Do You Really Know How to Use AI? Or Do You Only Know How to Type into a Chat Box?
Recently, I have been recruiting, and what I am looking for is not just someone
Recently, I have been recruiting, and what I am looking for is not just someone who “knows how to use AI”, but someone who genuinely knows how to “use AI to get work done”.
By Michael C.S. So | AiX Society
It sounds simple, but it was only when I actually sat across from candidates at the interview table that I realised how wide the gap is between their understanding of “knowing how to use AI” and my own.
In interviews these days, I usually start with a question:
“How many AI agents have you deployed that are currently working for you?”
Many candidates pause at this question, then answer:
“Agents? Well, I do use ChatGPT.”
So I follow up:
“Have you deployed any AI agents that run automatically on a set schedule? Or that trigger automatically and complete tasks when certain events, conditions or sensitive signals arise?”
At this point, most people cannot answer.
Because in many people’s minds, “using AI” still means opening a webpage, typing a question into a chat box, and asking AI to help write an email, revise an article, translate a document, or tidy up some information.
That is, of course, a form of using AI.
But if you ask me whether that counts as “knowing how to use AI to work”, my answer is: nowhere near enough.
I believe a person who truly knows how to use AI should not have an AI system that only “starts working after being given an instruction”.
A truly mature way of working with AI is for AI to keep working for you in the background, across different times and different contexts.
For example, before you even get up in the morning, an AI agent has already read last night’s important emails, organised today’s to-do list, and flagged the items that need priority attention.
While you are out at lunch, AI can carry on analysing data, organising documents or researching market information in the background.
When you are off work, resting or even asleep, AI can still monitor your business metrics, generate reports on schedule, track task progress, or watch for anomalies.
A more mature setup can even let AI take the initiative based on your working context.
For example, when you open a spreadsheet, AI has already cleaned, categorised and organised the data it has just received; when you mention a project in a work group chat, AI has already created the task, recorded the owner, and even arranged the meeting.
This state — where you do not need to issue instructions one by one, yet AI already knows what to do and completes the work on its own — is the AI working mode we should truly be aiming for today.
In other words, AI should not be a tool that merely waits for your commands; it should gradually become a digital colleague that keeps working in the background.
When you are on leave, it can still work; when you are asleep, it can still monitor, organise and execute tasks.
This is the real key to how AI transforms productivity.
If you ask the people around you:
“Do you use AI?”
Or:
“Do you use AI on a daily basis?”
Most people will answer quickly:
“Yes, I use it every day.”
But if you ask one more question:
“So, how many pieces of work does AI actually complete for you each day?”
Many people have to think about it.
The answer in the end tends to be:
“It helps me find information, revise text, and write a few emails.”
And there is the problem.
If your AI only does a few things for you each day — and each time it only starts working because you open the chat box and type in an instruction yourself — then in reality you are still just using AI as a slightly smarter search engine, dictionary, or writing tool.
You still have not tapped into AI’s two most important capabilities:
Autonomy, and automation.
I often tell my team that the truly difficult part is not learning how to give AI instructions; it is learning how to reach the point where “you no longer need to tell it what to do every single time”.
When tens or even hundreds of tasks in your daily work are completed automatically in the background by AI agents, without you having to manage them one by one, that is when you truly begin to grasp AI’s value.
So if I had to judge whether a person really knows how to use AI, I believe there is one very practical question:
“How many pieces of your work each day are completed by AI proactively and automatically, without you having to issue instructions one by one?”
If they can answer concretely, and the number keeps growing, that is the sign that someone has genuinely started to enter an AI-native way of working.
So how exactly does AI achieve this continuous, automatic and proactive way of working?
The following approaches are, in my view, the most practical yet most easily underestimated applications.
MCP, the Model Context Protocol, allows AI to connect to different external tools, data sources and enterprise systems.
Once AI can connect to external tools, combined with a scheduling mechanism such as a cron job, you can set up a huge number of automated tasks.
For example:
At 9am every morning, automatically pull the previous day’s sales data, tidy it up, analyse it, generate a report, and send it to a designated work group;
Check the server or system status every hour and send a notification the moment anything unusual is detected;
Every Friday afternoon, automatically compile the week’s work progress and produce a draft weekly report.
None of these tasks require anyone to remember them, nor to re-enter instructions each time.
The agent will simply run them on schedule.
This is the most basic form of “AI still works for you even after you have left the office”.
The next level is to stop using just one AI, and instead build a team of agents with different roles and specialisations.
