Tag: Notion

In 2022, when the rest of the world was only beginning to get a tangible sense of what AI could do, Notion had already become one of the earliest tools to integrate GPT-3. Its AI features have been evolving for years now. So if you ask me which “note-taking app” currently achieves the deepest and most practical integration with AI, I would, without hesitation, recommend Notion.

Yet over these past three years, while Notion AI has iterated countless times, I’ve rarely seen people on my timelines talking about how easy and powerful it is. That inevitably leaves me feeling a bit regretful. Model upgrades can easily ignite waves of excitement, but after the hype settles, what really matters is whether AI can truly optimize—or even reinvent—our outdated workflows, rather than becoming just another traffic-chasing keyword for influencers. That’s what I genuinely care about.

So in this article, I want to share a few topics:

1. Why Notion AI Is Worth Trying
2. How Much Potential Notion Agent Really Has
3. How I Personally Use Notion AI
4. Notion AI Pricing and Subscription Advice

Every time I write about Notion, I can’t seem to control the length. This article is long, but I’m certain these are details very few people ever talk about. Next, I’ll start with a brief introduction to Notion AI’s basic capabilities. If you want to skip directly to the core topic of this article—Notion Agent—you can jump to the second section.

1. The Basic Capabilities of Notion AI

First of all, just like every other AI tool you’ve used, the fundamental way you interact with Notion AI is through a question-and-answer chat. You can select a paragraph on a page and ask directly, or you can open the AI panel on the right side and carry out a longer conversation, as shown in the example below.

In addition, the selected paragraph is automatically added to the context in the right-hand panel, so you don’t need the extra step of copying and pasting.

Beyond content-specific Q&A, Notion AI can also perform semantic search across your entire workspace. When you only remember the general idea of a note—but forget its title or which database it’s in—you can simply describe it vaguely, and the AI will locate the relevant note for you.

Based on this ability, I built an item management hub in my Notion system. For certain important but infrequently used items, I’ve set their storage locations, so I can ask questions like this:

And because Notion AI is connected to the latest models from Anthropic, OpenAI, and Google, it has full multimodal processing capability. It can handle text, analyze uploaded CSVs, PDFs, and images, perform online searches, and even accept direct webpage links—reading the page content before responding.

After generating an answer, Notion AI can directly perform create/read/update/delete actions on your pages or databases. This means you can ask Notion AI to modify the original text of a note, or instruct it to create a new page and store the generated content in a specified location (including inside a database).

Basically, anything DeepSeek or Doubao can answer, Notion can answer too—and usually does it even better. But the real key is this: at every moment of writing, note-taking, summarizing, reflecting, or doing a review, when you need AI, you never have to open a second tool. All your AI needs can stay entirely inside Notion—truly achieving an All-in-One workflow.

In a note-taking setup without integrated AI, you typically need to switch between multiple windows repeatedly. If your task requires multi-turn conversations—or referencing several different notes—the number of these switches multiplies quickly. And when you finally want to save AI-generated content, you have to manually reformat it, add tags, delete unnecessary parts… everything must be done by hand.

This tedious workflow not only wastes time—more importantly, it breaks your flow. While waiting for the AI to respond, you often get distracted: you might check your phone, scroll social media, watch a short video… and ten minutes disappear without you noticing.

Notion AI’s real convenience is not just that it gives answers, but that it can handle everything inside and outside your workspace in one seamless environment. No more copying a paragraph into Doubao for analysis. No more pasting ChatGPT’s output back into your notes. Removing even a single context switch can mean a world of difference in user experience.

When you need to write daily/weekly reports, you can simply reference all the documents you wrote this week inside the Notion AI panel, let it read them directly, and it will generate the report for you. When reviewing the key decision-making of a project, you can feed it multiple meeting notes, let it extract all discussed options, or even read the page’s version history to analyze how each decision evolved.

You can also build a clipping database using the Save to Notion extension, add an AI field, and preset a specific processing prompt—for example, generate reading summaries or extract key information. In this way, you can have an unlimited AI-powered reading hub even without subscribing to Readwise.

