[00:07] What if I told you that one AI assistant did the work of 700 people in just one month? This really happened actually. In 2024, the payment company Klarna launched an AI assistant for customer support. In its first month, it handled [00:20] support. In its first month, it handled 2.3 million chats. That's 2/3 of their support chats. It did the work of 700 people and it solved the problem under 2 people and it solved the problem under 2 minutes. Sounds like AI won, right? But [00:34] here's a twist. By 2025, Klarna started hiring humans again because the quality of service actually dropped. So, here's a lesson. Building an AI agent is easy, but building one that is reliable, safe, and ready for real user is the real [00:49] skill. And that's exactly what companies are hiring for. They are hiring fast. Gartner predicts that 40% of enterprise app will have AI agents built in by the end of 2026. In India, the demand for agentic AI talent rose to 260% [01:07] in 2026. Skills in GenAI and AI agents can add around 25 to 40% and there's a big skill gap, which means fewer people to compete with. Lots of people can build a demo agent, but very few can [01:21] ship one that works in the production. And that's your chance. So, in this video, I will show you the complete roadmap of how to become agentic AI engineer. Now, here's the agenda of our today's session. First, we are going to [01:33] cover what is agentic AI and how does an AI agent works. step-by-step roadmap. First, we'll discuss programming basics, then we will move towards LLM [01:46] fundamentals, then we are going to deep dive into prompting and drag. Then, we will look into agents and tool use. After that, we are going to deep dive into memory and multi-agent systems. Then, we are going to discuss about [01:58] testing and security. After that, I'm going to tell you production and DevOps practices. Then I will also tell you what are the popular agent frameworks to actually learn. Then we are going to conclude our session with what employers [02:12] look for and what kind of portfolio projects you can build. Now before you move ahead, just a quick info guys. Simply learn has got program built around exact journey we just walked through. [02:24] The post graduate program in artificial intelligence and machine learning in collaboration with IIT BHU Varanasi. This program has Microsoft as its partner. It runs for 8 months in live online classes about 6 to 9 hours a [02:39] week. So you can learn alongside a full-time job. You can start with the foundations. You'll also get hands-on with 30 plus tools and build 18 plus So guys, hurry up now and join the course. The course link is mentioned in [02:54] the description box. Now here's a short quiz to test your knowledge. What makes an AI agent different from a chatbot? And your options are it answers questions faster, [03:06] finish a task, or it only works with a text, Please mention your answers in the comment section below. >> So let's get started. So first, let us discuss why agentic AI skills actually [03:23] So guys, [snorts] in the starting of the video, we discussed a case study about a video, we discussed a case study about a company and they are actually moving towards AI that can actually act. Gartner predicts that 40% of the [03:36] enterprise apps will have AI agents built in by the end of 2026. And by 2028, a third of enterprise software will use agentic AI. And 15% of [03:48] daily work decisions will be made by AI agents. That is kind of a huge shift and someone should build these agents. So guys, that's why the skill requirement is there in the industry. Now let us [04:03] discuss about this point why someone should aim for this role. First, the demand. The demand for agentic AI talent is very high. AI engineer is the fastest growing job title in the United States. Third, it [04:17] pays actually really well. Gen AI and AI agent skills can actually add 25 to 40% hike in your salary. And also fourth, hike in your salary. And also fourth, there is a skill gap of 38 to 60% in AI [04:31] cloud and security talent. So, you have a great chance to make a career in this domain. But first, let us try to understand what is agentic AI. In simple words, it's an AI that can plan steps, use tools, and [04:45] finish a task with little help. A chatbot answers your question you will ask, for example, like, "Will it rain today?" And it tells you. So, an AI agent gets the task done. You say, "Book a cab if it rains." And it is [04:59] According to that, it is going to book a cab. Every agent follows the same loop. It starts with a goal, makes plan, takes action, and checks the result. Then, it [05:11] repeats until the job is done. So, this is exactly what agentic AI is. Now, let us try to understand what an agentic AI engineer actually do. So, basically, they design agents, which means planning [05:26] what the agent does step by step, then connect tools so that the agents can use apps, APIs, and data. They test and secure the agent, so it's kind of safe it it does the right thing in the production scenario. [05:41] And they ship it finally to the production, which means deploying it on the cloud and keeping it running. Now, let us take a roadmap look at a glance. First, we are going to understand programming basics, then move towards LM [05:55] fundamentals. After that, we are going to discuss about prompting and drag. Then, we will deep dive into agents and tool use. After that, we will discuss about memory and multi-agent. Then, we will discuss about testing and security, [06:08] and finally production and DevOps. So, the first stage is programming basics. So, you can start here with Python because this is the primary language, and it has a very easier syntax as compared to other object-oriented [06:23] programming languages like C++ or Java. So, you can start with that, explore the basics of it like you can start with variables, data types, move into conditionals, then understand bit about exception handling, file handling, and [06:37] the list goes on and on. Um so, what you can do is you can go to the official documentation of Python, and if you are someone who don't know Python, then definitely explore our YouTube channel. We have a Python full course there, and [06:51] learning from that. After that, explore APIs and JSON, then explore version control system like Git and GitHub. After that, I'll also recommend you learning Linux command. So, you can try Linux command in the [07:07] terminal. So, there are a lot of sandboxes available online where you can So, this is one of the basic thing. After completing this, what you can do is you can build a mini