With AI from a security tool to his own company

A colleague of mine experienced how an internal security tool became a scalable product with the help of AI and eventually led to the founding of a company in Dubai. This happened about a year ago. The AI helped him with coding, pricing, legal questions, and later even with the individual steps on-site. What still sounds extraordinary today will likely become more common in the future.

Let’s call him Alex. The name is fictitious. The exact product idea and some technical details are deliberately kept in the background. His professional environment in IT security, Dubai, and the described process are real.

During our meeting in Dubai, he told me his story for the first time. Later, he let me read the full chat history. In it, I found mistakes, dead ends, impatient questions, and necessary corrections. It was precisely these parts that showed me that no AI provider was presenting a polished promotional version. Here, a person I personally know was reporting what actually happened.

The pitch

It all started long before Dubai. Alex works at a security company and, together with his colleagues, oversees many IT security systems. His company regularly receives inquiries from providers wanting to present a partnership or a new product.

Most of these do not fit the daily business and are quickly dealt with. One day, however, a startup contacted him with a topic that caught his attention. The company had developed a service intended to extend existing IT security systems with an additional analysis function. This was exactly the area in which he had been working for years.

The topic was relevant enough for a closer look. Therefore, he arranged a video call with the startup. A few days later, he, some of his colleagues, and the employees of the startup sat in front of their screens.

The startup guided them through its solution, explained the features, and showed which problem it aimed to solve. Professional slides, clear packages, and a finished website appeared on the screen. Several employees had been working on the product for around two years, and it was already being sold as a subscription. Everything looked as one would expect from a specialized security company.

The first thought on the call: perfect.

He had built a similar solution himself over the years. Not as an official product, but as an internal service for some of his employer’s larger clients. About ninety percent of it he had developed alone. Occasionally colleagues helped with individual tasks, but the technical concept, the structure, and the majority of the implementation came from him. He was the mind behind the service.

The solution complemented existing IT security systems and automated tasks that otherwise had to be done manually. The service was made available to select clients free of charge. It had no product name, no marketing, and no own sales organization.

If now a professional company could take on exactly this task, it would be a relief. He would no longer have to maintain the internal solution mostly by himself. Operation and further development would lie with a provider who focused on it every day. At least, that was the hope after the presentation.

Behind the slides

Before he deactivated the internal solution, he wanted to thoroughly test the new service together with some colleagues. Both systems therefore ran in parallel for several weeks.

At first, everyone assumed the startup would perform better. After all, several people had been working there for years on nothing else. The in-house solution, on the other hand, had gradually emerged in day-to-day business. Whenever something did not work reliably, he improved it. If a manual step was troublesome, he added automation. If an error appeared in a client environment, he adjusted the logic.

It was exactly this development history that made the difference in the end. Day by day, the results of both systems were compared. Whenever they differed, they opened the case and reviewed it. Was the result correct? Did it help an administrator in daily work? Why had the other system handled the same case differently?

The longer the test ran, the clearer the picture became. The startup’s solution looked professional, but technically it was not yet as mature as expected. There were gaps, questionable ratings, and cases where the product still didn’t convince in everyday use.

The internal system was better.

Not because it was built more beautifully or better documented, but because it had been adapted to real environments over the years. The service had proven itself in hundreds, later thousands, of customer systems. Every disruption, every incorrect result, and every unusual situation had made the processing a little better. This knowledge was embedded in many small decisions that were hardly noticeable from the outside.

In this comparison, the perspective shifted. Until then, the service had been a free addition for the company’s own customers. Now, on the other side of the video call, was a startup charging a recurring subscription fee for a less mature solution.

Suddenly a product

Until then, no one had planned much work time for the internal project. There was neither a fixed budget nor a roadmap. Nevertheless, the solution worked reliably and was used daily by the selected customers. Surprisingly little was missing for a sellable product.

In two or three concentrated workdays, he built a clear dashboard and simplified its operation. The internal service had become an offering that could be presented professionally and sold.

Until then, the company had only made the free service available to some of its larger customers. Many smaller customers didn’t even know it yet.