Notion is already ideal for organizing high-density, high-value information. With access to state-of-the-art AI models and the most seamless interaction experience, it naturally delivers better results than other note-taking tools.

And thanks to a decade of Notion’s open ecosystem development, you can even search external data sources directly inside the AI window—Google Drive, Google Calendar, GitHub, Gmail, and more—provided you’re already within those ecosystems.

Besides these built-in “connectors,” you can also link Notion with more tools through MCP, such as Cursor, Manus, Perplexity, ChatGPT, and others. You can send a Notion page link directly to these tools, and they can read the note content without any tedious copy-and-paste steps. These tools can also modify your Notion pages directly, based on your instructions.

For example, when you receive a long research report via Manus, you used to manually copy and paste it into another note-taking app. But now, you can simply send the Notion page link to Manus, and Manus can write the content directly into the designated place inside Notion.

Notion AI can take over your entire information-processing workflow—your notes, your tasks, your project documents, and even data from your connected third-party tools can all be processed faster and more directly. But these are only the basic applications of Notion AI; if it were merely “a chat window + quick access to information sources,” that alone wouldn’t be enough for me to keep using it long-term.

The real purpose of this article is to show you the key capabilities and potential of Notion Agent.

2. Notion Agent

What Is an Agent
Simply put, an Agent is an AI system capable of autonomously completing multi-step tasks on behalf of a human.

Most AI Q&A tools you’ve used can only provide information or ideas. After receiving the answer, you still have to manually execute the next steps. They don’t know who you are, what projects you’re working on, or what your habits and preferences are. Every conversation resets to zero—you must repeatedly explain context and clarify your needs. Of course, if an AI tool supports “memory” or “projects,” this can improve slightly.

But an Agent can do far more: once you give it a task, it can—within the scope of the information you authorize—intelligently make decisions, proactively call multiple tools, autonomously execute complex steps, and finally deliver the completed result back to you. All you need to do is enjoy the outcome. In addition, an Agent can not only store memory but also be trained. Through repeated interactions, it becomes smarter and more aligned with your expectations.

How to Build a Notion Agent
To use Notion Agent, you must first create a document specifically for the Agent. In this document, you define the Agent’s fundamental behavioral guidelines—for example: its identity and mission, communication and behavior rules, or the working scenarios and goals that shape its actions.

Suppose your goal is to make Notion AI better assist your content creation. Then your minimal viable Agent document might look like this:

After writing this document, go to Notion AI’s personalization settings and assign this document as the Agent’s system-level instruction, as shown below:

Once this is done, every time Notion AI generates a response, it will first follow the instructions defined in this document. It will interact with you according to the behavioral guidelines you set. As a result, when answering the same question, Notion AI with an Agent document and Notion AI without one will give completely different answers—the former precise and personalized, the latter generic and mediocre.

At this point, you might wonder: Isn’t this just giving the AI a prewritten prompt? If I paste the same prompt into Doubao or DeepSeek, won’t I get the same effect? To some extent, yes. But Notion Agent differs from ordinary AI chat tools in several key ways.

1. Documentation Is the Rule

Other AI tools require you to manually enter or copy-paste your prompts every single time. Notion Agent is different: its rules live directly inside a Notion document. They are automatically loaded, instantly editable, and immediately effective. More importantly, you can reference any existing Notion pages directly inside the Agent document—like this:

In other words, you can plug your existing creative SOPs, your writing notes, and your preferred methodologies straight into the Agent at high speed—no extra setup, no code, no complicated configuration—because the Agent instruction file is itself just a regular Notion note.

This creates a kind of “lift yourself by your own bootstraps” loop: Notion gives you a great environment for documenting your knowledge, and the Agent turns that accumulated knowledge into something executable. As you collaborate with the Agent, you’ll notice which parts of your notes work well, which parts need improvement, and where gaps exist. Then you refine your notes, and the Agent immediately becomes better. A positive feedback cycle naturally forms.

This is the essence of “documentation is the rule.” Your notes are no longer static archives—they become executable rules, reusable processes, and testable knowledge.

