Essential AI Skills For 2026

0:00 Here are the essential AI skills that I think you should know in 2026. Starting from beginner, intermediate to advanced. And there is one AI trend that actually really caught me by surprise, but I really think we'll be transforming [music] the world of AI in 2026, which for some reason nobody is talking about.

0:17 But before I get to that, let us start from the beginning, the [music] basics. A portion of this video is sponsored by Perplexity. Since the launch [music] of ChachiBT in late 2022, the rise of generative AI has been meteoric. So unless you literally live under a rock, I am certain that you have come in

0:33 contact with Gen AI at some point now in the form of algorithms, emails, images, videos, chat bots, products. It is literally everywhere. So if you still don't really have a grasp of what Gen AI is and how it works, I really recommend that you check out this video over here where I summarize Google's 8 hour AI

0:49 essentials course in just 15 minutes. That should be enough to get you started. Now, on top of this just general knowledge, there is one skill and a suite of tools that I [music] think any modern human that participates in society should know. Now, if you can guess what that skill is, please put it

1:03 in the comments and you'll get an imaginary cookie. Okay, did you get it? Yes, it is prompting. And when it comes to prompting, if you can just remember two frameworks, you would be better at prompting than like 98% of the population. The first framework is called the tiny crabs write enormous

1:16 iguanas framework, which stands for task, context, references, evaluate, and iterate. I'm going to put a screenshot now of an example of how to use this framework when prompting any general AI chatbot. This works well with any of your general chatbots like Perplexity, Chat BT, Gemini, Claude, Deep Sea,

1:30 Grock, etc., etc. And if for some reason you're still not really getting [music] the results that you're looking for, there is a second framework that you can try called the ramen saves tragic idiots framework. I'm going to put on screen now examples of how to use the different parts of this framework. Now, if you

1:43 want more details, explanations, and more examples of how to use these frameworks, as well as more advanced prompting techniques, please do check out this video that I have over here because I did a full video summarizing Google's 9-hour prompt engineering course in just 20 minutes. That one was

1:56 a really good course. And prompting is a skill that I highly recommend that you [music] really have a good grasp of because it is the single most important skill that will help you unlock all the tools and more advanced levels of AI. It's kind of like if you're going to be like a sword person, like a samurai, you

2:10 first got to learn how to swing your sword. Otherwise, it doesn't actually matter how good your sword is and how fancy it is if you don't know how to swing your sword, you know. Now, after you learn how to prompt correctly, there is a suite of AI tools that I think any modern human participating in society

2:24 should know. The good news is that you don't need to actually learn a lot of tools. The first tool is a generalized AI chatbot that can help you answer your questions, like what is this plant in this picture? Also, be able to help rewrite emails, simple reports, analyze pictures, data, [music] and videos. You

2:37 also probably want a tool that can help you look into news and research. For example, this is my personalized news feed. And here's an example of a AI generated research report diving deep into building products, which is what I'm interested in right now. Next up, I think everybody should master a tool

2:51 that can help you use AI to learn things faster. If there's a single most powerful meta skill that I can think of, it would be the ability to learn things quickly. And AI can really, really supercharge this for you. And finally, a tool that can help you execute and do stuff like making slides, dashboards,

3:05 simple apps. Now, the good news is that a generalized chatbot has a lot of different features now. So, most of them are able to help you like answer questions, write stuff, research for you, help you learn things and build things as well. So, it's not like you have to learn like a bunch of different

3:16 tools and like pay for a lot of subscriptions. So, you can really just choose one generalized chatbot like Chad GPT or Gemini or Claude and just like learn that really really well. But for those of you who really do want to optimize and just like get everything you can out of AI, I will put on screen

3:31 now the stack of tools that I personally use on a day-to-day basis. Like I personally use each of these tools on a day-to-day basis for specific types of task. I also do have deep dives into most of these tools that I listed here which I will also link in the description if you want to check them

3:45 out. Another tool that I've been using for the past few months that just like seamlessly integrates into my own workflow is comet which is Perplexity's free browser. I want to show you some other ways I use Comet now that is really really cool and useful. So, here is a mutual fund that I'm looking into

3:59 that I might want to invest in. And say like I see here it's like a style box and it shows this box and I don't really know what that means. So, I can type here, explain to me what does the VFAX style box mean. Okay. And it tells me a quick visual way of showing that this fund holds large cap US stocks with a

4:15 blend of value and growth characteristics rather than purely value or purely growth. And it goes to a lot more information about this. You can do this for anything that you're browsing that you're not sure about. Another example, I have an email chain going on. Uh, and there's like quite a few emails

