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
Welcome to Session 3 of Agentic AI Without Coding. In this session we build a personal AI assistant that truly understands your own business. In the first episode of this session, we meet the three options: Custom GPTs in ChatGPT, Projects in Claude, and Gems in Gemini. We see how they differ from regular chat, and when to use which.
Step 1: The problem with regular chat: always starting from zero
Let's start with a problem you may run into often. Dimas, the admin at Sunset Coffee, asks ChatGPT to help reply to a customer asking about the price of the one-liter iced coffee and delivery to Dago. The reply is neat and polite. But look at the price: seventy-five thousand rupiah. That's made up. The AI doesn't know Sunset Coffee's prices, so it guesses a number that sounds reasonable. And shipping is only described as based on distance. So Dimas corrects it. The right price is eighty-five thousand rupiah, delivery within the city is ten thousand, and Sunset Coffee usually greets customers with Hi there. Now the answer is right. The problem is, tomorrow morning Dimas opens a new conversation, and all of that has to be typed again from scratch. Regular chat carries no memory of your business. Let's do a rough calculation. Pasting the price list, delivery rules, and tone of voice at the start of every conversation takes about five minutes. If there are twenty new conversations a week about customer matters, that pasting happens twenty times. That adds up to about a hundred minutes a week just repeating context. These are illustrative numbers, and they don't include a costlier risk: forgetting to paste the latest prices, so a customer gets an old price. The solution is a custom AI assistant. The formula is simple: standing instructions, plus knowledge files. These instructions and files are set up once at the start. After that, every new conversation with the assistant already knows your prices, rules, and tone of voice, with nothing to paste again.
Step 2: What makes a custom assistant different
Let's see how it works. The first part is the standing instructions: what role this assistant plays, what its tone is, and what its limits are. The second part is knowledge. These are files you upload, for example a service SOP, a price catalog, and a list of frequently asked questions, or FAQ. Then comes a question, from you or from a team member who uses the assistant. The result is an answer that follows your data and your brand's style. The key is that the instructions and knowledge come along automatically in every new conversation. You don't need to remind it again. When you build an assistant later, you'll fill in four parts. First, a name and description, so you and your team know what the assistant is for. For example, Sunset Coffee Customer Service. Second, standing instructions. Think of them as an SOP for the AI: working rules it always reads before it answers. We'll write them together in Episode 13. Third, knowledge files, the reference documents the assistant is allowed to read. How to prepare them neatly and safely is covered in Episode 12. Fourth, extra features, which are optional, such as conversation starters or web search. Each app offers slightly different options, so there's no need to memorize them. Now let's put them side by side. Business context: in regular chat it must be pasted every time, while a custom assistant stores it permanently in its instructions and files. Tone of voice: regular chat can drift, sometimes formal, sometimes too casual. A custom assistant consistently follows the brand style you set. Prices and rules: regular chat may guess, as we saw earlier. A custom assistant refers to the official catalog you uploaded. Team use: with regular chat, Laras and Dimas write different prompts and get different results. With a custom assistant, everyone uses the same one. One honest note: a custom assistant needs maintenance. Every time a price or rule changes, its knowledge files must be updated. Otherwise the assistant will confidently quote the old price.
