Transcription
In 2026, these skills will become as basic as working with Google Sheets, [music] searching the internet, or making requests to ChatGPT. I'm talking about creating personal agent systems for automating [music] your routine. And it doesn't matter if you're a programmer, marketer, analyst, manager, producer, content creator, or anything else. If [music] you don't master this knowledge, you will fall behind in whatever field you are in. This video is my personal [music] plan for what I will continue to study in '26 to achieve more, do work of higher quality, and implement my ideas faster with the help of AI. Robots, hello. This is Prodsovet. My name is Uncle D. Happy New Year to you. And let's start it productively. Like, subscribe, bell, let's go. God, this is too bright. With your permission, I'll move the light a little. I don't intend to drag this out. Here are the five skills that, in my opinion, are super important in '26. And if something seems banal and obvious to you, you can skip ahead using the timestamps. But generally, I intend to go from the most complex to the more primitive. And I hope it will be clear why as we go. Creating an agent layer. This is the final skill to aim for, but it's also the skill of creating an agent layer around your life. I read various blogs of developers, startups, products, and scroll through authors on X. And in the last months of the past year, I've been seeing a lot more excitement towards how agent systems work. Some people launch five to six sessions and agents in parallel, working on several projects and tasks at once. Boris Cherny is the lead developer of Kladkod, I hope I pronounced his surname correctly. In a month, he made 259 pull requests and didn't write a single line of code himself. Kladkod did everything. Super experienced programmers and people in the machine learning industry lament that they feel they are falling behind in this race, while no-coders, on the contrary, rejoice, creating three versions of a project in a few hours instead of a week. With agent systems, you can work incredibly fast. You become not an employee, but a manager who orchestrates the work of your agents. And although I provided coding examples above, this is truly for various professions. My personal goal for '26 is to become such a manager. I will create specialized agents for deep research and reports. End to end. Creating prototypes for my ideas, transcripts of meetings with my colleagues, creating plans, and even performing various actions, like moving a Jira card or even going online to buy something, to order what we discussed. Content automation, of course, and so on, and so on. The idea is to wrap all my work in an agent layer and ensure that the executor is not me. And just imagine how cool that is. I am the manager, the idea generator, the entity that accumulates creative potential and explores new information. And the agent is the executor. The thing that performs all the routine and dirty work. And this is the first skill, but it won't come without several primitives. Firstly, the stack, of course, where to create these agent systems? Very briefly. I've chosen terminal agents for myself, and you can choose to use some agent frameworks like LangGraph, LangChain, and do everything in code, or no-code yourself a UI. I like the simplicity and efficiency of terminal agents. I've already made more than one video about them, two in fact, so you can go and watch. Prompt engineering is about what the user should write. Context engineering is about what the model should see. Here's a status I'm showing you from an experienced prompt engineer developer on X. Context engineering is the next step on our ladder, down to the fundamentals. And if, simply put, in simple terms, it's managing the neural network's context. This includes system and user prompts, information about available models, tools, responses it receives from these tools, the context of LLM servers, intermediate files that the model creates to complete a task, and possibly passes to some other model in this multi-agent system. Your task as a context engineer is not just to write a prompt, but to understand how your agent consumes context, sends requests, what responses it sees, how aware it is of the tools available to it. I liked the analogy that an Anthropic researcher provided. Tools for LLMs should not be like APIs, but like a UI for the user. That is, each tool should have a comprehensive description so that the model understands how to work with it. If you see a website like this, you'll agree it's very stupid if you first have to click a button to understand what that button was. It would be much more convenient if the buttons were labeled. Similarly, an agent should know what tools it has and what they do. There's a lot of different content, research, articles about context engineering. I've left several useful links to primary sources in my personal Telegram. Go and read if you want to delve deeper. Here's what Anthropic says about the capabilities of coding agents. And I think this can be applied to us humans as well. I believe that basic IT literacy will give you a strong boost in your personal work and will help you, among other things, in working with the previous step, context engineering. Well, simply knowing what a terminal is, what commands are there, what programming languages exist, how they differ from each other, what a development environment is, version control, how to run a project on Python or Node.js. Many opportunities open up when you stop being afraid of code, and I'm telling you this from my own experience. I have no IT background, but I'm learning this independently or with neural networks, and I'm diving deeper and deeper, and life and work with these silicon