4:28 that's going on here. So, instead of having to look through all of it, I can just say summarize this email chain and suggest next steps. And comment agent is telling me email chain covers your invitation, scheduling, recording, and post recording follow-up for your guest appearance on HubSpot's marketing

4:44 against the grain. Okay? Etc., etc. And then it says, okay, we've already done this. Preparing your own promotional plan and creating a simple internal doc. Those are the next steps that maybe I should consider. And then it says, here's for suggested next steps. Draft and send a follow-up email, ask for

4:57 estimated publish date, and then prepare your own promotion plans. So, here I noticed from this email that Ibrahim from our team has added a lot of comments into this Google doc that we're working on together. Now, instead of just like going back and forth on a Google doc, at this point, I kind of

5:09 just want to, you know, set up a meeting with him instead. So, with common agent, I can just write schedule a meeting with Ibraim for tomorrow at 9:00 a.m. Hong Kong time, 30 minutes. And Common Agent does its thing. It's able to plug straight into Google Workplace. I confirmed it and there you go. It

5:26 successfully scheduled a meeting with Ibrahim for tomorrow at 9:00 a.m. Hong Kong time for 30 minutes. Pretty handy, right? Don't need to contact switch between a bunch of different tabs. But that's not all. Here's an example in which I am having sleep problems, which is true. And I'm trying to look at

5:38 natural sleep aids to get better sleep. Telling me stuff like, "Oh, melatonin. Okay. Okay. Valyan root. All right. Uh, and then it goes like lavender. Hm. That sounds a little bit like pseudociency to me. I don't know about that." So, whenever I'm reading through an article like this, I kind of want to fact check

5:52 some stuff. Um I would just directly write check the usage of lavender for sleep for accuracy and sources. Comment was searched through perplexity showing me some different evidence that suggests okay like lavender essential oils associated with improved sleep quality in adults uh etc. meta analysis etc etc.

6:12 So you know maybe I can be convinced. Okay last example. Say you want to now buy some supplements for a friend of yours. Now, instead of going through all the different retailers, looking at their reviews and their prices, I could just ask Comet Assistant, compare melatonin 3MG supplements across

6:29 different retailers for price and delivery date. So, it's telling me that this Amazon website that I'm seeing here doesn't actually deliver to Hong Kong. But here are some other places that I can buy it from, different types of retailers, how much they cost, and tells me the best option here is that maybe I

6:44 could consider sites like I Herb, which I can attest because I have ordered from there before has good prices and good supplements. Great. Didn't have to open like 10 different tabs looking at all of them and trying to pick the right one. Comet is free right now, so if you want to try it out, check it out in the link

6:57 in the description. Thank you so much Plexity for sponsoring this portion of the video. And by the way, if you're interested in learning more about how I use AI in my daily life, literally a video called 101 ways that I use AI in my daily life. So you can check it out if you want. I think you may be

7:10 surprised by how much AI can augment your life in a lot of different tiny ways. Onto the next essential topic in 2026, AI agents. AI agents are defined as software systems that use AI to pursue goals and complete task on behalf of users. For example, a customer service AI agent is able to autonomously

7:28 answer questions from customers like through email or even phone calls while a coding AI agent is able to write code for you. So these are examples of independent agentic [music] AI products. But for the past year or so, many companies are realizing that in order to actually use these independent AI

7:43 products, they need to actually think carefully about how to integrate them into their existing workflows cuz they're not going to like just entirely change their existing workflow, right? And in a lot of cases, companies actually need to build their own custom AI solutions to really take advantage of

7:58 this agentic AI. Let me give you a couple examples because us as a company, we actually do quite a lot of B2B consulting and also building B2B AI solutions for different companies. For example, we built a custom product retention AI agent that specifically addresses when people try to unsubscribe

8:13 from their product. Now, prior to this agentic workflow, there's a very standard track. Like the customer will try to unsubscribe and they will have like one upsell saying like, "Hey, like here's 10% off this platform. Please don't go." And then that's pretty much it, right? But with this custom agentic

8:26 solution, when a customer tries to unsubscribe and they say what their reason is, we're able to craft a custom email back to them and offer them custom discounts and like credits to address the reason why they're unsubscribing. Like for example, somebody might be emailing saying like, "Oh, this specific

8:39 feature like there's just not enough credits, not good return on investment." So the AI agent can actually write a targeted email saying, "Hey, like I understand that you think that this feature is very expensive. Here's like 20 free credits that you can use for this feature." And because the email

