained their thematic contexts. Listening to this podcast in my daily walk helped me better understand the readings. Suggested prompt: I’m about to read [WORK] for the first time. Give me enough context to make sense of it — historical background, key arguments, things to watch for — but don’t spoil the experience of discovering it myself.

3. Connector Here’s yet another role you can easily do via chat: identifying thematic, philosophical, or narrative parallels between works. Note I wrote “works” — it’s fun and illuminating to ask for connections across media, genre, time, etc. Example: I watched Francis Ford Coppola’s The Conversation on the same week I read Oedipus Rex. For fun, I asked ChatGPT for possible parallels between the two works. Its reply was enlightening: it pointed out how the protagonists of both stories undertook an obsessive investigation that uncovered terrible knowledge. Suggested prompt: I’ve been reading [WORK A] and [WORK B]. What philosophical or thematic threads connect them? I’m looking for non-obvious resonances, not surface similarities.

2. Validator Another basic role for AI is validating your understanding. To do this, you ask it to review your notes for errors or gaps, do basic fact checking, or critique your reasoning. Again, you can do this via the chat interface, but I also experimented with passing my notes in Obsidian using the Copilot plugin and in Emacs using gptel. Example: After reading The Epic of Gilgamesh, I wrote a note in Obsidian summarizing its plot. When I asked ChatGPT to critique my summary, it pointed out that I’d given the central character a redemption arc that isn’t present in the text. I’m so accustomed to the standard hero’s journey, that I projected it onto the book — and an LLM helped me correct this ‘hallucination.’ Suggested prompt: Here are my notes on [WORK]. What important ideas did I miss or underemphasize? Don’t rewrite my notes — just flag the gaps.

1. Tutor The simplest role for AI is as a tutor. You ask it to explain a difficult concept, clarify a confusing passage, translate jargon, etc. I mostly did this via the standard chat UI (although I created a ChatGPT project to preserve context for the course.) Example: While reading Freud’s The Interpretation of Dreams, I came across three unfamiliar German terms: es, ich, and über-ich. ChatGPT helpfully explained these are more commonly known as id, ego, and superego — three terms I already understood. Suggested prompt: I just read [PASSAGE]. I understand [X] but I’m confused about [Y]. Can you explain [Y] in plain terms, without assuming I have background in [FIELD]?

It was a messy process. That’s what you do in a garden! And the outcome wasn’t an enthusiastic endorsement of AI. Instead, I landed at a map of roles and modalities for how AI can help at different points in the spectrum. Let’s look at nine of these roles.

Robots in the garden

I consider amanuensis to be the ideal role for AIs in your knowledge garden.

I get most value around the middle of the spectrum. There’s a historical precedent here.

Cognitive debt is likely a much bigger threat than technical debt, as AI and agents are adopted.

why asynchronous agents deserve more attention than they currently receive, provides practical guidelines for working with them effectively, and shares real-world experience using multiple agents to refactor a production codebase.

While the initial results fall short, the AI field has a history of blowing through challenging benchmarks. Now that the APEX-Agents test is public, it’s an open challenge for AI labs that believe they can do better — something Foody fully expects in the months to come.

“The way we do our jobs isn’t with one individual giving us all the context in one place. In real life, you’re operating across Slack and Google Drive and all these other tools.” For many agentic AI models, that kind of multi-domain reasoning is still hit or miss.

The new research looks at how leading AI models hold up doing actual white-collar work tasks, drawn from consulting, investment banking, and law. The result is a new benchmark called APEX-Agents — and so far, every AI lab is getting a failing grade. Faced with queries from real professionals, even the best models struggled to get more than a quarter of the questions right. The vast majority of the time, the model came back with a wrong answer or no answer at all.

MCP was donated to the new Agentic AI Foundation at the start of December. Skills were promoted to an “open format” on December 18th.

Qwen3-Coder Alibaba's performant long context models for agentic and coding tasks

Het zijn markdown bestanden met een persoonlijkheid, frameworks, en output templates. Die heb ik niet zelf geschreven - ik heb Claude gevraagd om ze te maken. “Maak een Product Owner agent die goed is in prioriteren en impact/effort analyses kan doen.” Claude schrijft dan het volledige bestand, inclusief werkwijze en voorbeelden.Als ik vervolgens zeg “vraag dit aan Tessa”, laadt Claude dat bestand en wordt Tessa.

In mijn werkmap heb ik een verzameling “agents” - tekstbestanden die Claude vertellen hoe hij zich moet gedragen. Tessa is er één van. Als ik haar “laad”, denkt Claude vanuit het perspectief van een product owner.

AI checking AI inherits vulnerabilities, Hays warned. "Transparency gaps, prompt injection vulnerabilities and a decision-making chain becomes harder to trace with each layer you add." Her research at Salesforce revealed that 55% of IT security leaders lack confidence that they have appropriate guardrails to deploy agents safely.