Step 3: Three options: Custom GPTs, Projects, and Gems
The first option is a Custom GPT in ChatGPT. You create your own version of ChatGPT, complete with a name, instructions, and files. At the top left you can see its name: Sunset Coffee CS. Now Dimas just pastes the customer's question as it is, with no extra explanation at all. The answer is right straight away. The price comes from the catalog, Dago is recognized as within the city, the customer is greeted with Hi there, and the emoji is kept light. All thanks to the instructions and files set up once at the start. The most noticeable advantage of a Custom GPT: it can be shared by link. Other people can use it without seeing or changing the settings inside. The second option is Projects in Claude. A project is a workspace. On the left you can see the Sunset Coffee project, holding several conversations that all use the same instructions and documents. Laras asks for a draft reply to a customer asking about a sugar-free hamper, along with the ordering deadline. Claude drafts it with the price of one hundred fifty thousand rupiah, the sugar-free version, and the one-day-ahead deadline. Notice the source too: Sunset Coffee FAQ, Hampers section. An answer that names its source is easier to check. Projects feel strong for work with long documents, such as SOPs or contracts. On team plans, a project can also be shared with coworkers. The third option is Gems in Gemini. A Gem is a custom assistant you build inside Gemini, with its own name, instructions, and files. This Gem answers the opening hours from the SOP, and works out that a two hundred ten thousand rupiah order is above the two hundred thousand threshold, so delivery is free. Gems feel most natural if your business already lives in the Google ecosystem: Gmail, Google Drive, and Google Docs every day. Let's line up all three. Custom GPTs are in ChatGPT, Projects are in Claude, and Gems are in Gemini. All three have standing instructions. In Claude they're called project instructions. All three also accept knowledge files. In Claude this is called project knowledge. The most noticeable difference is how you share. Custom GPTs are shared by link or through the GPT directory. Projects are shared with team members on certain plans. Gems are shared much like a file in Google Drive. So Custom GPTs are strong when many people use the assistant, Projects are strong for long documents and teamwork, and Gems are strong for Google ecosystem users. Remember, these features change often, so always check the official page. So where do you find them? In ChatGPT, look for the GPT menu, then choose to create a new GPT. You'll see a settings screen with a name, description, instructions, and a place to upload files. In Claude, open the Projects menu, then create a new project. Once the project is open, there's a place to write project instructions and add documents to the project knowledge. In Gemini, open the Gem manager, then create a new Gem. It's similar: a name, instructions, and files. There's usually a preview panel to try it right away. The screens in this video are simulations, and the real menu names may differ slightly because AI apps are updated often. Just look for the keywords: GPT, Projects, or Gem. We'll do the full hands-on practice in Episode 14.
Step 4: When to use which
So how do you choose? Use four questions. First, which app does your team use every day? Pick the one you already use, so you don't add a new subscription. Second, who will use this assistant? Just you, a small team at the shop, or many people through a link? Third, how many documents are there, and how long are they? If you have a thick SOP and lots of files, Projects are worth considering. Fourth, where is your data stored? If most of it is in Google Drive, Gems feel the most natural. Remember, no option is best for everything. The concept is the same, so the skills you learn in one app transfer easily to another. Before choosing an app, make sure the work really suits a custom assistant. The best fit is answering customer questions that repeat every day. Writing captions and product descriptions in the same brand style is also a good fit, because the tone can be locked into the instructions. So is preparing materials, quizzes, or reports in a fixed format. Mr. Hendra, for example, could have an assistant that creates practice questions in his tutoring center's format. On the other hand, one-off tasks, like brainstorming a name for the shop's anniversary event, are fine in regular chat. No need to go to the trouble of building an assistant. And work that needs data changing every minute, such as today's stock, isn't a good fit yet. Knowledge files are fixed. For live data, we'll connect AI to Google Sheets in Session 7. Let's apply this to our example world. Laras at Sunset Coffee wants Dimas and the cashier to use the same customer service assistant. A Custom GPT shared by link suits her. Dewi at Maju Bersama Co. manages long SOPs and contracts, and her team works together. Claude Projects on a team plan feels right. Mr. Hendra at Cerdas Tutoring, keeps all his materials in Google Drive and uses Gmail every day. Gems are closest to his habits. Nadia, a social media freelancer, handles several small businesses. She can build one assistant or one project for each client, so the tone and data of different clients don't get mixed up. One important point so our expectations are right. In Session 3, the assistant works as the admin's helper. The flow starts with a customer chatting on WhatsApp or Instagram. The admin, Dimas for example, copies the question and pastes it into the assistant. The assistant drafts a reply based on the catalog and the brand style. Then Dimas checks the draft and sends it to the customer. A human stays in control. An assistant connected directly to WhatsApp that replies on its own is something we build later, in Session 8. Here's what it looks like on Sunset Coffee's WhatsApp. The customer's question arrives at twelve minutes past ten. Two minutes later Dimas sends a reply. It's the assistant's draft, which he has read and price-checked before sending. The customer orders right away for three in the afternoon. A fast reply, the right price, and Dimas didn't have to type from scratch.