8:52 feels personal, is targeted to that person's problem. They would be less likely to unsubscribe. And this has very much been the case. Another popular custom AI agent type that we build is reporting agents. Many companies spend significant time and resources just on reporting, reporting their earnings,

9:05 reporting stuff to their investors or shareholders, just a lot of data having to be transmitted. So when we're able to make an agentic system to help with this reporting process that integrates directly into their existing databases, this can be really really helpful especially for companies that have

9:19 specific privacy and compliance regulations like those in finance and healthcare. So the skill that I really recommend you learning is building custom AI agents. There are a lot of agentic AI frameworks that you can learn in order to build these custom agents and I'll put some of them on screen

9:32 right now. So take a screenshot if you like. Some of them are no code, some of them are code, some of them are more niche for like specific type of agents that you're building. But regardless of whatever tool it is that you're using, the underlying skill of integrating a Gentic AI in a custom way for different

9:45 companies, that is such a valuable skill and so in demand for companies, both like internally and also as freelancing. Like that is literally what we do. If you are interested in learning these skills, I'll link some videos below in the description that you can check out. I made quite a few videos on it. We also

9:58 do run like an entire boot camp, like a 28-day AI agent boot camp. So there's like way more information than I can possibly fit into this video here. So if you are interested, do check it out. I'll leave a link in the description as well. >> [music] >> Okay, so here's a thing that really came

10:13 out of left field for me. Like I was really not expecting this and that is the rise of open-source AI pioneered by China. So just to make sure that we're all on the same page here, open source AI refers to AI systems where some or all of the core components are publicly available involving things like model

10:28 architecture, model weights, training data or inference data. Licenses for use, modification and redistributions with conditions. This is in contrast to closed source AI which refers to AI systems where the models weights and training process are proprietary and users can usually only access them

10:43 through APIs, web apps or enterprise platforms. Some examples of the most popular close source AIS would include stuff like opens chatbt, anthropic claude, Google Gemini, etc., etc. Closed source AI has really been dominating the field since 2022 and primarily led by the western countries mostly in the

10:59 United States. And the reason for this is that proprietary closed source AI has just been performing much better than open source AI until if you guys remember not that long ago when Deepseek came out. And this was a really big deal because DeepSseek is open- source and it was able to perform just as well as this

11:16 closed source AI while having all of the massive benefits of open source. Open source is a lot cheaper and at scale that difference is massive. Open source also allows the user to have full control over the model. So it can be run anywhere like on prem, edge, private clouds, whatever you like. You don't

11:33 need to go through other people's APIs, other people's systems. Now they're also much more customizable. Allows you to do stuff like fine-tuning, modifying architecture, even adding your own guard rails and building on top of it. So there's a lot more freedom and independent developers can improve them

11:45 by themselves. When you use open source models, there's also no vendor lock in. Like if you want to use OpenAI's models for example, you're kind of like locked into their entire ecosystem. And finally, it's auditable. This is massive because for many regulated industries like finance and healthcare and these

11:59 places, policies and regulations require transparency, which is something that you cannot have with closed source models because they are closed source. Like we don't actually know what's going on inside the private companies, but they can use open source models and products and tweak them to the

12:13 specifications that they want. So with all oops so with open source models performing just as well as closed source models plus you have all these advantages it kind of became a no-brainer for a lot of companies to switch to open source models which is exactly what happened and what makes it

12:27 particularly interesting is that these top open-source AI models are Chinese models. If you check out this graph here open source is kind of like gaining in popularity and then there was like this massive increase and a flip in which China started dominating the open-source landscape. In fact, A16Z recently

12:43 reported a major shift where 80% of AI startups pitching them are now built on Chinese open source AI models like DeepSeek and Alibaba's Quinn. I recently also gave a talk in a venture capital firm in which they are investing a lot into open source AI. I was honestly really shocked by how fast these

12:58 developer communities were growing around these open source models and yeah, like it totally makes sense. In fact, if I were to make a prediction, and I hope I'm not wrong about this, I was very wrong. Whatever. If I were to make a prediction, I think in 2026, this whole like open-source movement is going

13:14 to like skyrocket even higher with a lot of developers moving to open source models. And because of these geopolitical factors, I actually think a lot of Western companies are probably going to start open sourcing a lot of their AI models and like surrounding frameworks and architecture as well. And

13:28 yeah, like AI as a whole might just be moving towards open source, which I think that's actually great for everybody. Lower cost, more innovation, more transparency. I think that's great. I will probably be making a full open source video for like open source models and open source products. And we haven't