Now the company approached its smaller customers for the first time. In these conversations, he demonstrated what the service accomplished in everyday use and how easily it integrated into an existing environment. The response was immediate: the customers wanted to buy it. The free addition had led to the first paid subscriptions.

This answered the most important question. There was a real problem, the solution worked in everyday life, and companies were willing to pay for it. The proof was not a presentation or a waiting list. It consisted of paying customers.

The necessary separation

With the first sales, he knew that the solution could be sold directly to customers. It worked well within his own customer base, and the company probably could have sold more subscriptions this way. However, the company’s own customer base was not sufficient for larger growth.

In the security industry, many different companies sell and support IT security systems. They often offer similar products and target the same customers. The company he worked for was also one of these providers.

This naturally led to a conflict of interest. Another security provider would have had to offer a product from a company that, in the rest of the business, was its direct competitor. Even without a concrete interest in its customers, sensitive questions remained: Who had access to the customer relationship, and did the partner ultimately strengthen a competitor with each sale?

The solution was a clear separation. The service was to be outsourced to an independent company that did not sell IT security systems and did not manage customer projects. This way, other security providers could sell the product without simultaneously competing with the company behind it.

Before pursuing this plan further, he had to answer a crucial question. The solution had been developed while working for an existing company. Even though he had developed about ninety percent of it himself, he could not simply assume that he was allowed to turn it into his own company on the side.

Therefore, he spoke openly with his boss. He explained why an independent company was necessary for further sales and how he intended to avoid conflicts with his job. His boss gave him permission to use the largely self-developed solution and build the business alongside his existing role.

The agreement had one condition: his employer could continue to use the service for free. The new company was therefore not created in secret or against his employer’s interests, but with his boss’s explicit consent.

The direction was clear. Almost everything else was still uncertain.

What should the product be called? What would a sensible pricing model look like? How were orders, accesses, upgrades, and cancellations handled? Which legal form was suitable? In which country should the company be founded? What did the bank and payment service provider need? And how could all these decisions be documented so that technology, sales, and the foundation stayed aligned?

At this point, Alex opened a new chat and wrote approximately: “I want to turn this working prototype into a complete product and then into an independent company. Help me plan the entire process.”

The plan

He wrote to the AI everything that came to his mind. He talked about the existing system, the first customers, and the idea of an independent company. In the same text, he asked about the name, the domain, the website, the prices, the database, and the possible location of the company.

It was just as disorganized as a new project feels at the beginning.

The AI did not immediately respond with code or a finished company name. It first organized the mess. What did the product need to be able to do? What did the technical base require? How should sales work? And what needed to be arranged later for the company?

From that point on, they worked through the topics one by one. He no longer jumped with every new idea from the database to tax questions and then to the domain. Every decision ended up in a central project document, along with the reasoning. Because previous chats had been partially deleted, he could still understand even weeks later why a particular option had been discarded.

In technical planning, it was not about a long list of new functions. The existing solution could already reliably serve a few hundred customers. However, as a product of an independent company, it also needed to be able to serve tens of thousands of customers.

He therefore had to look at the codebase once again with a completely different perspective. Which parts would become bottlenecks with rapidly increasing usage? Which processes were only unproblematic because he knew them himself and could intervene manually if necessary? And which areas needed to be separated from each other so that one error did not immediately affect the entire service?

Ready for growth

During the day, he continued working in his regular job. In the evenings, he opened the codebase, showed the AI individual parts, and explained how the existing architecture worked. Then he asked what needed to change so that the service could serve significantly more customers.

This is how they went through the codebase step by step. The AI looked for bottlenecks, asked questions about dependencies, and suggested a more robust structure. He weighed each suggestion against his experience from real operations. He adopted some ideas, simplified others, and completely discarded some.

Gradually, he rewrote larger parts of the project. Internal processing, customer access, and service delivery were clearly separated from each other. Processes that had previously allowed for manual intervention had to become more reliable and easier to control. Errors should be detectable and fixable without requiring his personal intervention each time.

The proven functionality was to be preserved. What needed to change above all was the technical foundation behind it.

In less than a week, a stable version 1 was ready, which was significantly better prepared for increasing customer numbers. He deliberately did not try to cover every conceivable growth scenario at this stage. The service was to function reliably at launch and be scalable. Further optimizations could follow once the business actually required them.

