AI Use Cases - Boosting Work Productivity#

Now that we’ve covered key concepts as well as services and tools, let’s talk about what you can actually do with them. I’ll break it down into ways to boost work productivity and ways to generate business ideas.

Boosting work productivity#

As mentioned in the previous section on ChatGPT, you can use ChatGPT for information search, summarization, translation, writing, proofreading/editing, code writing and review, data analysis and visualization, image / video generation, idea generation, and problem solving. The same applies not only to ChatGPT but also to other AI services such as Gemini and Claude. Each service has its own strengths and weaknesses, and those can change over time, so it’s best to stay curious, try multiple services, and pick what fits the situation.

Information search#

In everyday life—at work, while studying, or even for hobbies—there are countless moments when we need to search for information. One of the biggest reasons the internet became what it is today might be “information search.” Traditional search services take keywords from users and list data that contains those keywords (or is judged to be closely related). Then the user explores the results to find what they need.

With AI services like ChatGPT, you can ask in natural language, as if you were asking a friend, and receive a well-organized answer. (Some services experiment with “showing multiple candidate answers and letting the user choose the most accurate one,” but it’s still unclear whether this approach will become standard.) Even if you don’t know what keywords to search for, you can describe the situation, and the AI will try to understand it, organize the problem, suggest what to do, and sometimes ask follow-up questions to give a better response. It also reduces the effort of hunting through lots of data, and since the AI can respond in the language you want, it greatly reduces translation friction.

However, because AI can hallucinate, it may produce incorrect answers—often in a very convincing way—which makes it hard to notice the mistake. So when you need accurate information, you should not blindly trust the AI’s response and should verify it.

As an example of when this can be helpful, imagine a new hire at an e-commerce company who receives an Excel sheet of orders and is asked to “add information so it’s easy to understand what each ordered product is.” The new hire doesn’t even know there’s a separate sheet containing product information and isn’t familiar with Excel functions, so they don’t even know what to search for. But they are a ChatGPT user, and by asking something like the following, they can get a direction and complete the task. Then they can look up what VLOOKUP is and learn from the official documentation.

Question: I have an Excel sheet with order history. It includes customer IDs, shipping addresses, product codes, quantities, and order dates. I was told to add information so it’s easy to understand what each ordered product is. What should I do?

Answer: With only a product code, it may be hard to understand what the product is. You can add product information using the steps below.

1. Gather product information for each code: Prepare a separate dataset (e.g., an Excel sheet) that contains product name/description/price for each product code. This sheet should include columns like “Product Code,” “Product Name,” and “Product Description.”
2. Use the VLOOKUP function: Add a “Product Name” column to the order history sheet, and enter the following VLOOKUP formula in the first cell.

=VLOOKUP(product_code_cell, product_info_range, column_number_for_product_name, FALSE)

For example, if the product code is in cell B2 of the order sheet, the product info sheet is named “ProductInfo,” and product code is in column 1 and product name is in column 2:

=VLOOKUP(B2, ProductInfo!A:B, 2, FALSE)

3. Copy the formula down: Copy the formula down to automatically look up the product name for all orders.
4. Expand with more fields: Add additional columns (description/price, etc.) using the same approach.
5. Validate the results: Verify the mapping is correct and adjust the range/column number if needed.

This will make the order sheet much easier to understand by supplementing product codes with readable product names and descriptions.

Summarization#

We live in an era of information overload. Some information needs to be carefully verified and memorized, but for many long articles, we end up retaining only a sentence or two. For example, after reading the Gemini API Developer Competition page, the one sentence that stuck in my head was: “Build an app with the Gemini API and submit it by August 12; first place gets $300,000.” In that case, it’s efficient to start by checking a summary and then dive deeper only when necessary.

With chat-based AI services like ChatGPT, Gemini, and Claude, you can paste a URL or attach a document and ask for a summary. You can also use AI browser extensions or AI-first browsers to request summaries of the web pages you’re reading. When a document is long, reading the summary first and then quickly scanning the original can help you understand the content faster.

Translation#

Translation was already widely used long before ChatGPT became popular. But as AI models evolve into multimodal models, translation becomes even more useful. In the past, systems often had separate models for image recognition, video recognition, summarization, and translation, and then connected them together—which can be slower and can degrade response quality.