For example:
An Analyst Agent reads data and finds insights;
A Research Agent gathers and verifies information;
A Writer Agent turns the analysis into articles, reports or presentations;
And a Reviewer Agent checks content quality, logic and accuracy.
Several agents can hand tasks to one another and complete an entire production workflow automatically.
The end result can be reports, social media content, code, or even product prototypes.
This is the real productivity gain that AI delivers.
The point is not that you do things faster yourself, but that you shift from being an “executor” to becoming a “manager of an AI team”.
When most people use AI, they only ask it to “do it for me once”.
But the more advanced mindset should be:
If this task is going to keep recurring, why not let AI build me a tool that can run automatically from now on?
For example, with a repetitive data-organising task, rather than asking AI to handle it every time, you might as well let AI write a script or automation.
From then on, whenever the same situation arises, the system runs it automatically.
In other words, a task AI completes for you today should have the chance to become a tool that runs on its own tomorrow.
It is when your thinking shifts from “using AI to do things” to “using AI to build tools” that productivity truly starts to multiply.
This is also one of the capabilities I particularly like to probe in interviews.
A mature AI system should not only work on a schedule; it should also be able to kick off work in response to certain events or conditions.
For example:
When an email related to a contract arrives, the agent automatically reads the terms, flags potential risks, and lists the areas that need human review;
When keywords such as “problem”, “deadline” or “customer complaint” appear in a work group, it automatically creates an urgent task and notifies the owner;
When a metric falls below a specified threshold, it automatically generates an anomaly report and suggests possible follow-up actions.
This kind of work is, in essence, event-driven automation.
You do not need to monitor constantly — AI can stand guard for you.
The real value lies in the system acting the moment a problem emerges, rather than waiting for someone to notice the problem before dealing with it.
The biggest difference between an AI agent and a typical chatbot is not just the interface or the name.
A true agent usually has several important capabilities:
It has a clear goal;
It can store or make use of context and memory;
It can call different tools;
It can break down tasks;
It can execute multiple steps;
And it can adjust its next move based on the outcome.
In other words, you are not just asking it a question; you are giving it a task.
It will try to judge how to complete it, which tools to use, and which steps to execute, and finally deliver the result to you.
It is this capability that brings an agent close to being a true “digital colleague”.
If you have read this far and found that you are still mostly at the stage of “opening the chat box and typing instructions”, there is actually no need to build a very complex system all at once.
The most practical approach is to start with the simplest task.
First, identify the one task that annoys you most and repeats most often each day, and try turning it into a task that an agent runs automatically on schedule.
Second, use MCP or other integration methods to connect AI to the tools you use every day, such as email, calendars, file systems, databases and work groups.
Third, try breaking one large piece of work into several roles and hand them to agents with different specialisations to complete collaboratively.
Fourth, re-examine your daily processes and find the tasks that actually have a clear “if A happens, then B should happen” logic, then turn them into event-driven automation flows.
When you start thinking this way, you will find that AI’s role is no longer just “answering questions”.
It starts to take part in your operational processes.
So when I ask in an interview:
“Do you know how to use AI?”
That is actually not the question I really want answered.
What I really want to know is:
How many AI agents have you deployed that work for you in the background?
How many of them run automatically on a schedule?
How many trigger automatically because of a specific event, condition or anomaly?
And how much of your work can AI complete on its own, without you having to remind it every time?
Because today, knowing how to type a question into a chat box is only the most basic level of AI literacy.
The people with a real competitive edge are those who can design workflows, deploy agents, set trigger conditions, build automations, and let AI run, execute and deliver results on its own.
What will truly set people apart in the future workplace is not who is better at asking AI questions.
It is who can build a “digital team” that genuinely works for them.
If you still have not deployed your first AI agent, perhaps the simplest way to start is to ask yourself one question:
Tomorrow morning, which piece of work should already have been completed by AI before I even get up?
Originally published in Chinese on HK01 (香港01). Read the original: 你真的懂得使用 AI嗎？還是只懂得在對話框輸入文字？｜蘇仲成
Recently, I have been recruiting, and what I am looking for is not just someone
Not long ago, as the host of a Metro Broadcast radio programme, I sat down
The reaction made me reflect: what people care about most is not how powerful AI
HKOEA Consultancy
Licence No. 59684
© 2026 Artificial Intellegence Application Research Society. All rights reserved
by Originals Group – Hong Kong, Malaysia, GBA