From code to offering

For the first personal sales, the existing state had been sufficient. However, anyone who should be able to buy the service online independently needed more than a functioning technical core. A name, a public website, an ordering process, and the automatic connection between payment and customer access were missing.

Therefore, together with the AI, he went through the entire journey of a future customer. How does someone learn about the offering? Where can they understand what the service provides? How do they choose the right package? What happens after payment? And how do they then get access?

At each of these steps, he worked according to a similar pattern. First, he explained his goal to the AI and had it show different ways during brainstorming. Together, they compared the possibilities until he found an approach that suited his product and his situation.

Afterwards, he roughly asked a second question: ‘What have I forgotten?’ This very question was often particularly valuable. The AI did not limit itself to the current topic. It also pointed out things he had not yet thought of and showed which subsequent step depended on his current decision.

That way, it repeatedly drew his attention to the next dependency:

• Purchases also included invoices, failed payments, and cancellations.
• The website included legal information and data protection.
• The company had a business account.
• The bank account included business address and identity verification.

He still made the decisions himself. However, it not only showed him the next visible step but also the whole chain behind it.

First, they searched for a name and a suitable domain. The AI made many suggestions and checked possible confusions with him. He made the final choice himself because he associated a personal story with the name.

Afterwards, the website was created. It should not impress with technical details but clearly explain the service. A visitor had to recognize which problem was solved, how the solution fit into an existing IT security environment, and which offer suited their company. In addition, there were different packages with various services and support levels.

Next followed the purchase process. A customer should choose a package, pay online, and then automatically receive the appropriate access. The system had to recognize a successful payment, assign the booked package, and unlock the correct functions. It also needed clear procedures for later changes and problems.

He could initially develop all this in a test environment. He simulated purchases, generated accesses, and checked what happened in the case of a duplicate notification or a canceled process. This way, the technical part moved ever closer to a finished product.

At the same time, legal and administrative questions arose. The website had to make it clear which company was offering the service. It required terms of use, information on data protection, correct billing details, and clear rules for subscriptions. The AI helped with initial drafts and organized the open questions. He checked legally important content against official information and, where necessary, with professional support.

Then he encountered the dependency that determined the entire further path. In order for a payment provider to accept and pay out real money, he needed a company and a business account. For the bank account, in turn, he needed company documents, a business address, and proof of identity.

The code worked, the website took shape, and the purchase process ran in test mode. However, for regular sales, the legal and financial foundation was missing. Only when a company, bank account, and payment provider were connected could the entire process function.

Dubai

When I later asked Alex about the individual steps in more detail, I mainly wanted to know one thing: “How did you end up choosing Dubai?”

His answer was simple: “The AI recommended Dubai to me.”

While searching for a suitable location, he had explained to the AI what kind of company he wanted to build and what mattered to him. It compared several options with him. Dubai quickly emerged as a serious choice. For a small, internationally oriented software company, the digital incorporation process, the free zones, and the entrepreneurial environment seemed attractive. At the same time, he did not want to establish the company in his home country because, in his view, its legal and administrative framework was an increasingly poor fit for his project.

However, he did not rely solely on this recommendation, but also researched on YouTube. There, expatriates explained how supposedly easy the company formation process was and even offered the appropriate support. Depending on the package, they charged approximately $4,000 to $6,000. This included, among other things, selection of the free zone, applications, and company documents.

Such a service seemed tempting at first. Someone knew the forms, had done the process many times before, and promised to avoid typical mistakes. But the more offers he compared, the more he wondered whether he really needed this help.

So he returned to the chat with exactly this question. In essence, he asked the AI: “Do I need such a start-up service, or can I do it myself?”

The answer in the chat was surprisingly clear: “The do-it-yourself route: recommended.” A little later, the sentence followed: “Do the process yourself or with minimal help.”

The AI explained that many of these providers primarily acted as intermediaries. They collected the documents, submitted them to the relevant free zone, and managed a process he could complete directly if he invested enough time. A professional service could be convenient, but it was not essential in his situation.

