Why Most People Struggle to Use AI: Transitioning from Passive to Active Paradigms

The cover image is generated by Recraft. Click the original link to try this workflow.

Current Situation Observation

Recently, while developing the AI workflow product “Vector Context,” I observed an interesting phenomenon. Although I have detailed the methods for using AI and taught how to write prompts hands-on, very few colleagues have truly integrated AI into their daily work. Most colleagues just dabble, using it a few times a month, and some feel that “using AI is not as fast as doing it myself.”

This phenomenon is often echoed when I communicate with other companies. Everyone talks about AI and believes it is important, but only a few have turned AI into a work partner. More people remain in the trial phase: hearing that the latest AI is good, registering an account to try it out, asking a few questions and finding it interesting, and then gradually forgetting it in some browser tab.

Why is this happening?

The Dilemma of Passive AI

At first, I thought it was because AI is not powerful enough. However, after a year of in-depth research and development on AI products, I gradually realized the problem might lie elsewhere: the vast majority of current AI products adopt a “passive” interaction model.

What is “passive”? Simply put, AI is always waiting for the user to initiate a conversation or action. You have to actively open the ChatGPT page, think about what questions to ask, organize the language to send a request, and only then will AI start working. When writing in Notion, you have to actively press the “/” key and select AI-related functions for it to help you. It’s like an assistant who is always waiting for you to give orders; if you don’t speak up, they will remain quietly in the corner.

But which kind of assistant do you prefer? One that you have to instruct for every task, or one that is perceptive and anticipates your needs?

Generated in the 2D art poster style using Recraft. Click the original link to try this workflow.

Clearly, it’s the latter. But why do most current AI products adopt a passive interaction model?

There’s a very realistic reason: the cost of tokens. At the beginning of 2023, the cost of calling AI models was relatively high; a normal conversation could cost several dollars. If AI were to continuously analyze and process in the background, the expenses would be considerable. Therefore, product design had to adopt this passive scheme, waiting for users to initiate requests before calling AI.

However, the situation is beginning to change. AI companies are competing fiercely, and while model performance is improving, the price of tokens is continuously dropping. Just like the emergence of Arduino lowered the barrier for hardware development, making “everyone a maker” possible, lower-cost AI models are also creating conditions for the popularization of “active AI”.

Model	Year	Input Price	Output Price
gpt-4-32k	2024	$60	$120
gpt-4o	2025	$2.5	$10
DeepSeek	2025	$0.14	$0.28

This means that this year you can obtain an AI model more powerful than last year’s strongest for a cost of only two thousandths (DeepSeek compared to gpt-4-32k). The cost has changed dramatically, but what about the product’s interaction paradigm? Currently, most AI applications we see are still primarily chatbot types. The benefit of chatbot applications is that users have virtually no learning cost; they can use it just like everyday chatting.

But no learning cost does not mean no cognitive cost.

Do you remember your first time using Excel? You might have thought it was just a tool for creating tables. Until one day, you saw someone using Excel pull down lightly, and the entire column of data filled itself according to the pattern, you suddenly realized: Excel can be used this way!

This is a typical cognitive cost issue: even if Excel’s smart fill function is very powerful, if users are unaware of its existence or do not think to use it while using, that functionality may forever be nonexistent to them.

Today’s AI products are no different. I often see scenarios like this:

A colleague is collecting and organizing information, searching through a vast number of pages via search engines, struggling to edit and write. Clearly, we have introduced various AI tools, but they completely forget that they can ask AI for help. When I remind them, “Why not try using AI?” they suddenly wake up: “Oh right, I forgot I could use AI!”

Another colleague is categorizing product images manually, classifying and tagging images across different dimensions. I asked, “Why not use AI for classification?” They replied, “Huh? AI can understand images?”

For customer service colleagues, I specifically designed workflows to query tickets, access professional knowledge, and optimize response language, but from the backend observations, we found that usage frequency gradually decreased after a few days. Upon inquiry, many customer service colleagues felt it was still faster to answer customer questions directly themselves.

This reflects the reality for most people when using AI:

1. They either completely forget that AI is an option.

2. They don’t know what AI can do.

3. Or they feel that using AI is more cumbersome than doing it directly.

Even if they know they can use AI, they still face a series of cognitive burdens:

• They have to open a new webpage or software.

• They need to think about how to describe their needs.

• They worry whether AI will understand incorrectly.

• They need to check whether AI’s output is reliable.

• They have to copy and paste the results back into the working interface.

Each of these steps seems simple, but combined, they lead people to think: “Forget it, I’ll just do it myself.”

It’s like having an assistant in another office; every time you need help, you have to:

1. Walk to their office.

2. Explain in detail what you need.

3. Wait for them to finish and check if it meets your requirements.

4. If it’s wrong, you have to explain it again.

5. Finally, you need to bring their work back to your office.

Over time, unless necessary, who would actively seek out this assistant for help?