Step 5: Availability, plans, and what to check
Before you start building, there are a few things you must check on the official page of the app you chose. First, can your plan create assistants, or only use assistants made by others? This often differs between free and paid plans. Second, the limits on the number and size of knowledge files. For a small business they're usually enough, but you still need to know them. Third, sharing options: is the assistant just for you, can it be shared by link, or with team members? Fourth, privacy settings. Check whether your conversations are used to train the model, and how to turn that off if you don't want it. Fifth, plan prices. I'm deliberately not mentioning numbers in this video because prices change often. Check the latest pricing page. Sixth, feature availability for your account and country. New features sometimes roll out gradually, so not everyone gets them on the same day. On top of that, there are three things to always keep in mind. First, a custom assistant is still AI, so it can still be wrong. Make a habit of checking answers before sending, especially prices and promises to customers. Second, assume the contents of the knowledge files can be read by anyone who uses the assistant. So never upload customer personal data or business secrets. We'll go into detail in the next episode. Third, the assistant needs maintenance. As soon as a price or rule changes, its knowledge files must be updated too. We'll build this habit in Episode 15.
Step 6: Material from Session 2 and the Session 3 map
The good news is, you're not starting from zero. Remember the library of ten prompt templates from Session 2? This is one of them, the customer reply template from Episode 10. The role section sets the assistant as a friendly customer service admin. The context section holds the catalog variable. Before, every time you used it, you had to paste the price list here. In a custom assistant, the catalog moves into a knowledge file. The task section still holds the customer message variable. This is the only part you'll still paste each time, because its content is always different. The format section, at most three sentences, the "Hi there" greeting, and light emoji, moves into the standing instructions. So does the limits section: never invent prices. So your Session 2 templates are the raw material for your assistant's instructions. The work you've already done isn't wasted. Here's the Session 3 map. Episode 11, which you're watching now, introduces the three custom assistant options. Episode 12 prepares the knowledge: an SOP, a price catalog, and an FAQ in tidy documents, plus what must not be uploaded. Episode 13 writes the standing instructions and the brand voice, including when the assistant should hand a customer over to a human admin. Episode 14 is hands-on: building Sunset Coffee's customer service assistant step by step. And Episode 15 tests the assistant with trap questions, improves its instructions, then shares it with the team. By the end of the session, you'll have one custom AI assistant for your own business. Before moving on, do these four small things. First, pick one app you'll use: ChatGPT, Claude, or Gemini. Second, check on its official page whether your plan can create custom assistants. Third, gather the raw material: price list, delivery rules, opening hours, and the customer questions that come up most often. It's fine if it's still messy; we'll tidy it up in the next episode. Fourth, have the customer reply prompt template from Session 2 ready. We'll turn it into standing instructions in Episode 13.
Done!
Let's sum up. Regular chat always starts from zero, while a custom assistant carries standing instructions and knowledge files into every conversation. Custom GPTs, Claude Projects, and Gems share the same concept. Pick the one your team already uses, and check its availability on the official page. In the next episode we prepare the assistant's most important material: Sunset Coffee's SOP, price catalog, and FAQ in tidy documents, plus what must not be uploaded. See you in Episode 12.
Practice exercises
Try these on your own after watching to lock in what you learned.
Pick one app (ChatGPT, Claude, or Gemini) and check on its official page whether your plan can create custom assistants.
Write down the three customer questions you answer over and over every week.
Gather the raw material: price list, delivery rules, opening hours, and the customer reply prompt template from Session 2.
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 classes1,278 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.