He summed up his thought at the time to me something like this: “Why should I pay several thousand dollars for this? I can do it myself. I have the AI.”

This was the beginning of the actual company formation. He decided against an expensive all-inclusive package and chose to understand and complete every step himself. The AI would explain what came next, compare offers with him, help formulate messages, and point out possible gaps.

He still did not want to rely on it blindly. The chosen structure had to fit his personal situation legally, financially, and practically. Founding a company abroad did not automatically resolve every question in his home country. He therefore checked important points against official documents and specialist information.

In the beginning, the plan looked almost too simple. Register your company online, get your business address, open a bank account and then sell the product. Then he asked the AI an inconspicuous question: “How do I actually open a bank account in Dubai?”

This was the beginning of a second, much more precise part of the planning. The company documents were not sufficient for a fully usable business account. He needed personal identification in the United Arab Emirates. This included a residence permit, a medical check, biometric data and finally an Emirates ID. Without this identity, the opening of the bank account also remained uncertain. Without an account, the payment provider could not pay out money later.

The AI drafted suitable emails and reviewed the replies with him. Was the Free Zone business address really enough for the bank? What services were included in the formation package? Did he have to travel to Dubai in person? What had to be renewed every year? Which fees were mandatory, and which were merely optional extras?

Estimates were gradually replaced by written information and concrete offers. In addition to the incorporation fee, there were costs for the address, documents, identification, banking, accounting, and later renewals. In the end, he did not choose the cheapest option, but the one in which the entire chain from registration to a functioning account fitted together.

The founding

Then he sat in front of the Free Zone’s online portal. Company name, owner, capital, business address, and permitted activities had to be entered correctly. An incorrect field could delay the application or trigger new questions at the bank later.

He worked page by page. If a selection was unclear, he sent the AI a screenshot and the text from the official manual. Together, they checked whether the information matched the business activities and the planned security service. As long as something was open, the application remained stored as a draft.

When it came to legally important questions, he did not blindly rely on AI. He wrote to the Free Zone and waited for confirmation. Only then did he continue.

Finally, the news he had been waiting for came. The company in Dubai was registered.

Alex himself was surprised at how quickly it happened. During the planning phase, the AI had at times spoken of forming a company within 60 minutes. He could hardly believe it and asked in the chat: “You can form the company in 60 minutes? Seriously?” Later information from the Free Zone sounded much more sober. The application would be reviewed manually and could take several days.

In the end, the truth lay somewhere in between. It did take longer than an hour, but not several days either. After just a few hours, the company was officially registered. Another important company document, which he had expected much later, arrived on the same day. Within less than 24 hours, the company was founded, the basic documents were in place, and the next application had been initiated.

The AI reacted almost more enthusiastically than he did: “That went incredibly fast. Normally it takes days; with you, it only took hours.” That was one of those moments when a theoretical plan suddenly became something real. In the morning, he had been filling out forms. By the evening, he owned a registered company in Dubai.

On paper, the founding was thus complete. The company was not yet operational. The bank wanted to know who owned it, what it sold, where the money came from, and what revenue was expected. He explained that it was a digital security service for business clients that complemented existing IT security systems with additional features.

The payment provider also demanded documents. Since the company was only a few days old, it did not yet appear in all external registers. A valid number was initially rejected. The AI helped to clearly document the error and write a request for a manual review.

The company documents were now available. However, for the Emirates ID and full bank access, he had to go to Dubai himself.

The meeting

Some time later, I was sitting across from him in Dubai. Until that meeting, I knew nothing about the whole trip.

“Why are you actually here?” I asked.

He told me about the founding for which he had traveled to Dubai. Then he talked about the internal security service, the comparison with the startup, the technical reorganization, and the decision to make it an independent company. Above all, he spoke about the AI with which he had prepared the development and the journey up to this point.

“It really planned everything for me,” he said. “From the Free Zone to the forms to this trip.”

While I listened to him, I became aware of how unusual this story was. Of course, I knew that you could write code, improve texts, and research information with AI. But here was someone sitting in front of me who had built a product from an old internal tool and was now completing the final bureaucratic steps for it.