devices become more and more interesting. You'll be able to do more complex automations, write various scripts, create more stable, robust systems, understand how the same agents work at a deeper level. And you won't need a developer, for example, to test some idea of yours, to create an MVP for a future startup. I think that's great. And here we've reached the atomic unit of any agent system. I still believe that prompting is important, and being able to write requests to LLMs, choosing the right approaches for different models, is the key to success in creating a reliable agent system. I think it's right to start any automation and system with a prompt. In Kladkod, for example, this is implemented through slash commands. Don't overcomplicate, don't try to immediately create sub-agents or skills or LLM servers with AGI. Start with a prompt. You can take, create a slash command and then use it, substitute it at the right moment, as I do here, so that the model critiques my script and helps me improve it. The prompt itself needs to be added to the root folder of slash commands and the name of your slash command in Markdown. And keep in mind that different models actually require slightly different prompting. For example, requires more structured, detailed prompts, and you can use them, this prompt optimizer, which will help you write these prompts. Claude reacts very well to CSS and you can often write critical and describe something to it so that it definitely doesn't forget it. I've noticed this myself and in Anthropic's system prompts. And, in general, many users also pay attention to this. And finally, all of the above makes little sense if it's not applied. My area of development for this year is product management, including not just creating agent systems, but creating them for something, for some personal productivity, automating tasks in my routine, creating a side project, to either learn or just become a solo developer, all that. Essentially, this skill, I would even call it more of a stage and some preparatory work, it comes before everything we've discussed above. It's about why I should create a particular system, who the end user is. And how to do it as simply as possible and not over-engineer it. Here, I believe it's important to make quick iterations, gather feedback from your users, clients, change your approach if necessary, and generally monitor the market, look for trends and ways to create added value. If everything I've said above resonates with you, seems interesting, a fundamental thing to invest your time in, as it seems to me, then pay attention to the following announcement about our program, which we are launching very soon, and after that, we'll summarize. Essentially, the main skill is creating an agent layer. And all the skills I've listed after this skill are preparatory, leading to this wonderful, marvelous ability. I think it's important to learn everything through practice, of course, because you can find a ton of information, but finding people who will push you and give you feedback, who have already gone through a certain path, that's, in my opinion, cool. On January 12th, a three-week sprint called "AI Dev - Side Project with AI" starts, and in 21 days, we will be creating our first product, or not necessarily, for some it won't be the first, but we will go from idea to the first results and first users. We're not just learning, we're doing. During the intensive, we will have five to six live workshops where we will go through the following path together from idea generation to creating an MVP product. There will be weekly project and idea review sessions. You'll be able to start implementing your ideas with AI under the guidance of mentors, myself, and my colleagues, individually or in a team. And the main value of the program is, of course, the sprint and networking, but we also provide you with a theoretical foundation. Access to a library. There are 200, even 300, probably plus lessons on no-coding, creating AI agents, LLMs, graphics, and video, AI avatars. We have a lot of content there on working with AI tools. So, at the very beginning of '26, I suggest you go straight to results, and not just write down plans and goals, which, as we know, are not always realized. Therefore, the start is January 12th, we have only 20 20 spots at a discounted price. 10 have already been bought for sure, as far as I remember. The registration link is below this video, and more detailed information is also there, so welcome. Let's summarize the results of my semi-philosophical speech from this video. The first skill and the main skill, which you need to reach by mastering the other skills, is the agent layer. We wrap all our work in agents and explain to them how to perform it for us. Context engineering without understanding how context works, what tools models can use, whether they have access, how the context management system is built, it will be impossible to create agent systems. To master context engineering, it's good to understand basic coding. If a model calls an API and you have no idea what it is, what endpoints are, then you won't get very far. Prompting is the atomic essence of all agent systems. Start with prompts before moving towards complexity. And a product approach. Create useful systems that will either solve specific user or your tasks, help you be more effective, again, to create products and achieve your goals. Start simple, the rest will follow. Don't forget to subscribe to the channel, to all our resources, Telegram channels. We have different ones. Our general personal Telegram, Igor's Telegram. All links are in the description, to our bot with useful knowledge too. And see you in future episodes. Like, subscribe, bell. Bye. [music] >> [music]