So when we say, “Most people struggle to use AI,” it’s not a question of technical ability, but rather that the current design of AI products imposes unnecessary cognitive burdens on users. Users need to constantly remember, “I can use AI,” need to actively switch to another tool, and need to learn how to communicate with AI—these are all stumbling blocks preventing AI from truly integrating into daily workflows.

So how should we design to allow users to seamlessly use AI?

Transitioning from Passive to Active Paradigms

Let me start with an excellent product design example: Cursor AI. Cursor AI is an AI-assisted code editor.

The scene of AI-assisted programming has existed for a few years, with basic operations including auto-completion and AI dialogue generation of code.

How does Cursor AI, as a newcomer, stand out in these basic operations and gain favor from so many developers?

First, regarding the auto-completion function, Cursor goes a step further, not only predicting the code that needs to be completed but also predicting which line you might need to edit next (or which several lines), automatically moving the cursor to that line, saving you the time of moving the cursor using the mouse or keyboard. Don’t underestimate this little bit of time; for developers, it allows us to simply hit the Tab key to achieve auto-complete -> move to the next position -> continue auto-completing. Over time, this can save a lot of time. Cursor calls this Just hit tab.

For the standard feature of code generation, Cursor also goes a step further. Not only does it generate code, but it has also trained a proprietary model specifically for handling how to insert AI-generated code into the user’s files. It’s important to note that a user’s code can be lengthy; after analysis, AI may find that the user’s new requirements only require changes in two places, but those two places may be far apart. If AI cannot automatically help users insert the modified code into the designated locations, users would have to manually copy and paste. Cursor saves this step for users, automatically helping them find where new code should be inserted/modified. Cursor calls this Instant Apply, and they have written a blog to introduce this proprietary model.

Additionally, Cursor has a feature called Shadow Workspace. It’s not an intuitive feature, but it can greatly improve the development experience. At times, it may even make you exclaim: how did you know to change this! Simply put, this feature opens an invisible workspace, allowing AI to continuously observe and understand the user’s code in the background. This is not just analyzing a single code file currently being edited, but analyzing the entire project. You can imagine it as having an expert sitting next to you while you write code, who not only knows what code you are editing but also understands the structure of your entire project, and when you need it, they will prompt you: this should be written differently. This is an excellent example of an “active” AI assistant, not waiting for me to ask before responding, but working alongside me while I work, providing prompts when necessary.

So how can we transform existing tools according to the design of “active” AI?

For example, in the earlier mentioned Excel smart fill example, to use the smart fill function, you need to select content and then pull down. This operation is not intuitive for a user who only knows basic Excel operations; they may not think of such a feature at all. But if we change our thinking: AI continuously observes the user’s actions in the background. When AI detects that the user filled in the first row with student number 20240101, and the second row with 20240102, it can immediately recognize this pattern. Instead of waiting for the user to “discover” the smart fill function, it would be better to actively display a semi-transparent preview in the following cells. If the user agrees, they confirm; if not, they continue entering manually, completely not interrupting the workflow.

Similarly, this logic can be applied to other parts of the table. For instance, if I input the columns for “Name” and “Gender,” can AI not actively suggest whether I need to add columns for “Date of Birth,” “Contact Number,” “Home Address,” etc.? Or if I input a column of date data, can AI not actively suggest generating a chart to visualize this data? A truly intelligent AI should act like an experienced craftsman, always providing useful suggestions while I work on the table, rather than arriving late only when I seek help.

Another example is an intelligent browser assistant. Could it automatically query unfamiliar keywords/professional terms in the background while having some understanding of my background knowledge, organizing a mind map and related material links for me? When I see the corresponding professional term, it automatically presents me with prompts?

When the interaction paradigm of AI applications shifts from “passive” to “active,” users no longer need to constantly remind themselves to use AI tools for assistance. Active AI tools will continually reinforce users’ cognition, gradually helping them accept the concept of AI assistance.

Product Design Considerations During the Transition Period

Active AI is undoubtedly the design trend for AI products moving forward. However, during the period when most users have not yet established the mindset of using AI assistance, making significant strides can easily lead to products that are well-received but underutilized, with many users merely coming to try it out.

So how can we design AI products during this somewhat awkward transitional period? Here are some of my humble opinions.

First, let’s look at the current situation during this transition period:

1. The cost of large language models has dropped to a relatively low price, and the technical cost of active AI is manageable.

2. Users’ understanding of AI remains at the “knowing” stage, without forming a daily usage mindset for AI.

Therefore, I believe that AI products during the transition period can consider integrating active AI into existing products and tools that users are familiar with. Without interrupting users’ current workflows, we can continuously enhance users’ cognition through active AI, letting them know the capabilities and boundaries of AI, thus gradually transitioning to the next phase.

An old example comes from the cash register at 7-11 convenience stores. To collect customer data for analysis while minimizing the cognitive burden on cashiers, 7-11 directly designed quick buttons on the cash register to record customer gender and age. The cashier simply makes a judgment and presses the button. There’s no need to record specific customer information in another notebook after the customer leaves. This design does not interrupt the original workflow and tries to combine the design with the original scenario, making it easy for cashiers to use.