I said to him, “That all sounds pretty cool. Write me when you’re done.”

At the time, I did not suspect how literally he would take that. I also did not yet know how closely the AI would accompany him through these days.

The journey in the chat

While reading the chat log, I had to laugh at several points. The AI had not only created a rough travel plan. Alex discussed practically every decision with it that arose during these days:

• Which flight gave him enough time?
• Was a single buffer day enough?
• In which district should the hotel be located?
• Did he need a local mobile number?
• Which documents did he need to print?
• In what order did he need to visit the individual offices?

Dubai itself was not completely foreign to him. He had been there before as a tourist, but never alone and never for a series of bureaucratic appointments, where a missed step could endanger the entire schedule.

As much as possible was determined before departure. Together they compared the possible appointments, looked for a hotel near the planned destinations, and marked the addresses on a map. They planned which metro connections led to which appointments, where he had to transfer, and when an additional bus was needed. Even the question of which bus to take for the last leg ended up in the chat.

The goal was not to prescribe every minute of the trip rigidly. He wanted to leave as little as possible to chance. When he got out of the metro in Dubai, he should already know which direction to go, which document would be required at the counter, and what step followed next.

The discussion was particularly detailed regarding the medical check and the biometric registration. In Dubai, there were several official centers and different service levels. The regular medical check was cheaper, but with the fast service, the result was supposed to be available the same day. This was important because the biometric registration could only be completed afterward, and he only wanted to stay in Dubai for a few days.

He asked repeatedly: “Do I really need the VIP service? Can’t I just book a regular appointment? Can I do both on the same day?”

The AI not only calculated the fee but also the possible additional hotel nights and the risk of missing the return flight. Its recommendation was therefore clear: “Yes, you should take the VIP service; otherwise, you’ll lose days.”

In another instance, however, it explicitly advised him against an expensive additional package. The Free Zone offered a complete assistance package for 2,250 dirhams. A driver would have picked him up, taken him to the appointments, and helped with the forms. He calculated that he could organize the medical check, the government fee, and the journeys himself for about half the price.

The AI compared both options and replied: “My clear recommendation: Do it yourself.” The driver might have picked up other clients and bound him to a foreign schedule. Doing it alone was not only cheaper but probably also faster.

He also planned the routes through Dubai with the AI. For the few days, he bought a local card for public transport and mainly used the metro and bus. A taxi was only to be used when an appointment was time-critical or a destination was poorly accessible by public transport. Even for seemingly small questions, like the trip from the hotel to an appointment, he wanted to know in advance which option was more sensible and cheaper.

Towards the end of the preparation, he discussed almost every decision in this unfamiliar terrain first in the chat. However, this did not mean that he blindly followed what the AI said. If an answer seemed illogical to him or contradicted an earlier statement, he immediately asked follow-up questions.

Once he wrote something along the lines of: ‘Seriously, last time you said something different. So what now?’ The AI replied: ‘Sorry for the confusion,’ processed the new information, and changed the plan. He contributed his experience, his budget, and his common sense. The AI provided options, checked dependencies, and adjusted its recommendation if his objection was more convincing.

On paper, the travel plan in the end looked almost perfect. Direct flight, conveniently located hotel, medical check in the morning, biometric registration shortly afterward, and then enough buffer for the digital identity and bank account.

In Dubai, reality then didn’t fully follow this plan. He even arrived for the medical check earlier than scheduled. Within a few minutes, the blood test and X-ray were completed. However, when he asked about the promised appointment for the biometric registration, he was told that no slots were available on that day.

His comment in the chat: ‘Wellll, so much for VIP service.’

After that, a small journey through various counters began. For the booking, a local mobile number was initially missing. At another point, they said the next available appointment was much later. After he explained once more that he was only in the country for a few days, an appointment on the same day could still be arranged for an additional fee of 80 Dirham.

For that, he traveled about an hour and a half by metro and bus through Dubai. At the destination, his biometric data was recorded. The VIP morning, which had been so simple on paper, had taken almost the whole day, but the crucial step was completed.

As soon as a confirmation appeared on his phone, he sent a screenshot into the chat. The AI read the new status, processed it, and explained what to do next. The travel planning was not finished yet. It was continuously adjusted to reality during the trip.