13:42 publicly announced this yet, but it has been decided already that we'll be hosting a free in-person open source workshop in Hong Kong as well in January. So for those of you who are based in Hong Kong, I hope you guys can come. For those of you who are not based in Hong Kong, that is totally fine as

13:56 well because we'll be sending summaries and key insights to our mailing list. So either case, you should sign up for the mailing list. It is completely free if you're interested in our future like events and workshops and things like that. In the meantime though, I highly recommend that you check out some of the

14:08 open source AI models and products. And if you are a builder or developer, you should definitely check out open source. I will put on screen now some of the different models and products that you can try out. Take a screenshot if you like. And [music] finally, I want to talk about the last essential skill in my

14:23 opinion for 2026, which is AI assisted coding, aka vibe coding. I am genuinely convinced at this point that for anybody that wants to build a product can build a product. Whether you know how to code, don't know how to code, have any experience building products, it is so easy to build a product these days.

14:38 [music] You legit can just go on a vibe coding tool and just type what you want to build, like build a manga creator tool where the user inputs a manga concept and description characters and outputs manga and the tool will build it for you. Of course, there are things that you need to tweak and certain

14:51 things that you need to add in order to make it scalable. But you can do all of this without even knowing how to code now, which is insane. Like literally a year ago, this was not possible. like you need to be a developer to build stuff like this. And for those of you who are developers, if you're not using

15:05 an AI assisted coding tool, I don't know what to say because using a tool like this, like I'm able to code things and build things like 10 times faster. I'm also able to learn things like 10 times faster. The cool thing is that over the past year or so, there are a lot of AI assisted coding tools now. They range

15:20 from products that are built for people who have no coding experience, like very beginners, all the way to those for pro developers. A full spectrum of different products. I'm not going to go into too much detail about all of them right now cuz I don't want to make this video super long. So, I'm going to put on

15:32 screen kind of like a spectrum of different coding tools and you can like kind of try out the ones that you think fits you the best. In the past couple months, we've also seen a lot more open-sourcebased vibe coding tools which are a lot cheaper. So, I would not be surprised to see even more options in

15:46 2026. In any case, regardless of which tool that you choose, I do think that AI assisted coding, vibe coding, just like building products is very much a skill that is essential for 2026. I also have an entire video that covers the fundamentals of vibe coding. So, if you want to check it out, if you want to get

16:01 started, then you can check it out over here. All right, so before I end this video, I do want to mention a couple more trends that I really see emerging and significantly developing in 2026. I wouldn't consider them like essential yet, [music] but I do want to get them on your radar. The first one is

16:17 multimodality. And by multimodality, I mean using Gen AI for images, video, and audio. Multimodality models are just getting better and better. In my humble opinion, audio is pretty much there already. AI generated audio is virtually indistinguishable from people audio now. >> So Tina here was just mentioning how

16:34 real AI voices are getting. And well, here I am to prove it. >> Like, could you tell that that was an AI voice? Like I have a hard time telling that was an AI generated voice and AI music as well. >> This sounds really legit to me. Like granted, I'm not a musician, so maybe like if you're a musician, you would be

16:54 able to catch that it's doing strange things, but to me, I'm like, this is really good. And when it comes to images and video as well, it's just getting better and better. I think the release of Google's Nano Banana model was a very defining moment because prior to the Nano Banana model, there's this really

17:07 big issue of character consistency. Like if you generate a specific character and you try to generate like multiple versions of that character doing different things, the character would be like slightly different for each image. And in video as well, like does the frames progress, there's like slight

17:23 differences in the character, so it doesn't feel real. Now, the Nano Banana model is way better at keeping character consistency. It's still not perfect, like it's definitely not 100%, but it was a very big leap and I would not be surprised if this problem is solved in 2026. There are also a lot of open

17:39 source multimodality models that are popping up as well. And the final trend I want to touch on today is AI safety. I actually made a video about AI safety. And that video really did not do well. It was not a popular video at all. Probably because it's like not a cool topic, right? It's like not like visual

17:56 and trendy and cool because it's like safety. It's like boring sounding. But regardless, this is a really really important topic cuz as AI progresses and all these things become available, the safety part of stuff that also becomes really important as well because with great power comes great responsibility.

18:12 So please, I am imploring people who are watching this video if you are interested in AI safety. First, I guess like check out that video and if it is something that you're interested in, please do dive deeper into this trend because I promise you this is going to become increasingly more important in

18:27 2026. You can quote me on that one. All right, thank you so much for watching until the end of this video. I hope you feel inspired. I hope you feel excited to learn these new skills now. And let me know in the comments what AI skills are you particularly interested in learning right now. And I will see you

18:41 guys in the next video or live stream.