2. Agents Can Directly Operate Your Notion Workspace

This is another fundamental capability of Notion Agent: it has permission to perform create/read/update/delete operations on your pages and databases. Other AI tools can only generate text responses, but Notion AI can directly execute underlying actions. Here are a few concrete examples:

1️⃣ Create and Modify Notes

When you finish discussing an idea with the AI, you can simply tell it to organize the conversation into a note and save it to a specific database. The AI will automatically read that database’s property fields, understand what each field is for, and then intelligently populate the content: applying the correct tags, linking related projects, setting priorities, etc.

For example, when I was learning about Claude Skill, I asked Notion Agent to search the web for information, summarize it into a note, and store it in my Notes database, as shown below.

Notion Agent not only organized the content correctly—it also knew which database was my Notes database, and automatically filled in all the database properties. It understood the meaning of the six basic tag categories for note-taking that I described in this article.

2️⃣ Batch Operations Across Database Pages

When you need to process tasks in bulk, Notion AI can act on an entire database at once. For example:

• Mark all overdue tasks as high priority
• Identify all tasks completed this week and generate a summary

Actions that would normally require you to click through each item manually can now be completed in one sentence.

Also, the example database in the screenshots was created entirely by Notion Agent—I simply told it: “Please understand the context and create a demonstration database for this sentence.”

3️⃣ Workflow Automation

Going further, you can ask the AI to automatically execute complex multistep sequences based on specific conditions. For instance, “Help me generate a weekly report” is not just simple data retrieval—it’s an entire workflow: accessing multiple data sources → filtering pages → reading content → applying a report template → saving it to the correct location and filling in properties. Every step is executed automatically according to your preset rules, without manual intervention.

I’ll go into much more depth on automated workflows in later sections, so we’ll pause here for now.

4️⃣ Modify Rules and Memory in Real Time Based on Your Instructions

When you ask the Agent to generate a weekly report for the first time and find the summary too brief, you can simply say: “Remember, each task in the weekly report must include specific details of what was done.” The Agent will then proactively update the rules inside the Agent Document, and next time it will automatically follow this standard. Or if you notice that the Agent always over-compliments your writing during review, you can say: “From now on, just point out the issues—don’t praise me.” It will immediately adjust its tone and update the Agent Document accordingly.

Once you get used to this interaction style, refining the Agent’s behavior becomes incredibly easy. One sentence is enough for it to remember and adapt—no need to rewrite complex rule documents. Your collaboration will naturally become more and more seamless.

For example:

And the effect:

These foundational features together form the core capabilities of Notion Agent:

• Document as Rules: your notes directly become the Agent’s behavioral instructions
• Database as Memory: the Agent knows where to read and where to write
• Conversation as Training: one sentence is enough for the Agent to remember and improve

But underlying capabilities alone are not enough. The real challenge lies in how to organize these abilities and apply them to real work scenarios. Next, I’ll share some design principles for crafting effective Agent Documents, helping you connect these building blocks into truly useful workflows.

3. Agent Design Principles

Scenario Routing

Real work scenarios are complex. When you say “take a look for me,” you might be asking the Agent to review an article, check a video script, or examine a project’s progress. The same sentence can imply totally different needs depending on context. Of course, you could write every possible situation directly into the Agent Document—but then the Agent would need to load all instructions for every conversation, wasting valuable context space.

That’s why I recommend building your Agent Document with a progressive disclosure approach—layering information and loading details only when needed, instead of everything at once. My personal method is to define four core documents that must be loaded at the start of every conversation. Each new chat loads only the minimal necessary context.

These core documents vary by person, depending on your unique workflow, but generally you should at least include:

• Identity & Mission: who the Agent is, its core purpose, and whom it serves
• Interaction Style: tone of communication, response format, when to be brief, when to elaborate
• Continuously Updated Memory: user preferences, latest habits, ongoing requirements
• About the User: the user’s identity, background, work style, values, etc.

