Last week, I had dinner with a friend who works as a product manager at an internet company. Halfway through the meal, he suddenly slammed his phone on the table and sighed:
“I’m going crazy. I have ChatGPT, Claude, Notion AI, Perplexity, Gamma… seven AI tools in total. I spend half an hour every day just switching between them. And the result? My productivity is actually lower than before.”
I put down my chopsticks and asked him a question: “Have you ever thought that what you lack isn’t more tools, but figuring out which tool is best suited to your hand?”
This isn’t a joke. Microsoft’s 2025 research data shows that 75% of knowledge workers worldwide are using AI tools, with heavy users reporting efficiency improvements as high as 93%. But on the other hand—a Zapier report indicates that about 80% of companies struggle to effectively integrate AI into their workflows.
The abundance of tools has ironically created a paralysis of choice.
Today’s article isn’t about “15 AI Gadget Recommendations”—you’ve probably seen those a hundred times already. I want to talk to you about a more fundamental issue—how to choose the two “knives” that are truly right for you, and then use them effectively.
01 First, look in the mirror: Where exactly is your workflow stuck?
Most people choose AI tools based on this logic: This one’s popular, install it. That one’s new, install it. A friend recommended it, install it.
This isn’t choosing tools; it’s collecting junk.
A professional chef doesn’t buy a knife just because it looks nice; before buying a knife, they think, “What do I need to cut?”
Similarly, before opening any AI tool, you only need to do one thing: find the most painful point in your workflow.
I’ve broken down a typical professional’s daily work into four core stages. See which one you’re stuck on:
Stage 1: Information Input. You spend a lot of time reading reports, searching for information, and digesting it, but you forget it immediately afterward, unable to find the key points.
Stage 2: Content Output. Writing documents, emails, proposals, and PowerPoint presentations—writing itself isn’t slow; what’s slow is the process of going “from a blank page to the first draft.”
Step Three: Communication and Collaboration. Too many meetings, and you don’t know what was said afterward; too many emails, and you have to write each one yourself.
Step Four: Task Execution. There are many tasks, and you don’t know where to start; large tasks can’t be broken down, and you just stare blankly at them.
Don’t be greedy. You can’t expect AI to save you in all four steps. Choose the most painful one, and then only provide tools for it.
For example, if your pain point is “three or four meetings a day, and I forget everything afterward”—then you need a good meeting transcription tool, not an AI that can write code.
Targeted solutions are a hundred times more effective than casting a wide net.
02 Decluttering: The Three-Piece Rule to “Slim Down” Your Toolbox
Once you’ve identified your pain point, the next step is to simplify.
My principle is called the “Three-Piece Rule”: Keep a maximum of three items in your daily AI toolbox. Anything beyond what’s needed should either be deleted or sidelined.
It’s not about being stingy. It’s that each additional tool adds another layer to your cognitive load. Switching costs, learning costs, monthly bills—all of these combined can easily eat away at your efficiency gains.
Below are four “three-piece” combination solutions for you to choose from, based on your role:
Solution A: Document-Oriented Worker (Product Manager/Operations/HR/Administration)
📌 Claude — Full-text writing and long document processing; large context window, suitable for writing proposals, reports, and polishing copy.
📌 Fathom (Free) — Meeting transcription + automatic summarization; unlimited free personal version; automatically extracts action items after a meeting.
📌 Goblin.tools (Free) — Breaks down vague, large tasks like “The boss said to make a proposal” into actionable subtasks.
Solution B: Programmer/Technologist
📌 Cursor or GitHub Copilot — AI-assisted coding; Cursor excels at cross-file refactoring, Copilot excels at high-frequency inline completion.
📌 Claude — Technical documentation writing and code review summarization; not all AI is suitable for writing technical documentation.
📌 Perplexity — Searching for technical documentation and solutions yields results with cited sources, not fabricated ones.
Option C: Sales/BD/Client Communication
📌 Otter — Use AI Chat to follow up on meeting content like in a conversation, “How was that client’s quote negotiated last time?” Instant answer.
📌 ChatGPT — Draft personalized emails and WeChat messages, with customizable tone.
📌 Notion AI — Turn client information and follow-up records into a searchable knowledge base.
Option D: Creative/Content Creators
📌 ChatGPT — Strongest in brainstorming, topic selection, and multimodal capabilities.
