How do you build an AI agent that doesn't break?

So, how do we design an AI system that does not break? And what do I mean by breaking over here is that your AI system that you've paid an freelancer, let's say, or a person from Upwork or a certain AI automation agency to add within your business is now answering the wrong answers to certain people, to certain questions, let's say. It's quoting the wrong prices. It's giving them the wrong offers. It's answering clumsy English for Arabic inquiries. And overall, it's making your brand look very cheap when you've paid a lot in the premium positioning within the UAE market that you have for your small business. And you only find out it has broken, not when it stops working. It can work fine for the next six to 12 months, but you only find out when leads go down, cash flow goes down, and everything's like just a mess right now, and you don't know how to fix it. This is exactly what we're going to talk about today. So we're going to go into three main steps. Point number one is we're going to begin with the end in mind. Meaning that what are we designing this AI system for? Why do we have this AI system within our business in the first place? I'm going to give you two numbers over there. Very important numbers if you're going to run AI systems within your small businesses. Point number two is you're going to build it right. Even if you're not the person building it, you're going to be adding AI more and more within your business because we have this agentic AI wave coming in. A lot of your competitors, a lot of small medium businesses within the UAE will be adding AI to some degree. And you must be aware of all these first principles or fundamentals to how to in order to audit or understand whoever you pay or bring on board or even if you build it yourself to understand the whole picture and what's happening and how to do it right. And point number three is you need to prove it survives the extreme conditions which you have extreme conditions that may happen within your business. I'm going to come into the methodology over here and all this is taken from my experience in building a space systems and satellite systems and launching them into space because once you launch something up there you can't go and fix it right that's how we approach AI systems but a less let's say rigorous form of that methodology and a combination of also how we've applied this to UAE businesses for the past two years going down on the channel you can see proof we've been talking about that for the past two years. So now let's give let me give you where does this breaking come into the picture. So the whole game here is building an AI system that does not break. And many people what I see within the UAE and small business right now, they're just paying people and adding this AI automation kind of thing within the business. They're just bolting it on top of that. We started with this wave where everyone thought AI is just the chatbot or the chat widget that it's that's at the bottom right corner over here that says hello how can I help you or do you have any questions and so on and now it's transforming more into let's attach this thing to a proven or unproven system and let's see how and what it does the problem with that and what everyone is telling me right now and the biggest fear which I'm going to share in a bit over here is that one if they touch it everything breaks down and two if they add any new integration, let's say they shifted their business model. Let's say they introduce a new marketing channel to their business, they add that and everything also breaks. So it's kind of this fragile kind of piece of flower. Let's say they have built the AI system as a flower that dies as soon as they touch it. So this is something the agencies or whoever built it out for you understand. And during my days when I used to sell websites and SEOs before AI was cool and AI did all these cool things for small businesses and build this website many people were telling me that okay you're going to build this website for me and this is small business and you business owners you're going to build it for me yes but I don't know how to maintain it I don't know how to operate it can you do that part for me and they were much fearful let's say or scared to being hooked by that meaning that they need to pay me for a really long period of time if they want a website. And right now, the same thing is happening with UA business owners when they talk to automation agencies and everyone else from Upwork and Fiverr and so on. How do I operate this AI system once it's done because it's broken in the past? I'm going to show you how you can make sure that doesn't happen or you don't come to that situation and I don't want to be locked and that's what they're saying to a vendor or a person where I need to pay this person a retainer every single month to maintain it and fix the bugs and so on. So let me give you a secret over here. The reason novice beginners low quality agencies put that retainer is one other than their their let's say vicious plan of getting more money out of you but they put it over there because they know they've not built the right solution and they say that okay my time is worth my my my time is worth money and your your system I know it will break after a couple of days pay me I'm gonna keep on fixing it when it breaks because I did not design it which is not they're not gonna say that but I did not design it properly and this is how it looks like. That's part number one where the honest way of having let's say retainers and how I think why genuinely people need retainers is that because there are going to be updates because it's not all just a certain software because you've attached the AI system to external providers which are we're not we we don't have control on them they're very uncontrollable they may update a certain let's say parameter they have and the system gets disrupted as So this is the evil way what I'm trying to say this is the good way fine retainers are fine but it's misused within this world and I understand why small business owners right now or medium alsomemes let's call them are just fearful about this I don't want to say I system I don't want to get locked into this person so what you do and the best way to do and do it over here is to build an AI system that does not break on the basic level and keep that retainer the honest one which updates only the small things or even provides updates to the software itself. Meaning that okay, they