Only when the Agent detects specific keywords during a conversation does it load relevant sub-documents—like the example below. Each scenario sub-document includes a complete SOP: detailed workflow, evaluation criteria, and output format. This avoids loading all sub-documents at once and keeps the Agent’s responses focused and efficient.

For example, while writing this very article, I can highlight a paragraph and ask the Agent to “generate an image.” Notion Agent will detect the keyword “generate image” and activate only the corresponding sub-document, Scenario N: Content Illustration Generation, as shown below:

According to the SOP defined in the “Scenario N: Content Illustration Generation” document, Notion will automatically follow these steps:

1. Select a style: default to the pre-determined illustration style
2. Understand the content:
   • For partial illustrations: analyze the meaning of the selected text and the intended purpose of the image
   • For article covers: extract the core theme and emotional tone of the full piece
3. Generate a prompt: Generate the image prompt: Based on the selected text + the default design-style document + contextual information + any additional notes from the chat window, directly output a complete image-generation prompt.

In Step 3, I require the Agent to prioritize my predefined top-level design style, which specifies the default aesthetics, ratios, and stylistic preferences for images. This is why all images in this article generated with Nano Banana maintain a consistent visual style.

The generation process and results are shown below:

The same execution logic applies to other scenarios as well. For example, I created a “Diet Log” sub-workflow, and now I can simply send a photo of my food to Notion Agent and trigger this SOP with the keyword “what I ate today.” Notion Agent will automatically analyze the food items in the image, log calories, carbs, fats, and other data, and save everything to the designated database.

It’s worth noting that the calorie estimates produced after image recognition are not completely accurate—they should be treated as a reference. However, identifying the types of food in the picture is quite straightforward, so… could this be used to build a dietary evaluation system?

Suppose I am a patient with diabetes. I create a note called “Type 2 Diabetes Personal Health File,” place this note inside the required documents for the “Diet Log” workflow, and instruct the Agent to always compare any recognized food against the “restricted foods” list in the health file, and to clearly explain the food’s impact on blood sugar in its feedback:

After eating, I send the photo to the Agent and trigger the workflow with the keyword “what I ate today.”

Here is the feedback the Agent gives me:

1. It logs the dietary data
2. It provides clear health warnings

Although AI models inherently have the ability to offer general medical advice, integrating personal health records and medical reminders greatly increases the relevance and usefulness of the output.

From the above examples, we can see that the true power of Notion Agent lies in the combination of “keywords + sub-documents.”

By using trigger keywords, the Agent enters a specific scenario, and the sub-documents nested inside that scenario (such as the health profile) further refine the execution rules. At the same time, this health profile is just a Notion page — easy to edit and adjust at any time. This layered structure allows the Agent to remain general-purpose while still becoming highly specialized when needed.

If you’ve used Claude’s Skill feature, you’ll find the logic very similar — both follow a progressive loading approach based on “keyword trigger + sub-doc execution.”

By comparison, Notion Agent’s limitation is that it can only call tools inside the Notion ecosystem, and cannot run custom scripts the way Skills can. But the advantage is that Notion Agent only needs to interact with documents — the barrier to entry is extremely low. As long as you can write a document, as long as you can articulate your idea — even poorly — you can simply keep talking to Notion Agent and let it ask you questions. Even if your answers are vague, the AI model can gradually infer your intentions and intelligently assemble the entire workflow for you.

The Boundary of Information

With scenario routing in place, the next challenge is determining where information comes from and where it should go — a concept I call the “boundary of information.”

If you ask the Agent to generate a weekly report, it needs to know where to read this week’s task data. If you ask it to store a new idea, it needs to know which database to save it in. If every time you have to manually specify “read from this database” or “save to that database,” the use cost becomes far too high.

And without clearly defined information boundaries, the answer quality will inevitably drop, because Notion Agent has access to a huge amount of workspace data.

At the same time, we cannot predefine every possible rule in the Agent document, such as:
“If the user asks A → read page X; if the user asks B → read page Y.”
That would be exhausting to maintain and inflexible. So my solution is structured database design + scenario presets.