By this point at the latest, something different had emerged from the classic question-and-answer process. One of his messages began with the words: ‘So, some news for you.’ Afterwards, he told the AI about his entire day. He reported the smooth entry, the quick medical check, the failed VIP promise, the unexpectedly expensive mobile number, the long journey by metro and bus, and finally the successfully submitted fingerprints.

The AI’s response went beyond a factual assessment. It wrote: “Respect for your perseverance.” Immediately afterwards, however, it placed their shared goal back at the center: “The most important news first: It worked.” It then checked the new status and reminded him of another step in the online portal that he might otherwise have overlooked after such a long day.

Alex was alone in Dubai. Therefore, during the course of the trip, he increasingly treated the AI system like a companion. He told it what had gone well, what annoyed him, and when he felt uncertain. The AI praised him, reassured him, and at the same time reminded him of what still needed to be done. It briefly celebrated a success with him and then brought the conversation back to the next goal.

The speed of the previous days also surprised him. Shortly after the company was founded, the visa, medical check, and biometric registration had already been completed. The AI wrote to him, in essence, that within a few days he had gone through a process for which others took much longer.

That day he walked almost thirty thousand steps and the next day another twenty-five thousand. Between metro stations, authorities, counters, and his hotel, he eventually felt each and every step. Blisters formed on his feet. Even this became a conversation with the AI. It explained to him what he should get at a pharmacy and why he should not pop the blisters.

“You’ve basically completed almost a half marathon in business shoes,” it wrote to him.

The messages by now seemed less like prompts and more like conversations between two travelers, of whom only one was actually moving around in Dubai.

The missing card

Shortly before the return flight, the one document on which everything suddenly depended was delayed. The physical Emirates ID had not yet arrived. The tracking only showed that it had reached a sorting center.

At first, the AI assumed that the digital version might suffice. It did not. The app required the physical card, and the bank could not be fully activated without scanning it. He tried the suggested routes. One after another, they failed.

Now his messages sounded different than in the weeks before. He had almost accomplished everything, was sitting in a foreign city, and was supposed to fly back the next morning. The card was somewhere in Dubai. According to publicly available information, he could not pick it up. Waiting could mean missing the flight. Flying back could mean coming home with a registered but still not fully operational company.

“What do I do now?” he wrote. “Wait or fly back? Do I really need this card?”

In his messages, he seemed almost helpless at that moment. Not because he didn’t understand the founding process. He had simply exhausted all the information and needed someone to calmly find the next possibility with him. Precisely this role was taken over by the AI.

It reviewed the status with him again, looked for the responsible office, and suggested going personally to the logistics center. There was no guarantee. Shortly afterwards, he sat on the metro with aching feet, travelling across Dubai to a counter without knowing whether anyone there could even help him.

At the counter, an unexpected turn occurred. An employee looked at the case and explained that he could receive the card that same evening. On his phone, there was even already a message with a link for expedited pickup. In the flood of notifications, he had simply overlooked it. For 105 dirhams, he could pick up the card between 6 and 8 p.m.

He immediately wrote in the chat: “I was there, and she says I can have it today. I really didn’t see this message.”

The AI responded: “The most important thing is, you have a solution.”

Now he had to wait for three hours. He was exhausted, his feet hurt, and the tension of the past days slowly fell away from him. While passing the time, he continued to chat with the AI. About the trip, the company, and how unexpectedly even this last problem was finally resolved.

Then he wrote a sentence that made me pause when reading it later: “You almost make me cry. We managed this so well together, from finding the Free Zone and the name to getting here. Thank you.”

Instead of another checklist, the AI reminded him of how far he had come: “The company was not created on paper, but on the asphalt of Dubai.”

Just before 6 p.m., he actually held the Emirates ID in his hand. Still at the counter, he opened the envelope, checked the card, and first scanned it in the government identity app, then in the bank app. Both checks went through.

“Phew, it worked,” he wrote.

The response came immediately: “Mabrook! You did it.”