delivered to you, they gave you that AI system right now and now maybe they want to update it. They want to add more functionality to it. It's not that okay, I just you you just paid me for this price. I'm not going to do anything else. I'm just going to maintain it. But more let me add more features. Let me see what's your concerns. Let me see what's happening or what do you need as a business owner. Let me add it to you. So this is just the two ways you can test that as a small business owner. You can say okay what is the retainer for? Go into details before buying if you're fearful about that fact. And for agencies try to stay on this side please because you know we don't want to to build something that breaks. Don't don't create a problem to and give them a solution for the problem you created over here. So that's point number one over here. And let me give you a small story of how this started because I'm not just telling you because this is just like there over there because this was me at some point and it was me during my senior design project. So in my senior design project during my bachelor's we were under pressure to deliver because of course I wanted to graduate right and we were given this task you can see over here the image we've given this task where there's this drone that we have and then there's this Linux operating system that's completely let's say at that time for me it was an alien kind of operating system and it they told me you need to learn C++ and you need to learn Python and all these programming language to control this drone over here with an Oculus Rift headset which I had to order from eBay and it's a used one. It wasn't a new one as well to connect them and make it work together. So for me to graduate what I did was I was patching things up. AI was not there. I did not understand what am I doing over here on this operating system and I just patched things up. Took me a lot of time like you know how you put things with duct tape. In the UAE we call the industrial regions let's say in Abu Dhabi Mustafa it's kind of like Mustafa work or we call like anyways you understand lowquality garage places it's kind of like duct taping things together and so on. I can give you a lot of stories about that as well but what happened is that we I spent a lot of time on this particular project duct tape duct taped everything and it eventually worked. But what what I understood within me at that time is that this was not built the right way. Anyone if they just change one small parameter or introduce one small change within the whole system everything breaks down the drone will hit the wall. So me going into satellite systems after this after the drone systems I had a kind of imposttor syndrome saying that okay people believe that I've built this for a drone the control and the system and I've I've connected them all together and people now believe that I can do it for satellites but internally I understood that no I duct taped it and I didn't understand the fundamentals so that took me like a year and or two with industrial space deep tech experience to get this process or framework in order to how to make a certain thing work and from that day on I've carried that throughout every work that I do and now we've achieved to AI systems and that's exactly how we approach AI systems and what I'm sharing with you and the take I'm showing you and this by the way is in NASA in the US when we launched one of the satellites so beginning with the end we have two numbers over here when you're going to build and operate AI system engineer ing starts from the end. What is the outcome we are solving for? In our case, in a business case, what do we want out of this AI system? Is it 20 leads per month? Is it five sales per month or five clients per month? What is it we want? And how long has this thing or we need this thing to run in order for this to be successful? For satellite missions, for example, there's a hard disposal deadline for satellites. for example, it's 5 years if you send it up to space and then you can push it towards a an external orbit just so you don't make a mess in in space itself. There are some laws that say 25 years and so on. This it's there's a lot of details about that but the point I'm trying to make here is what is that defined life or design life for your AI system. And for this case and what the number I'm giving you if you're building an AI system for a small business is 6 to 12 months. It should operate for 6 to 12 months with full effectiveness untouched and give you the results that you want. And why do I say 6 to 12 months? Because it takes for any system that you implement within your business today. Not one day, not seven days, not 14 days, which is two weeks, but six weeks to see signs of potential results. You can't just test something in one day say this does not work. It takes minimum of six weeks and to reach its full potential it takes six months and that's why I say six months untouched so we can actually assess whether this system is working or not and point number two so point number one is all about time and point number two it's what must it produce for this to be worth the hassle in order to implementing AI within or agentic AI within my small business over here will it make me more money will it give me more time will it convert that time into money for me. So what is that number that I'm satisfied of? And a nice way to imagine this is that what is that number if AI got you in terms of more revenue per month or per day per week whatever numbers in your mind that would say yeah this is worth it actually if we can do this within the next six months let's say we're going to achieve an extra of 10,000 durhams per month or whatever numbers in your mind for example the way you're going to calculate this is let's say I want 20 I want 50 leads per month I want like it's going to be worth it to me this AI system if I get 50 leads per month or 100 leads per month and each lead to me is of let's say 200 Durham's value and the cost to run this AI system is let's say 1,000 over here meaning that my AI number is $19,000 if dirhams if it gives me 19,000 dirhams then this is worth it for me to use it and the AI system and so on. So this is what I mean by an end in mind. How long should it run and what should it produce? That is the outcome we're going for within our business. For example, if it's a go to market, more leads. If it's sales, it's closed deals. If it's operations, hours saved. And convert that hour saves into a monetary value. So