I took the first ever AI Certification 2H Exam..

0:00 Yesterday I took the first ever AI certification exam. In my previous life, I worked as an architect and I had to have license which meant that I knew how to make buildings that would not

0:12 collapse on people. Doctors have licenses, lawyers have licenses, but AI people can walk into the business and introduce nondeterministic AIS and call it a day with no proof that they know

0:28 what they are doing or they are doing it right. So on traffic, the leading AI lab introduced the first ever certification to show that you can make critical decisions when building AI systems. Now

0:43 I'm not talking here about the certification of a free course that you can take on Entrophic Academy. That certification is for completion. I'm talking about full 2hour 60 question

0:55 proactored exam when you are taking it either in a test center or you are taking it online and you are being recorded. In this video I want to talk more about what the certification does

1:07 and what it enables. I will make a completely separate video on all the prep that I did and in my community now we have two more people who are literally certified this week. So we are

1:18 going to park all those lessons and how we prepare and how we all pass for another video. The first real question is why should you even care? If you if you work with AI either in your business

1:30 or you are servicing clients, you most likely you're not going to work alone in isolation. You can, but you know like as a single person just to manage clients, you're going to it's it's not really

1:45 scalable. Therefore, you won't surround yourself with people who you can trust. The issue with AI industry is that we don't have really any standards. If I am talking about AI and how I build my

1:58 systems and you're talking about it too, how do we know that we both follow best practices or that our approach is actually the right approach or the way we judge systems or work on constraints

2:11 is the right way. Certification does this, right? Certifications allow us to speak the same language. And that is my primary reason why I wanted to take that exam. I did take it and drum rolls I

2:26 passed. I wanted to see if my approach how I build systems is aligned with number one AI lab where they deploy the systems into the businesses and they know the best practices they know the

2:40 constraints and the all the problems that come. To pass this exam you have to score 720 out of a,000. I scored 886. So close to the nice round number of 900 but I passed. The thing is that two

2:55 other members in my community also passed which means that when I talk with Sheila or Jeff we are speaking now the same language. Sheila and Jeff learned on the same materials practiced with us

3:08 been with us in a community for a long time and then we talk about systems. I know that they know what I know. So it means that if I bring them to the client and we talk about clients problems that

3:22 Sheila and Jeff will be able to properly evaluate and judge it. Now if you want to take certification for the sake to have a stamp on your LinkedIn, no shame for that. It is completely valid reason.

3:36 It can open new doors. It shows to external employers that you went through the hard process and you sat on an exam and you did the thing and you passed and you know what you are talking about. So

3:49 that in itself is good. I want to build a business and I want to build a community which is the top talent in AI which means that we need to speak the same language. And if my members are

4:01 certified claude architects or certified claude professional architects that is insane proof that what we are teaching how we are approaching system design it is the right direction. Now you might be

4:15 wondering okay so what is the exam and what it's testing? The exam is not memorization. you don't have to code and it is not something that you know there is right answer to the question and that

4:28 is the critical thing and I think that's what everybody who already took an exam really appreciate because it is not easy and the reason it's not easy is because you have to judge situation and there

4:42 are multiple right approaches but you have to really identify what is the constraint what is the limitation what is business even able to do what would be consequences of my actions and

4:55 critically when are we overengineering project when do should we use AI maybe we shouldn't use AI the way I would describe you are building a system for the business or the process it could be

5:08 multi- aent research system it could be extracting data and clarifying data and so on so you build a system and inside that system sits one little piece which is LLM

5:20 And that LLM is a good data processor. That's kind of what it is. And so you have an input that comes in. It processes and then there is an output. And because in the business

5:35 deterministic process, we are placing this one piece which is nondeterministic. We really need to know what are we giving up, what we are gaining. What do we have to build and

5:48 constrain this nondeterministic element in the whole system so that when input comes and it gets processed that the output is correct and validated? What type of input we should never give to

6:00 LLM? I would say for the 90% of the businesses, they do not need to introduce an agent. It's just the fancy hyped term that we use in AI space. Businesses need workflows that work and

6:17 they do not care if you call it an agent, if you call it multi- aents, whatever it is. Every agent by the design of it introduces nondeterministic workflow, which means that it's harder

6:31 to verify and audit. And if you introduce this and your business demands for it, right? For any agent in a business, this is what AI browser is not going to tell you. In a real business,

6:44 when you introduce a a agents or agentic workflows, you have to track every decision point and prepare it for audit that would come in a business at any given time. Okay? So in real life

6:58 scenario agent would be probably the last thing you would go for because it can inflate your cost. It is way bigger part of engineering. You have to trace every decision. You have to calibrate

7:11 the tools that that agent is using. You have to run evaluations for that agent and if you change your model the behavior changes completely. And this becomes really clear once you are

7:23 preparing for the certification and then you take that certification. The exam is testing your judgment. I would go as far as to say that you can prepare all you want and there is free mockups, free

7:37 exam guides, there is entropic preparation things and free courses. But if you have not built tools, if you have not set up MCP server, if you have not set up the system in a real business or

7:50 a real production with the real users, you have not run into those issues where you learned those critical judgment calls because through test and error and for the last three years, we've been

8:01 building systems and I've been involved with multi-agent systems and that was probably the main thing that helped me to understand because I would read an example and I would be like, "Oh, yeah.

8:12 I remember running into exactly this issue and I remember what were the consequences of my decisions." We take this whole system and we stamp the name of AI. It's actually just business

8:25 operations and processes. And AI is great at transforming data, extracting data, um, converting data, whatever you want. It's just you have an input, it massages the whatever it needs to

8:39 massage, and then you have an output. LLMs best used in operational processes which usually falls under the business. If you learn how to actually use it, where to place it, what are the benefits

8:52 and tradeoffs and what actually you are introducing into the business, where you put your name on the audit logs because believe me, business is going to call you when an auditor comes and is asking

9:05 how did this agent decide that refund should be issued or how did this agent decide that this person and not that person needs to get hired? That's going to be you who needs to answer those

9:17 questions. And the issue right now is that people build systems, sell it as oneoff to their businesses. Businesses run into bunch of issues using those systems because they do not understand

9:30 what is actually happening under the hood. They bought a solution. Right now they have to maintain it. And if you have not thought about every single step that you introduce LLM, what are the

9:43 consequences for the business? good and bad. What do they need to know? If you have not introduced and clearly communicated and thought about it, those will come out eventually. And this is

9:54 where projects fail and AI systems break and they are not good enough. I'm glad that I took the exam. I already knew a lot, but I learned a ton more and it also exposed my own gaps because I

10:10 didn't pass thousand out of a thousand, right? So there are small gaps where I'm like okay so I need to make sure that I drill those and build more and learn and experience real situations so I can fill

10:25 those right if I didn't take exam I would be like hey I'm confident I built really great systems and how would I know that maybe my approach is not exactly right in this exact situation

10:38 you wouldn't you're not going to learn that from watching a YouTube video or downloading random skills from the internet. If you are interested in certification, drop your comments below.

10:48 I will take a look at them and answer them in the next video. And I'm going to also cover how I prepared, what mockups I used, how I built my insanely hard mockup, which was probably at least five

11:01 times harder than an actual exam. um how we have a leaderboard in our community and we are preparing other members to take it with the visual examples and the projects that members build and now we

11:13 can say that they also pass exam. We already have two and two more members are taking it later this week. So it's working. If you're interesting in that then video comes out.