At most one email a week. Unfollow anytime with one click.
Transcript
The full narration of this video. The current sentence is highlighted while the video plays; click any sentence to jump there.
Show transcript
Introduction
Hi, and welcome back to Agentic AI Without Coding. This is episode five, the last episode of Session 1. Today we produce the first output of this class: a list of five repetitive tasks you'll hand over to AI. We map them in Google Sheets with a simple score: frequency times time times safety.
Step 1: Finding repetitive work
The first step is finding the work that repeats. Don't rely on memory. Look at the evidence: your calendar or work notes from the past week. Here's a sample work week for Dimas, the Sunset Coffee admin. Every morning, the first hour goes to replying to WhatsApp chats. Monday to Friday, the pattern is exactly the same. And that's only the morning round; chats keep coming in all day. On Tuesday afternoon, Dimas writes a whole week of Instagram captions at once, which takes about an hour. On Wednesday afternoon, Dimas checks the ingredient stock and goes shopping. And every Friday, two and a half hours go to the weekly sales recap for Laras. Seen this way, the repeating pattern jumps right out. But a calendar has blind spots too. Small tasks done in between, like logging orders from chats into a spreadsheet, never make it onto the calendar. Yet they're done many times a day. That's why we use five sources. First, your calendar and work notes, like the one we just saw. Second, chat and email history. Look for the same questions coming in again and again, such as prices, delivery fees, and opening hours. Third, your Sent folder. Emails or reports you recreate every week in a similar format are strong candidates. Fourth, the most boring tasks, the ones you often put off. They're usually repetitive tasks that don't need much creativity. Fifth, ask your team. The people who do the work know best which tasks are just copying and pasting. Often, the most boring work is invisible to the business owner. Let's put the second source into practice. These are chats coming into Sunset Coffee's WhatsApp on a Monday morning. Rina asks the price of a liter of iced coffee. Budi asks about the delivery fee to Buahbatu. Putri asks about today's opening hours. And Joko asks the price of the one-liter iced coffee again. In one hour, four questions whose answers are already written in the catalog. This is the clearest sign of repetitive work: the questions are similar, the answers are always the same, and the data already exists. If you work in a team, collect repetitive tasks with a simple Google Form. One form per task, so every team member can fill it in on their phone between jobs. Here's an example filled in by Dimas: replying to price and stock chats on WhatsApp. The next question: how often, and how long does it take each time? Dimas answers many times a day, thirty to sixty minutes per round. Then, what happens if it goes wrong? Customers get the wrong price information and may complain. This question becomes the basis for the risk score later. Finally, which data does it use: the price catalog, stock, and delivery fees. Tasks whose data is already written down are much easier to hand over to AI. If the data isn't written down yet, note it as homework: tidy up the data first.
Step 2: Completing the list with AI's help
The list from the calendar and forms is sometimes incomplete. This is where AI can help. We ask Claude to list common repetitive tasks for a business like Sunset Coffee. Notice the last sentence: don't ask for customer data. Claude gives a table of eight tasks with their frequency. Some are already on Dimas's calendar. But some were missed, such as logging orders, replying to complaints, paying suppliers, and researching competitor prices. Remember, this is only raw material. You know your own work best. Cross out what isn't relevant, and add what's specific to your business. You can use ChatGPT, Claude, or Gemini, whichever you chose in Episode 4. Move the final list into Google Sheets. Create a new file called Repetitive Work Map, with six columns: task, frequency, time, safe, score, and how. Fill in the task column first. The first four tasks come from Dimas's calendar and form: replying to chats, logging orders, writing captions, and checking stock. The next four are added from the discussion with AI: the sales recap, replying to complaints, paying suppliers, and competitor price research. Eight rows, with the other columns still empty. Eight to ten rows is enough to start. It doesn't have to be perfect.