For example, I have an Agent Scenario F that automatically generates daily, weekly, and monthly reports. The trigger keywords look like this:

In the execution document for Scenario F, this is how I define the sources of information:

With this setup, when I say “Generate this week’s report,” the Agent immediately knows:

• where to query data (which specific databases)
• what filtering conditions to apply
• that it should not search unrelated pages or other databases

Here is the query result:

After retrieving the necessary information, I then tell the Agent how to process it:

Following that, there are additional rules for analyzing and handling the data — but the core outcome remains the same:
the Agent will automatically generate a complete, structured weekly or monthly report based on the predefined templates.

This clear boundary-setting brings three major benefits:This kind of clearly defined boundary brings three benefits. First, the Agent will no longer wander aimlessly through your entire workspace — instead, it retrieves information precisely from the designated data sources. Second, clear data sources mean faster query speeds, without wasting time on irrelevant content. Most importantly, you always know where the Agent is pulling information from, making its behavior predictable and controllable. And if a result turns out to be suboptimal, you can quickly identify the issue — whether the data source itself is incomplete, or the Agent’s extraction logic needs adjustment.

But all of this relies on one essential foundation: Your Notion workspace must be built on structured databases:

• Tasks have a dedicated home
• Notes are stored and categorized by type
• Projects follow an organized hierarchy
• Saved articles have a consistent clipping hub

In other words: The power of Notion Agent depends entirely on the organizational strength of Notion itself. If your workspace is a mess, the Agent cannot perform well — no matter how advanced the model is. Most people find Notion AI “not useful” for two fundamental reasons: They don’t record enough information. Their workspace lacks structural clarity. Only when you have both rich content and a well-designed structure can Notion Agent unleash its full potential.

Once we’ve solved where information comes from, the next step is to solve where information should go — using the same approach.

If every interaction with the Agent still required you to manually specify which database to save into, which fields to fill, or which tags to set, the experience would be terrible, and true automation of information flow would never happen. So my solution remains: preset storage rules + intelligent field filling.

The strength of Notion Agent lies in the fact that it can not only create pages inside databases, but also understand the structure of a database and intelligently populate its fields.

Continuing the monthly report example:
The Agent’s generated report doesn’t sit in the chat window waiting for me to manually copy it — it is automatically saved into “My Notes DB”, because in the workflow document I have already specified:

1. The storage location for monthly reports
2. The format template for monthly reports

After accessing the “Notes Database,” the Agent interprets the semantics of its fields. It knows, for example:

• Exp = experience review
• Idea = inspiration
• Log = log entry

So the monthly report is automatically tagged as Exp. This semantic understanding is what makes the entire information flow truly automated.

Looking back at the previous two sections: “Scene routing” solves how the Agent should think.
“Information boundaries” solve where the Agent should look and where it should write. Only when these two are combined can the Agent be both smart enough to understand your intent and constrained enough to avoid mistakes.

Behind both design principles is one shared philosophy: The more automated the system becomes, the more it needs clear boundaries. If you don’t constrain anything, the Agent’s behavior becomes unpredictable — you’ll never know where it will pull information from or where it will save the output. But once boundaries are clearly defined, the Agent becomes controllable and predictable, and debugging becomes easy.

Of course, these boundaries are not permanent. As your workflow evolves, your database structures change, or you discover loopholes in certain scenarios, you can update the rules at any time simply through conversation. This “iterable rule system” allows Notion Agent to combine the reliability of structured systems with the flexibility of AI.

Custom Agent

This next part involves Notion’s upcoming Custom Agent feature, which has not yet been officially launched — and which I currently don’t have access to. So the following is based on publicly shared information, but enough to explain what it is and what it can enable.

Everything discussed so far — scene routing, boundaries, document-as-rules — operates within the Personal Agent model. Meaning: the Agent only acts when you initiate the request. You must open Notion → open the AI panel → type the instruction → wait for the result.

But Custom Agent attempts to answer a different question: Can an Agent run automatically in the background, without me manually triggering it each time?