📌 Claude — Deeply polished, consistently styled content.
📌 Zapier (Free) — Automates repetitive processes like “writing an article → automatically sending it to a Lark group notification.”
Note that each option has three components. Not five, not seven. Three pieces—that’s enough.
If you think, “Wait, I need XX,” go back and review Section 1: What exactly is your pain point?
03 It’s Not Just About Opening: Three Practical Tips to “Use” Tools Effectively
After choosing a tool, most people encounter the second hurdle: installation is complete, but usage remains elusive.
An app might lie dormant on your phone for three months, opened less than five times. Why? Because a few crucial steps are missing between “knowing” and “habitualizing.”
Here are three tips I’ve personally verified; give them a try:
Tips 1: Be a “unicycle” for two weeks; don’t give up immediately.
After selecting a tool, give yourself a two-week “forced usage period.” Force yourself to use it for everything you can. Don’t try three or four tools simultaneously—you might think you’re comparing options, but you won’t actually learn any of them.
For example, if you choose Fathom for meeting minutes, use it for all your meetings for those two weeks. After two weeks, you’ll find you no longer need to “think” about it—a conditioned reflex for meetings has been established.
Habits aren’t learned, they’re developed through use.
Tip 2: Create a “What I Learned” memo, not a bookmarks.
Most people use AI tools like this: discover a useful use case → get excited → take a screenshot → send it to a file transfer assistant → never look at it again.
Try a different approach. Create a note page called “What I Learned,” and jot down every practical tip you discover—not a list of features, but the specific scenario where you actually used it.
For example: “Used Claude to merge three competitive analysis reports into a single comparative summary, saving 40 minutes.”
After a month, this page will become your personal AI efficiency manual, more useful than any tutorial you’ve bookmarked.
Tips 3: Understand the tool’s “boundaries,” not “what it can do.”
Every AI tool has a “jack-of-all-trades” image in its marketing, but in reality, it’s only good at “one or two scenarios.”
For example, Claude’s context window can reach 200KB, but when you fill it with 200KB of content, its ability to follow commands noticeably decreases after 60%-70%. Knowing this will allow you to break large tasks down into several self-contained sessions, instead of a marathon window.
Knowing where a tool might fail is more useful than knowing how powerful it is.
04 Avoid These Three Pitfalls: Scenes of Failure I Witnessed in 2026
Having discussed how to choose and use tools, I want to talk about three of the most common pitfalls. These aren’t theoretical deductions—they are real-life experiences I and those around me have encountered.
Pitfall 1: “Just let the AI write it”—Send it without reading it.
I saw a colleague use AI to write a client email, the content of which looked beautifully written. He sent it without reading it. The result? The email used a brand name that shouldn’t have appeared, “filled in” by the AI from its training data.
The client replied: “Who is XXX?” The scene was incredibly awkward.
AI is a crutch, not a prosthetic limb. Your judgment is always the last line of defense.
Pitfall 2: Automation for the sake of automation—making simple tasks more complicated.
A friend spent three hours setting up an automated workflow on Zapier to handle a task that would normally only take her five minutes a week.
Let’s do the math: three hours equals 180 minutes, divided by five minutes—enough for her to do it manually for 36 weeks.
Automation is about saving time, not proving you can automate. If the importance and frequency of a task don’t justify the cost of automation, doing it manually is the optimal solution.
Pitfall 3: Using free versions to process sensitive data—you think it’s free, but it’s actually very expensive.
Many people don’t know that most free and personal paid versions of AI tools reserve the right to use your conversation data for model training. Only enterprise versions have explicit contractual guarantees that data will not be used for training.
Are you an HR professional using the free version of ChatGPT to analyze a resume containing salary and performance information? Sorry, you’d better pray that that data isn’t included in the next round of training.
Before processing sensitive company information, scroll to the bottom of the product page and take a look at the data processing agreement. Without reading this, everything else is pointless. The most ironic thing about AI tools is that—instead of giving you freedom, they’ve made many people more anxious. This is because you’ve mistaken “owning tools” for “knowing how to use them,” and “having enough space” for “being able to use them.”
In 2026, what truly differentiates you in the workplace won’t be how many AI tools you use, but whether you dare to discard those that are unsuitable.
You don’t need thirty functions on a Swiss Army knife; you just need a handy chef’s knife.