DEV Community

Cover image for From Dialog Box to Cyber Assistant
熊钊
熊钊

Posted on

From Dialog Box to Cyber Assistant

Translated from Chinese.

Preface

After getting into AI, I realized there are many types — or more precisely, under different packaging, AI comes in so many forms that it's overwhelming. So in this article, based on my own experience, I'll provide a user‑perspective classification of current AI tools for your reference.

I believe the value of AI lies in collaboration and productivity. Therefore, I roughly divide the AI tools I frequently use into four categories: Chat AI, Tool AI, Search AI, and Agent AI.

Chat AI is like a knowledgeable, sharp‑minded colleague who is a bit wordy and not good with their hands. It can think with you, but when it comes to taking action, it can't get the job done.

Tool AI is like the notebook, drawing board, or toolbox on your desk. It usually works only in a specific vertical domain and has very little conversational functionality.

Agent AI is a knowledgeable, sharp‑minded, concise assistant that can also get work done for you.

Search AI is like a folder that organizes all your files for you, making them easy to find.


Chat AI

Form: Usually a relatively clean dialog box. Open it and you can start talking. Different companies' AIs have different language styles.

Market positioning: This type of AI is the most popular, most common, and has the lowest barrier to entry. Doubao, Grok, ChatGPT, and the Copilot Chat I use all fall into this category. They generally have little hands‑on ability (can't read/process files or generate files); the best they can do is generate images. But that doesn't mean they are useless — you can ask one to check the weather or crawl the entire web for information, for example: "What are the popular conversational AIs on the market, and what are their reputations?"

How to choose: Most people never push AI to its limit. The differences you can feel are mostly in the manufacturer's language style settings (warmer vs. cooler tone), the logic for handling problems (prioritizing reasoning vs. retrieval), and capability boundaries (whether it can generate light documents, context length, companionship features, safety‑restriction differences, etc.). Then there are price and ecosystem. For example, I'm a Microsoft ecosystem user; by paying a bit more I get Copilot, unlocking chat functionality, Tasks capabilities, and Copilot integration in other apps within the ecosystem.

Summary: Chat AI has a low barrier to entry. For most people, there's no huge gap in capability — the biggest difference lies in the conversational experience, unless you are a deep user like me. You should choose based on the style you like (more companion‑like vs. cooler, more talkative vs. less talkative). So if you don't have a strong urge to learn about AI, every platform actually offers some free quota. You can try them out — even use 10 free queries from one platform and 10 free queries from another — hop from one to the other and enjoy the benefits for free.


Tool AI

Form: Specializes in a specific vertical domain, such as video generation, image generation, or code generation. The advantage is that it understands your language in the direction of its own capability, making the information more focused. But its chat functionality is weak — how can a tool chat?

Market positioning: There are many types of this kind of AI. My Copilot embedded in Excel falls into this category. Why is Tool AI the type most likely to replace humans? Because they reduce the "execution barrier" in a vertical domain directly to zero. In the past, you needed to master 80% of Excel functions to complete a complex spreadsheet, or years of artistic skill to storyboard a short drama. Now, you only need clear input. It's not taking away your job — it's taking away the premium that used to go to those "who only mastered basic execution tools but lack structured thinking."

A colleague of mine used to spend 7 out of 8 hours writing code, and still might produce buggy, messy code. Now he spends 1 hour explaining clearly what he wants, then takes it easy. He writes code faster, with fewer bugs. At the same time, AI‑generated short dramas have already reduced the market demand for short‑drama actors.

How to choose: Look at how relevant it is to your work. I work with spreadsheets often, so I don't need image or video generation tools. I use Copilot inside Excel to get most of my work done. Copilot in Excel has cut my spreadsheet‑related work time by at least 70%.

Summary: Choose this type of tool based on your needs — you don't have to have one. But it is the category that truly gives you the immediate feeling that "AI is gradually replacing humans."


Search AI

Form: Usually embedded inside a search engine. It summarizes based on what you search for, giving you more efficient results.

Market positioning: This type of AI often goes unnoticed — because it doesn't converse with you; it just summarizes. You think you haven't given any instruction, but when you type something and hit Enter, it's already activated.

How to choose: There's not much to choose — it depends on which browser you use. You wouldn't switch browsers just for one feature. Moreover, it usually can only summarize the content it finds — if the search results are messy (e.g., ads, misinformation), the summary will be messy too. Don't worry too much about this choice. It's unlikely you would abandon a browser you're comfortable with just because another one has this feature.

Summary: This type of AI is usually free; it's just a feature enhancement that search engines add to their own browsers. It is the only type of AI that "doesn't invite you to participate" — all others involve some interaction with you. Moreover, it is a "non‑reasoning, summarizing AI." To some extent, it can avoid the hallucinations caused by reasoning, which is one reason Perplexity is well‑liked.


Agent AI

Form: This type of AI is usually more than just a dialog box. It connects to your email, calendar, browser, cloud storage, and other tools. You can tell it, "Check tomorrow's meeting schedule for me and send an email reminder to my colleagues." It will then look through the calendar, write the email, and send it — rather than giving you a block of text for you to do it yourself. Some agents can even execute tasks in the background and notify you when they're done. It possesses the closed‑loop ability to autonomously break down tasks and call external tools.

Market positioning: If Chat AI is a colleague who only offers ideas, then Agent AI is an assistant that actually gets work done for you. This is currently the fastest‑growing and most closely watched direction in AI, because it solves the problem of moving from "talking" to "doing." There are already many consumer‑grade agent products on the market. For example, the Copilot Tasks I use, OpenAI's ChatGPT with Agent Mode, Google's Gemini, and Anthropic's Claude have all introduced agents. In China, ByteDance's Coze lets you build your own agents (the previous ones bring the food to your table; this one is more like giving you a kitchen). There are plenty of tutorials online, and the barrier to entry is not too high.

However, this track is still very early. In practice, stability, accuracy, and the ability to handle complex tasks are far from the point where you can "completely let go." I often encounter misunderstandings, operation failures, and situations that need correction. So today's Agent AI is more like an intern that needs supervision, not a fully trustworthy veteran. But even so, it is already a big step ahead of Chat AI — at least it's willing to take action.

How to choose: Choosing an Agent AI is more complex than choosing Chat AI. Its core value isn't "how well it chats," but "what it can connect to and what it can operate." You need to see which of your daily tools it can integrate with. A Microsoft ecosystem user will find Copilot Tasks smooth; a Google ecosystem user will go with Gemini. This isn't grocery shopping — it's buying a phone. You really need to choose carefully based on your habits.


Summary

Agent AI is the most worthwhile AI tool to learn about right now. Tool AI is just a single point; Agent AI is the whole plane (though in vertical domains, it may not offer a better experience than Tool AI). It is the only category that can most fundamentally change "how humans collaborate with AI." It is still very young, and the experience is not yet mature. When choosing, don't look at whose experience is more mature — look at how well it fits your work. Even if you tell me Google's experience is the best, I wouldn't give up the Microsoft ecosystem for it.

Above are the four types of AI tools classified from a user perspective. The fastest to experience is Chat AI; the most intuitive way to feel the changes of the AI era is through Tool AI. But if you are willing to spend time learning and adapting, the efficiency gains from Agent AI are something the other three categories cannot match.

However, Agent AI often has a relatively high barrier to entry, and you may not be clear about its capability boundaries. Jumping in recklessly will only lead to unnecessary expenses. Later, I will write specific practical cases so everyone can directly experience how AI helps with productivity.


Medium@dskeke

Top comments (0)