In OpenAI’s GPT-4o launch demo, real-time voice conversations feel natural with barely noticeable latency, and the system can recognize a language switch and respond in the new language immediately. Reducing end-to-end latency for speech recognition and translation doesn’t just save time—it opens up new possibilities.

Writing and proofreading#

One of the strongest capabilities of LLMs is writing. Instead of merely retrieving stored data, they can understand the user’s request and context and generate text accordingly. They can write fiction, scripts, reports, proposals, names, and planning ideas.

However, if you already have a very specific piece of writing in your head, using AI purely to reduce the “typing work” is often not a good fit. To get a very specific output, you must write very specific prompts, which can take longer than writing the text yourself. So it’s more reasonable to use AI when you have an abstract structure and can delegate some details, or when you want to get ideas from the draft the AI produces. For formal business documents like emails or reports, you can generate a first draft with AI and then edit it yourself. You can also use AI to check for grammatical issues or awkward phrasing.

Code writing and review#

In some ways, coding is similar to writing. Writers write text, and developers write code. Just as AI can write because it learned from a massive amount of text, it can code because it learned from a massive amount of code. The difference is that code tends to be more structured than natural language, which reduces the chance of drifting off-topic. In other words, AI coding capability is extremely strong and useful.

I believe developers are among the groups benefiting the most from LLM-based AI services. Some people worry that developers’ roles will be fully replaced by AI, but I think the right direction is to use AI as a tool—like a “mathematician using a computer”—and spend more time focusing on what matters.

With Cursor’s arrival and the introduction of agent features, the term “vibe coding” became popular. It refers to the idea that you can build an application simply by describing what you want in natural language. While there’s still a gap between the ideal and reality, one thing is clear: the time spent on “typing code by hand” has decreased significantly. This dramatically reduces coding and debugging time and makes it easier to handle more than one task in parallel. (The gap is that for people unfamiliar with programming, the barrier can still be high. Generating an initial draft often goes smoothly, but steering that draft toward the desired direction can be surprisingly difficult without a solid understanding of programming.)

AI is also very good at explaining code. Taking over someone else’s codebase and updating it is often one of the most stressful situations for developers. Handover is rarely painless, and after the handover period ends, the previous owner may no longer be available. But as long as you have internet access, AI can answer questions whenever you need.

Data analysis and visualization#

As AI becomes integrated into data analysis tools, users can automate more tasks and perform complex analyses more easily. For example, Tableau and Power BI are improving by simplifying workflows with AI-powered auto-insights, predictive analytics, and natural-language querying. These tools play an important role in business intelligence (BI) by supporting decision-making based on real-time data.

However, advanced analytics tools can be complex, have a steep learning curve, and be relatively expensive—especially for individuals or small businesses starting with data analysis. In contrast, AI services like ChatGPT let you request analysis in natural language without learning a complex tool. That flexibility can also help with requirements that are hard to solve within a traditional BI tool.

That said, AI services can have limitations in precision and visualization compared to dedicated analytics platforms. A reasonable strategy is to start with an AI service for quick, simple analysis, and later, as your business grows and analysis needs become more complex, combine it with tools like Tableau or Power BI for more refined analytics and visualization.

Let’s also talk about Excel. Excel is an excellent tool for storing, analyzing, and visualizing data, but beginners often use it only for entering values into tables. With Microsoft Copilot, you can request tasks like chart creation, adding columns, writing formulas, or building pivot tables using natural language. You can do similar things with AI services like ChatGPT, but a key difference is that Copilot is integrated into Excel, so you don’t need to separately explain the context. With ChatGPT, you typically need to upload the file or describe the table in detail before asking.

There are also many ways to integrate AI into Excel beyond Copilot. For example, you can connect Python to Excel and apply data analysis or machine learning. Using libraries like xlwings, you can bridge Excel and Python for more advanced workflows. You can also extend Excel via Excel add-ins, use Azure Machine Learning in Excel, or use VBA to call external AI APIs to analyze data or automate tasks. These approaches can help you leverage AI in Excel even without Copilot.