project to test your [07:21] understanding. So, you can take weather app basically which takes app basically which takes the data from an API and portrays it on your, let's say, application whatever you are building. So, this is very [07:35] simple application, and you can use HTML as your front end, and back end you can use simple Python. Next [snorts] step, I would say LLM fundamentals is very, very much important because LLM is the brain of an [07:50] agent. So, whatever you are going to say, it is going to go through LLM. So, first understand tokens, what is context window, then what is temperature like [08:02] how creative are, and what exactly is hallucination. So, these are some of the hallucination. So, these are some of the basics of LLM and from that we can go towards prompting and rank because a prompt design is very very much [08:16] important. How to write clear prompt like for example giving it a role showing an example or you know asking for a format or give you can also give a format to your LLM that can also be fine. After that you have to build [08:29] application that utilizes rack. So, rack is very very important concept. So, after learning all this thing we have a video on rack also on our YouTube channel. So, you can explore that. So, for example if a user asks something [08:46] let's say um for a refund policy. So, the agent is going to look into the company policy data or the policy book and from that knowledge source it is going to give the what what will be the next step to the [09:03] user like what he can do to actually get the refund. So, it's kind of a very personalized you know thing which rack helps us to achieve. for this you can just start with a chatbot that answers [09:17] After that I would suggest to you exploring agents and tool use. So, basically agents think they pick tool and uses it and then it gives you [09:29] the required result until the task is completed. So, it's a kind of iterative And what tools it could use it could use web search calendar email database code runner. So, if you are using chat GPT or [09:44] cloud then you would see how the agent is actually using certain tools to give you the given data or the result whatever you are asking to the agent. So, that is very very much important and you have to understand how to build [09:59] agents. So, that is very much into our context. After that you have to explore memory and multi-agent system. Explore memory because memory is very very much important. So, agent uses different kinds of memory like [10:15] short-term memory, long-term memory, episodic memory, procedural memory. And also explore multi-agent architecture, which is very much important for building multi-agent system. [10:27] So, these are very much important concept. After that, you have to explore concept. After that, you have to explore testing and security aspect of building an agent. So, run evils on real test cases, you know, ask a human before any [10:41] risk is involved. So, human in loop is very much important. And give each tool the access it needs, but not for every tool, only tool which is required for that particular task. After that, you have to learn DevOps, which is very very [10:57] much important because you have to ship the AI application. So, version control system is is your first thing, then understand the CI/CD process. Then, you have to also explore Docker, and you should be very proficient with [11:11] and you should be very proficient with the cloud concepts. So, you can use AWS or Azure as your starting point. And after that, you have to look for monitoring aspect. So, for monitoring, you can use Azure Monitor if you are [11:24] using Azure as your platform, or you can use Prometheus or Grafana also, that also works fine. Now, these are certain uh popular [clears throat] agent frameworks like LangGraph, which is basically made [11:37] by LangChain. Then, there is CrewAI, um Microsoft Agent Framework by Microsoft, OpenAI SDK, and Google EDK. So, these are very much important. And also, you have to understand what is MCP, how you can use MCP as a common [11:53] MCP, how you can use MCP as a common plug to connect agent to tools. And it is very very much important concept. So, these are the thing. Now, let us look at what employers are exactly looking for. Employers are actually looking for [12:07] Employers are actually looking for engineers who can ship AI products very And they will definitely look for your portfolio, so you should have your GitHub profile with pull requests. And [12:20] you should be able to run Docker on cloud. evil. The CICD thing that we discussed how you are going to manage keys, basically which are very much important. [12:34] You should have diagram or architecture of your project. So update your read.md file. And also the cost optimization parts like how you can trace or how you can study the logs and look into the cost [12:49] aspect also, which is very very important dimension for building AI application. Now just a quick check guys. This was your complete roadmap for becoming agentic AI engineer. Now [13:04] becoming agentic AI engineer. Now the question is your agent API key was pushed to a public GitHub repo. Then which skill would have stopped this? And your options are better prompting, secret management with scan [13:16] and CICD or multi-agent design. Please let us know what is the answer in the comment section below. Now these are the portfolio projects which you can build. So you can take as a beginner level company docs [13:29] and QA bot. So you will have to use fast API as a endpoint and you could check all the test in GitHub action whether it is passing the tests or not. Then meeting scheduler agent kind of a [13:42] mid-level project. Basically runs in Docker. You are managing all your keys. So in a secret manager that's also you can show. An advanced a research agent team. So this is based on multi-agent architecture. You have to deploy it on [13:58] any cloud platform. We have to give the CICD traces, logs and LangFuse is one of the thing which you'll be using. So these are some of the basics project, but it will [14:10] definitely give you a clear idea or picture of how to use the particular tools to be very proficient and you can also speak a very in front of your employer during the interview. So guys, this was all for [14:26] today's video. I hope so you would have enjoyed our today's video on agent QA engineer roadmap. Thank you guys for watching this video and if you like these kind of videos then do not forget to hit the subscribe button and click [14:38] to hit the subscribe button and click the notification bell icon.