Imagine scenarios like:

• 9 a.m. every morning — the Agent scans your task database and compiles a list of today’s due tasks, then pushes it to you
• Friday afternoon — the Agent automatically reads this week’s completed tasks, generates a weekly report, and saves it in the designated database
• Every weekend — the Agent crawls the web for the latest AI news and compiles a digest into your clipping database

This is the core value of Custom Agent: upgrading from “you ask, it answers” to “it acts proactively.”

Reviewing the Agent design principles introduced earlier, Custom Agent is essentially an extension of the same logic:

1. The logic of scene routing still applies — except the trigger shifts from “keyword detection” to “time- or event-based triggers.”
2. Information boundaries become even more important, because an autonomous Agent must know exactly where to read from and where to write to.
3. The philosophy of “documents = rules” remains unchanged. You still define the Agent’s behavior by writing documents.

If you are already using the Personal Agent and have built solid scene documents and information structures, upgrading to Custom Agent in the future will be extremely smooth: you only need to convert tasks that previously required manual triggering into automated triggers.

In everyone’s workflow, there are countless repetitive, predictable tasks — daily summaries, weekly reports, data cleanup, information syncing, periodic reviews… None of these tasks are hard individually, but precisely because they’re easy, they are often delayed or forgotten. Custom Agent transforms these “should do” tasks into “automatically done” tasks, allowing your energy to focus on work that requires creativity.

Of course, this also places higher demands on the organization of your Notion workspace. A messy, unstructured database cannot benefit from Custom Agent, no matter how powerful the feature is. So if you’re interested in this feature, now is the perfect time to start cleaning up your information structure and preparing for the future.

Agent Design Template

If you are completely new to this, the previous sections may feel scattered or complicated — but the core idea is actually very simple.

A typical Agent document contains four basic modules: Identity & Mission , Interaction Style, Scenes & Trigger Words, Memory Area . You don’t need to write everything from day one. Start with one scenario you use the most, test it in practice, then gradually expand.

There’s also a much easier way to get started — feed the AI with your past notes.

Many people feel lost when facing an empty Agent document. They don’t know how to define their “identity,” describe their “style,” or articulate their “values.” But the truth is: you don’t need to invent these out of thin air. Just dump all your old notes, articles, project reviews, random thoughts — everything — into the AI. Let the AI analyze and extract patterns, then generate a profile of you. It can infer your communication style, areas of expertise, and the standards you use to judge good work — all from your writing.

This is the idea of using existing material to bootstrap the new system, which makes starting effortless, fast, and — most importantly — authentic. Because the content is originally yours; the AI is only organizing it.

This also highlights a deeper principle: In the AI era, recording is infrastructure. Only with a habit of documenting your work and thoughts can you provide material for the AI to analyze now. It’s never too late to start — you never know what new AI tools the future will bring. No matter how advanced models become, they are not mind readers; they still rely on the material you feed them. No input, no output.

To help you get into this loop more quickly, I created a Notion Agent starter template that you can copy and use directly. Click here to get the template link.

All you need to do is follow the structure and instructions in the template, start talking to your Agent, and then gradually adjust and refine it through real usage. Add new sub-documents, tweak trigger keywords, and supplement your own methodologies and preferences according to your work scenarios.

Of course, if you want the Agent to perform at its full potential, a structured Notion workspace is a prerequisite. If you haven’t yet built your own information architecture — or you’re unsure how to organize tasks, projects, and notes — you can refer to my FLO.W template. It includes a clear pre-designed database structure: tasks, projects, notes, and saved items each have their own dedicated storage, and every field has been refined through repeated iterations so that the Agent can understand and use them right away. You won’t need to build your information system from scratch; the template itself is the foundation for unlocking Agent capabilities.

This template has already been included in the Minority Co-Creation Project — you’re welcome to explore or purchase it:

1. Minority Co-Creation — FLO.W Template Purchase
2. Full workflow video walkthrough
3. 10,000-word deep dive into the template’s underlying design principles

Notion AI Subscription Recommendations

Most of what Notion Agent can do has already been covered in the previous sections. But given how wildly diverse Notion’s capabilities are, I’ve also compiled a list of things that Notion Agent cannot do. You can refer to this link for the complete list. Before subscribing, you should review this document to evaluate whether it meets your needs.