the payback is your AI number over here. Two things I want to keep you keep in mind. The time aspect and the AI number over here. Amazing. That's one part. This is the fundamental to build it right and assess any piece of AI system that's going to come to you. One is a piece I want you to understand the three things over here. A software or a piece of software is a computer doing a human's job but by its own it's completely dumb. And that's the base we're going to use. So what is a software? We all know software. We call them in the UAE apps. Like anything we call it an app like I want an app. I want to build an app for myself. It used to be called portals by the way but I want an app and so on. And automation is that software running itself but with fixed rule meaning that I have a workflow. The way I onboard people is they pay then they get on a call then they fill out a form then they get the contract whatever I'm just making it up but whatever the the flow is. So that's an automation. You can make that workflow an automation because it's a set fixed set of rules. Let's say this is the sequence of steps we go to achieve this certain outcome over here which is to onboard someone this is an automation but AI when you add it to the mix this adds chaos by the way and I'm going to come to that part but it's that intelligence layer on top meaning that when I let's go back to the onboarding example first I get I I get the payment the payment goes to a call let's say sends a calendar link or a calendar link an invite then I get on that call that call gives me a transcript then I perform that or get them the report that they want. AI where it fits in is that it takes that transcripts understands what was said on the call and maybe that report for every individual is different. Maybe the form the onboarding form for every individual is different. So AI just gives that you that intelligence where it thinks what was spoken within this call and how does this person or business owner do the forms let me adapt that form for this person to get maximum value and we extract the right knowledge so we can deliver the service properly to them. So this is just an onboarding example. So I hope this makes sense in terms of difference between software automation and AI. So, so far you know the time, you know the your AI number, you know the difference between software automation and AI. But AI by itself and I want you to understand this because a very big complaint I hear or something I hear by a lot of business owners, not just business owners but in like in general a lot of people over here but AI hallucinates but AI is not predictive but AI is just dumb. It's just every model that's coming out like just now they released Fable 5 I think. Yeah, Fable 5 by Claude and they're saying what is Fable 5? It's it's just dumb. Everyone is losing their mind and so on. AI by itself and you need to understand the nature of AI by itself is is very unpredictable meaning that it's just based on probabilities that it predicts the next step based on the current step that it has. meaning that it predicts the next word based on the current word. So it goes in small steps to give you an answer to whatever question you have and this is the formula over here. You can go look into base theory and all those kind of things. But just a implemented example for you let's say you tell AI the cat sat on the he just wants now now AI comes here and wants to guess what's the next word. It's going to guess okay mat rug floor chair car whatever keywords that may come over here and it says what is most likely in this case it said okay this most likely I have five 58% 58% confidence that it's Matt so the cat sat on the mat then full stop then it goes another sentence and just keeps on guessing the next word I hope you get what this means right like it's just guessing the next word and it's based on probability abilities. So if you just give it again and say the cat sat on the and give it a blank, it's going may give you rug, it may give you floor, it may give you a chair, it may give you a car, it may give you anything else. So it's just based on probabilities that point to number one and the way you control this for your business because this is not good for business. You can't just let AI guess for your business, for your clients and so on because money is involved and you don't want to lose money out of this. is that you contain the AI, the hallucination as they call it or whatever. Hallucination will always be there by the way accept it because that's inherently how AI is. And I've seen a lot of comments by the way with my previous videos, people saying that but AI hallucinates. How would you deal about that? But how will you have AI employees if they hallucinate and so on? AI yes will hallucinate as do humans hallucinate on their job as well and they do their job wrong. So it's not some magical thing that it's unlike human robotic let's go humans also deal with probability actually probability is a very nice field if you want to get into but the way you contain the AI hallucination or this unpredictability is you have a set of workflow let's say a workflow a sequence of steps you put it within the sequence of step this is what I've seen the best you don't let AI say okay do this full steps by itself you don't tell AI AI to do that. You don't let AI or build an AI system that does a complete flow. You build an AI agentic AI system for that particular step that's contained within a full sequence of steps. You get what I mean, right? You don't say go onboard this client to AI. You say create this report for the onboarding process which is step number three within the onboarding process. And the report looks like this. Now AI is less hallucinating and understands where where you want to go. We're going to get into details of this by the way in future videos. But just understand this concept over here. So this is the basic principles when you're talking to anyone and or or building AI yourself. You understand the sequence of steps you're building AI for. You're containing it within a certain let's say substep. You're understanding AI is just adding that intelligence layer and also you're understanding that this it must run for six months and produce me 20,000 durhams let's say by on a monthly basis after running it for 6 months and showing early signs at 6 weeks let's say so this is like how you assess or the frame you go for assessing every AI system for a business or small business purpose and then for the most important part where