Therefore, instead of designing completely new AI products, it’s better to think about what existing products and tools can be improved, using active AI to assist users. This may be a more suitable research direction.

Design Principles for Active AI

What should be noted in the design of active AI?

Product Design Principles

1. Do not disturb the user’s workflow. Active does not mean forced; AI’s suggestions should be like sticky notes—visible at all times but easily ignored. Active AI should not overshadow the user’s normal use of the product.

2. Accurately grasp the timing. Just like an excellent assistant knows when to proactively report work and when to remain quiet, AI also needs to learn to provide help at the right moment and in the right place.

3. Provide clear control options. Allow users to easily adjust the degree of AI’s proactivity; some users may want more proactive suggestions, while others may prefer a quieter working environment.

4. Establish a comprehensive feedback mechanism. One benefit of active AI is that you can clearly know from the backend whether users have accepted the suggestions given by AI, as accepting AI suggestions requires users to perform an action to confirm. This behavioral data must not be overlooked, as it can provide substantial help for product optimization.

5. Continuously learn user preferences. By recording users’ acceptance of AI suggestions, continuously optimize the frequency and content of recommendations to achieve true personalized service.

6. Leave the final decision-making power to users. Until the illusions of large language models are resolved, AI can never fully replace human decision-making. Most products should leave the final decision-making power to users, who should be able to take over at any time to make final decisions.

7. Privacy information handling. With active AI, unlike passive AI, users cannot intuitively see what information is being processed by AI. Therefore, it is necessary to inform users in advance about which information will be accessed by AI to achieve active AI effects.

Technical Implementation Points

1. Strictly control costs. Due to the design of active AI, costs may significantly increase compared to passive AI. Therefore, design must strictly control costs. There are many ways to control costs, such as:

• Automatically matching different large language models based on task difficulty. There’s no need to pursue the strongest model; simple tasks can be handled by smaller models.

• Adjusting the structure of prompts to utilize many caching features provided by large model vendors. For public parts, try to abstract them into a single dialogue round to hit the cache. For example, in the case of DeepSeek, the cache hit prompt costs only 0.1 RMB per million tokens.

1. Asynchronous processing mechanism. Active AI needs to run analysis continuously in the background. To avoid affecting the user’s normal operation experience, an asynchronous processing mechanism must be adopted. The tasks running in the background can not only provide active AI suggestions but also continuously perform preprocessing analysis on more user data to provide more accurate suggestions at the right time. For example, tasks such as document segmentation, formatting, summarization, keyword extraction, and terminology organization.

2. Layered architecture design

• Sensing layer: Real-time capture of user behavior and context.

• Analysis layer: Understand user intentions and needs.

• Decision layer: Determine whether to proactively provide help.

• Execution layer: Generate and display suggestions.

1. Reasonable context length. Design the content that needs to be included in prompts reasonably. Simply cramming all conceivable content into prompts is easy but leads to high costs and increased initial token delays. Therefore, the second point above can come into play here; some non-essential information can rely on previous preprocessing summaries as substitutes, reducing length without losing information.

Attempts with Vector Context

The “Vector Context” I developed is a no-code workflow platform that also faces the issue of user learning barriers. Therefore, I am currently trying to add active prompts on the workflow canvas interface, allowing AI to suggest that users consider using a certain node or connection, while also providing reasons to facilitate user understanding. I hope to reduce the design barriers for users through this approach.

In terms of workflow usage, although we have developed browser plugins and PC software, users still need to actively invoke them. According to the active AI design paradigm discussed in this article, I hope that in the future, AI can automatically determine when to invoke which workflow to assist users in their current work, making Vector Context a versatile and intelligent active AI assistant.

Final Thoughts

Looking back at the entire development history of computers, every significant paradigm shift has been accompanied by innovations in interaction methods. From command lines to graphical interfaces, from keyboard and mouse operations to touch screens, each change has brought computers closer to users.

Now, AI is ushering in a new wave of revolution. However, we must think not only about how to make AI more powerful but also about how to make it more human-centric. Just like a newcomer to the company filled with fresh knowledge, their abilities may be strong, but if they do not learn to be proactive and provide help at the right moment, their value will never be fully realized.

The transition from passive to active AI is not just an improvement in product design; it is also an important step towards AI becoming an “assistant” rather than just a “tool.” When AI begins to actively observe, think, and suggest, we draw closer to the true concept of an “intelligent assistant.”

I remember when I established the Tsinghua Maker Space, I proposed the slogan of lowering technical barriers: “Make everything, everyone a maker.” Today, when we discuss the design of AI products, we are essentially still lowering the usage barriers of technology, allowing more people to easily use and experience AI. Future AI product design should not wait for users to “learn to use” it but should actively “understand users.”

After all, the first step toward benevolent technology is to make technology more in tune with human hearts.

I am Maker Bi, and I’ll see you next time.