In addition, Notion’s official pricing strategy has already undergone one major adjustment. Now, if you want access to Notion’s AI features, you must subscribe to the Business or Enterprise plan. For personal users, the Business plan is sufficient — but the yearly cost of $240 is certainly not cheap. Therefore, my recommendation is: If you are new to Notion, do not subscribe to Notion AI right away. Instead, read my article “5 Beginner Tips for Notion” first and see whether Notion’s way of capturing and organizing information feels natural to you. Only after you truly feel that Notion is a good tool for you should you consider subscribing.

Some users may notice that they can still subscribe to Notion AI as an add-on while staying on the Plus plan. According to Notion, this policy applies only to legacy subscribers. As of May 2025, the AI add-on is no longer available for purchase by new users. This means only users who subscribed to the AI add-on before the policy change can continue using AI in this way. However, the AI features available via add-on are incomplete, and you will be missing several major capabilities:

• AI Agent: A personal AI assistant capable of multi-step tasks
• Enterprise search: Global search across workspaces and connected apps
• AI meeting notes: Automated voice transcription and meeting summaries

Also important: Once you cancel the AI add-on, you cannot re-enable it. You will only be able to regain full AI functionality by upgrading to the Business or Enterprise plans.

Conclusion

Most people’s relationship with AI still stays at the stage of “open a chat window when I have a question”: use it, leave it, and come back next time as strangers again.

But Notion Agent gives me a different possibility: AI is no longer an external tool, but a collaborator that can be trained, shaped, and grown alongside you. It won’t keep asking, “What format would you like?” because it already remembers your preferences. It won’t wander aimlessly around your workspace, because you’ve already told it where to look and where to store things. It feels almost magical — like training an assistant who becomes more and more in sync with you over time.

Of course, all of this only happens if you’re willing to invest time to build, refine, and iterate this system — and more importantly, if you’re willing to get your hands dirty and record things honestly. Notion Agent is not plug-and-play magic. It requires you to think clearly about how you work, and express those rules in documents. This process itself becomes a form of self-reflection; you’ll discover habits and preferences you never realized before.

This leads to another point I want to emphasize: as the performance gap between AI models shrinks, what truly determines output quality is the input you give them.

And “input” does not mean the overwhelming flood of second-hand information everywhere on the internet. It is not clipping other people’s articles or collecting other people’s ideas — those might even be AI-generated leftovers. The inputs that matter are your own:
your biases, your ignorance, your narrow perspective, your clumsy processes, the wrong turns you’ve taken, and the mistakes you’ve made.

Only when you honestly record these “imperfections” can AI truly help you. Because then it is no longer looking at generic, mass-produced correct answers — it’s seeing your unique thinking patterns. It learns your real confusion through your mistakes, understands your real needs through your preferences, and recognizes your true standards through the revisions you make again and again.

That’s why I strongly disagree with the “AI era makes note-taking useless” argument. To me, that is resignation in the face of technological change — an excuse for laziness. No technological revolution can replace independent thinking. AI can execute, organize, accelerate — but it will never decide for you what you actually want.

The greatest value of Notion Agent, to me, is that once I saw the huge potential of automated workflows and the clear path to building them, I realized I truly couldn’t delay any longer. I must seriously rethink my workflows:

• What repetitive processes are draining my time?
• How can those be optimized or automated?
• What can be redesigned so I can get more done with the same time?

Once I think through these questions — that is when Notion Agent becomes powerful.

Lastly, this article was originally planned to include comparisons with Obsidian’s AI plugins and Heptabase’s newly redesigned AI (several iterations in), and also dig deeper into the current limitations of Notion AI. But since this article is already over ten thousand words, I’ll save those topics for next time.

If there is anything specific you’d like me to cover, feel free to leave a comment — I will evaluate it and consider including it in upcoming articles.