Step 3: Scoring criteria: frequency × time × safety
Now let's score them. There are three criteria on a scale of one to five. The score one row is the weakest candidate: done once a month, finished in under five minutes, and a mistake could lose money or leak data. In the middle are scores two to four. Take a score of three, for example: two to three times a week, fifteen to thirty minutes each round, and a mistake just means apologizing. Then the score five row: done many times a day, more than an hour each round, and a mistake has almost no consequences. This is the strongest candidate to hand over to AI. The frequency column measures how often a task comes up. The time column measures how long each round takes. The more often and the longer, the more time you can save. The third column needs attention. It's called safety, the opposite of risk. A high score means low risk. We flip it so the three numbers can be multiplied directly: the bigger the result, the better the task fits automation. So a high score means three things at once: frequent, long, and safe. The formula is simple: frequency times time times safety. The maximum score is one hundred twenty-five, which is five times five times five. As a rule of thumb, a score of thirty or more is a strong candidate. Why multiply instead of add? Because one very low value, such as high risk, pulls the score down sharply. A high frequency can't rescue a risky task. On the other hand, safe and frequent tasks rise to the top, even if each round doesn't take very long.
Step 4: Filling in the scores in Google Sheets
Now fill in the scores. The first row is replying to price and stock chats. Its frequency is five, because it happens many times a day. Its time is four, thirty to sixty minutes per round. Its safety is four: mistakes can be corrected, especially if the AI only answers from the data. The score is eighty. That's the highest score in our table. The second row is logging orders. Frequency five, time two because each order only takes a few minutes, and safety four. The score is forty. Writing Instagram captions: once a week, so frequency is two, time is four, and safety is five, because you can always read a caption before posting it. The score is forty. Checking stock and shopping: once a week, about an hour, safety four. The score is thirty-two. The weekly sales recap is also only once a week, but each round takes more than an hour, so time is five. The score is forty. Then two risky tasks. Replying to complaints gets a safety of two, because it involves customers' feelings and money. The score is twelve. Paying suppliers gets a safety of one, because a wrong transfer means losing money. The score is only four. Finally, competitor price research: once a month, but long and safe. The score is twenty-five. The score column only needs a multiplication formula once in the first row, then you drag it down. If you're unsure about a score, AI can be a discussion partner. We give Claude the same scale, then ask for starting scores for three tasks, with a short reason for each. Notice the last sentence: I'll correct them. Claude gives starting scores with reasons. For paying suppliers, it gives a safety of one, because a wrong transfer means losing money. That makes sense. But for complaints, it gives a safety of three. Dimas knows the reality: complaints at Sunset Coffee are usually about refunds, and the customer is upset. So Dimas corrects it to two. The AI accepts the correction and adds a useful note: complaints can still be helped with drafts. This is a healthy division of roles. AI gives quick suggestions, and a human decides based on experience. Three tips for scoring. First, being consistent matters more than being exact. Nobody knows for sure whether a task is a three or a four. What matters is using the same yardstick for every row. Second, score with the people who do the work. Laras might think replying to chats is quick, but Dimas knows how long it really takes. Third, scores can be revised. After a task has been automated for a few weeks, you'll know whether your risk estimate was right.
Step 5: Sorting and choosing the top five
Sort the table by the score column, from highest. At the top is replying to price and stock chats with a score of eighty, far above the rest. Next come three tasks with a score of forty: logging orders, writing captions, and the weekly sales recap. In fifth place is checking stock and shopping, with thirty-two. These five green rows are our candidates. If you have candidates tied at the cutoff for the top five, pick the one whose data is most ready. The three below them don't make the list: competitor price research, replying to complaints, and paying suppliers. Look at the how column. Complaints still get help from AI as drafts, but they're sent by hand. Payments stay with a human. The how column is filled in using the three deciding questions from Episode 1: are the steps always the same, does it need understanding or writing, and does it need several apps? The answer also shows which session we'll build that task in. Before deciding, run a sanity check on every candidate. First, is the data written down? A catalog, a sheet, or emails. If it's still in Laras's head, write it down first. Second, are there rules or an example of a good result? AI understands what we want fastest from examples. Third, it's clear who reviews and approves the results. Fourth, is the benefit real: time saved, or customers answered faster? If nobody feels the difference, that task isn't worth doing first. And the red cross: if a task involves money or sensitive data without supervision, take it off the list. All five Sunset Coffee candidates pass this check. How big is the impact? Here's an illustration for Sunset Coffee. Replying to price and stock chats could free up about five hours a week, because most routine questions are answered by the agent, and Dimas only handles special cases. The other four tasks, logging orders, writing captions, the sales recap, and checking stock, add up to about five hours a week as well. That's about ten hours a week in total, more than one working day. Dimas can use that time for things AI can't do, like serving customers in person at the shop or trying out new menu items. These numbers are an illustration, and in Session 10 you'll measure the real savings in your own business.