everybody skips this part is that it must be tested for extreme conditions before launching it or let's say as a beta launch as well. You can launch it but to a small number of people but it should be tested in extreme conditions. The same way a satellite was tested because once you send the satellite to outer space as I said you can't just go up there and fix it right and the same goes for your AI system. If you apply that rigor to it it's it's going to perform. It's going to give you the money that you want. And let me just give give you the or let you imagine how that was done for satellites. So for a piece of let's say metal or a satellite or an object to be spent sent sent into outer space, it's going to go on a rocket and it's going to go through that rocket to outer space. It's going to come out of that rocket and start rotating around earth. Let's say that satellite is around earth, right? So what are the conditions involved over here? Once it goes in the rocket, it shakes. So once it shakes, pieces may come off, right? So that's the vibration test you do to make sure the pieces don't come off as it goes to outer space and reaches space safely. So you test it on extreme conditions based on the rocket specifications that you have to make sure it's going to reach space safely. You don't want a pieces. You don't want a broken satellite in space. It's just waste or garbage. Right? So that's point number one. Point number two, when the satellite is in space operating. So I want you to think in these two frames, which is initiation, then operation. That's the test you're going to do for your AI systems, by the way. So initiation is sending that satellites, which is the vibration. Now testing for the operation, the satellite will either be in front of Earth or behind Earth. In front of Earth, it will be exposed to the sun. Behind Earth, it's going to be exposed to nothing, which is let's say complete vacuum dark. Meaning that in front of the sun out of the ozone layer let's say it's approximately let's say 120° behind the sun is minus 150° C which is extreme cult cold and extreme hot environments and your AI system most probably does not face such extreme conditions but people can be extreme as well. People can be unpredictable as well. They fluctuate as well. This all the like if you think about all of this is game of probabilities, right? So they can be unpredictable as well. What if someone sends you a voice note in Arabic with some English words and it just messes your or confuses your AI AI up because you've set this automation and your AI says API error. It just responds with API error to their client or prospect. What if you get 100 in inquiries because Eid is coming or Ramadan is coming and you have the certain store that sells during seasons and a season is when the store store sells and you get 100 inquiries in one hour because they just announced a tomorrow. Let's say you're a barber or a clothes you you tailor clothes. So what would happen to your AI system? So these are edge cases that you will test your AI system for at its complete extreme scenarios while you initiate it and while you operate it as well which is the example over here messy inputs hostile customer what if this customer so there was this case I'm not going to mention who but a very very popular brand that that built in that wanted to get on this agentic AI wave and built in customer call agent so people would call and say I want help with this and that and this so there's one person that called and started insulting the AI the AI and this is a very by the way the I want to put two underlines under under a very important entity the person said you're stupid AI the AI said no you are stupid so this is when they took the AI voice agent down so hostile customer you don't want that to happen you don't want you want to test it for volume spikes you wanted to test it for dependency change what if you're taking data from a certain in database. Let's say you're taking from ampify and this actor updated their terms. The output is different which is the output you rely on as your input and then your system completely breaks. So you test all of these out and make sure it it just passes and it's fail safe. So the big question you're going to be answering with all of this that I just explained is that would it run to its design life and hit the number that you want the potential your AI number with no one touching it let's say for 6 months that's the standard you're going to hold your AI systems for your small business to it has to meet everything that I just talked about the design life the AI number because if does not meet the AI number for example, it's for you not worth it. You're going to be say saying, "Why am I wasting my time for just 10,000 dirhams in 12 months? It doesn't make sense because you're going to put some time into this to build it." It's going to be an extra, let's say, cognitive load. And does it survive the extreme conditions? And the orbit test, which I just gave you, answers to all these three things. And it either passes untouched or fails. So the test is design life is set. The AI number is defined. You understand the difference between AI and automation and a software. The AI is contained within a substep and it it survives extreme environments. And the reason I keep on repeating this because it's super super important. It may seem logical but I'm trying to give you different perspectives to just make this idea stick as you build these AI system. And an example here is let's say in Business Bay or Marina there's this clinic and this clinic there's an inquiry that lands on WhatsApp and Instagram and it's in Arabic and English and they bought this AI system to book calls on their calendar. So how they would score it using this over here is that have they set the design life have they had the AI number is there any AI and automations and is the AI contain and does it survive the extreme environments and they have this checklist and say yes it's met this is open open let's test it let's do it and let's fix it and so on so essentially what we talked about we check if it breaks or not because most people are now complaining that their AI system is breaking we understand difference between software SAS and AI and automation. We make it contained. We set the design life. We set our AI number and we do the orbit test to make sure that AI works. And that way you're going to run or run an AI system and build an AI system that actually produce the results that you want for your small medium Jesus.