Image / video generation#

Image-generation AI and video-generation AI show strong potential across many industries. Image-generation AI plays an important role in quickly producing creative visual content for marketing, advertising, game design, website and app UI/UX design, fashion design, and art. It reduces repetitive work so designers can spend more time exploring new ideas. It can also automatically generate patterns, styles, and backgrounds to simplify complex design work.

Video-generation AI can be used for automated content production, short-form social media videos, animation and game development, educational videos, and personalized video editing. It can generate and edit videos based on user requests, or convert text-based material into richer visual content for learning and marketing. A common workflow is to generate images first and then produce videos from those images—for example, creating a promotional video using product images generated by an image model.

Overall, these tools help increase the speed and efficiency of content production and make complex tasks easier. As a result, they can boost productivity and creativity, while enabling cost reduction and innovation across many industries.

HTML page generation#

With AI, you can quickly create HTML pages for a wide range of purposes. Here, “an HTML page” doesn’t mean building a full production website—it refers to a simple UI that runs in a browser (input forms, buttons, result areas, and so on). This is especially useful when you want to validate an idea before investing in full development, or when you want to build a small tool to solve an immediate pain point.

The first use case is prototyping to communicate product requirements. With slides or documents, you can describe requirements in text, but it’s hard to convey the actual “execution flow”—what the user clicks, which screen comes next, and what result they see. If you provide the spec (or a summarized version) to an AI and ask it to “build an HTML prototype that demonstrates this flow,” you can share a clickable draft quickly. This reduces the amount of guesswork for teammates and makes feedback much more concrete (e.g., “this screen needs one more field”).

The second use case is a personal tool to solve a specific need. For example, if you don’t like the default calculator installed on your computer, you can create a calculator as an HTML page with the exact features you want (frequently used buttons, keyboard shortcuts, a history panel, unit conversion, etc.). You may still need iterations to improve the quality, but getting a “working first draft” is fast enough that small, practical tools can be surprisingly easy to build.

Using Cursor for non-development work#

Cursor is primarily a tool developers use to write code, but as I used it, I started to feel it could be used effectively for many non-development tasks as well.

In software development, there’s a concept called a project—typically a folder that contains all the files needed to build an application. Cursor starts by selecting a project. In a non-development context, you can do the same: create a folder, put in the text-based files you need for work (notes, requirements, meeting notes, checklists, work logs, etc.), and then select that folder as your project in Cursor.

Here are a few Cursor capabilities worth highlighting:

• Chat with an AI agent: You can chat like you would in a chat-based AI service such as ChatGPT. A key advantage is that the agent can automatically reference the files in your project folder and use them as context.
• Browser: The agent can open a web browser, navigate to pages, type into inputs, click buttons, and understand the current page state. (However, authentication steps like logging in often still need to be handled by a person first, depending on the environment.)
• MCP integrations: You can connect external services like Notion or Google Docs tools via MCP. If connected, the agent can not only read but also write—enabling workflows like “summarize and save to Notion.” (Whether and how this works depends on the MCP server and your permission settings.)
• Request work from Slack via @Cursor: For example, you can design a workflow where you mention “@Cursor” in a Slack thread and ask Cursor to do work based on the conversation. (Availability depends on your organization’s Slack/integration setup.)

Based on the above, here’s a simple use-case example:

1. You discuss a task with coworkers in a Slack thread. (Assume the task is: “Go to a website, retrieve a table of data, and add it to a Google Spreadsheet.”)
2. In that thread, you mention “@Cursor” and ask it to write a plan document for what needs to be done.
3. You open Cursor, review the plan document, and in the chat panel instruct it to “create a task document in Notion, open the browser and perform the task, add the results to a Google Spreadsheet, and then add the spreadsheet link back to the Notion task document.”

Isn’t that exciting? To keep the example simple, I didn’t mention details like MCP setup or having to log into websites in the browser, but the scenario itself is absolutely plausible.

The downside is that Cursor is a developer tool, so it includes many features that non-developers won’t recognize, which can make it difficult to learn. I expect that products designed specifically for this kind of need will emerge soon, but until then, even if it feels a bit challenging, trying Cursor once can still be a worthwhile experiment.