Step 6: Session 1 output: your list of 5 tasks
To picture this for other kinds of work, here are the mapping results from our sample characters. For Dewi at Maju Bersama Co., the highest score goes to summarizing email and preparing reply drafts, which we'll build with an email assistant in Session 5. For Nadia, it's writing captions and content calendars for four clients. Nadia uses prompt templates from Session 2, followed by an automatic content pipeline in Session 9. And for Mr. Hendra, it's giving feedback on practice questions every week, with a custom assistant and a rubric in Session 3. The work is different, but the mapping method is the same. Pick the character closest to your own work as a reference. Here's the Session 1 output for Sunset Coffee: a list of five tasks to automate. Number one, replying to price and stock chats on WhatsApp, which becomes a customer service agent in Sessions 7 and 8. Number two, logging orders in Google Sheets with automation and an AI step in Session 6. Number three, writing Instagram captions with a prompt template in Session 2. Number four, the weekly sales recap, which starts getting AI help in Session 5 and could become the final project in Session 10. And number five, checking stock with an automatic, scheduled shopping reminder, also in Session 6. This list is Sunset Coffee's roadmap until the end of the class. Put it somewhere easy to see, so every session reminds you of the end goal. One more thing: the highest score doesn't have to come first. We start with the safest. Instagram captions in Session 2, because the risk is lowest and you see the results right away. Then the sales recap in Session 5, which only reads data and doesn't send anything to customers. Next, logging orders and checking stock in Session 6, using automation with an AI step. And the WhatsApp CS agent, which has the highest score, in Sessions 7 and 8, once you have enough experience. The score decides what to automate, while the class order decides when, from the easiest and safest. Now it's your turn. Gather eight to ten repetitive tasks from your calendar, chat history, team input, and AI's help. Use your notes from Episode 1 as a starting point. Take it easy; about thirty to sixty minutes is enough. Score frequency, time, and safety on a scale of one to five, then multiply them. Sort them, run the sanity check, then choose the top five. Fill in the how and session columns, then save this file carefully. We'll open it again in many of the next sessions, and calculate the real savings in Session 10.
Done!
Congratulations, Session 1 is complete. You now understand the difference between chatbots, automation, and agents, how an agent thinks, real examples and their risks, how to compare three AI assistants, and you now hold a list of five tasks to automate. In Session 2 we learn prompting for consistent results, starting with Episode 6: the anatomy of a good prompt, with role, context, task, format, and examples. See you in the next session.
Practice exercises
Try these on your own after watching to lock in what you learned.
Gather 8–10 repetitive tasks from your calendar, chat and email history, your team, and AI's help.
Fill in Frequency, Time, and Safety scores (1–5) in Google Sheets, multiply them, then sort from the highest score.
Run the sanity check, settle on the top 5 tasks, and fill in the columns for how and in which session you'll build them.
Ask the class material
Type a question or keyword; we find the part of this episode that explains it and jump the video there.
My notes
Notes are saved with the current video time and are only visible to you.
Questions & answers
Stuck on a step? Ask here; the instructor and other learners can answer.
No questions yet.
Your instructor
Cerdas Inovasi Digital Team
Software developers, BIM & 3D engineers, and AI practitioners
53 video classes1,287 episodes
We build software, BIM models, games and AI automations for clients every day, and turn that hands-on work into step-by-step classes: every step is shown on screen and comes with a PDF guide you can keep.