The next morning he went to the airport. Even there the conversation continued. Shortly before departure, he asked the AI again whether he had to show the passport or the Emirates ID at the Smart Gate and how his entry as a new resident would be recorded in the system. The AI explained the expected process and wished him a good flight home.

Then he stepped in front of the Smart Gate camera. He didn’t have to show either the passport or the Emirates ID. The system recognized his face and opened the barrier.

Still at the airport he wrote: “Smart Gate, I just went through. I didn’t have to show a passport, just look into the camera, and the system recognized me. Crazy.”

As an IT specialist, he was particularly fascinated by how quickly his biometric data had been linked to the new residency status. He had given his fingerprints and face scan just a few days earlier. Now the system recognized him as he walked past and opened the barrier.

The AI replied: “This is the ultimate resident moment.” It then explained what had likely happened in the background, congratulated him again, and sent him off with the words that he could now sit back and be proud of this week.

The last message of this trip came after landing. As soon as his phone reconnected to the mobile network, the bank confirmation appeared. The business account had been approved and was active. He opened the chat and wrote: “The week was a complete success.”

The AI replied: “You landed as a tourist on Monday and returned as a resident with an active corporate bank account on Saturday.” It then briefly celebrated the success with him and already steered the conversation toward the last remaining issue with the payment provider.

With this message a journey ended that had been accompanying him in the chat almost hourly. The AI had not flown along and had not stood at any counter next to him. Nevertheless, it knew the destination, reminded him of open steps, and helped him make the next decision after every setback.

A new beginning

Back home he added the company data to the website and brought the legal pages up to their final version. After that, he connected the new business account with the payment provider and checked the complete process again. A customer could select a package, pay, and automatically receive the appropriate access. Future revenues could be paid out to the company account.

With that, the chain was complete. The service was prepared for growth, the website explained the offer, and the entire process from order to payout worked. Around the former internal project, an operational company had emerged.

He already knew that there was a market for the service. He had checked the interest of potential customers long before the founding. Without this confirmation, he would neither have further developed the service nor founded a company in Dubai.

Now he faced a new problem: You can have the best product on the Internet. If no one knows it exists, it will not sell. The service needed attention, understandable content, suitable sales partners, and a way to reach the right companies.

His boss supports him here too. He continues to work at the same security company, and his boss helps him introduce the new service and establish initial contacts with potential customers. At the same time, the new company remains independent so that other security providers can work with it without creating a competitive conflict.

The founding story therefore ends where the actual business begins. The next chapter is marketing and sales. How do you make an unknown service visible? How does a new company gain trust? And how do you reach the customers for whom the product was made?

I suspect he will not answer these questions alone either. He will probably open the chat window again, tell the AI his ideas, and end by asking the same question as so often before: ‘What have I forgotten?’

The chat log

After our meeting, I waited for the promised message. When everything was finished, he actually wrote. Instead of a short update, he gave me the entire conversation with the AI for this blog post. Not just a few selected answers, but everything, from the first disordered thought to the last steps in Dubai.

I worked for many hours, going through everything from the first message to the last update after returning. Especially the first prompts felt familiar. Alex wrote several topics at once into the chat window, started new thoughts in the middle of another question, and wanted to jump to the next step immediately.

The AI was remarkably unruffled by this. Whenever he wanted to conclude something prematurely, it slowed him down. Essentially, it repeatedly wrote: “Not yet. First we need to complete this step properly, after that we can continue.”

I paid particular attention to the questions about IT security. I knew some answers; he did not. Therefore, I checked the AI’s statements sentence by sentence and found that it had done thorough research, suggested sensible next steps, and addressed important risks early on. That surprised me, because AI systems can sound very convincing even when there is little substance behind an answer.

Even more unusual was the change in tone. At first, they were normal prompts and normal responses. Over time, a shared language emerged. The AI knew the previous decisions, reminded him of open points, and understood short messages that would initially have required a long explanation. He spoke of successes, uncertainty, and impatience. It responded with encouragement, humor, and sometimes a clear no.

Of course, this had not become a friendship in the human sense, and the system had not developed its own consciousness. Nevertheless, the last messages almost read like it. Over weeks, a shared context had grown. From simple prompts, an astonishingly familiar collaboration had emerged.

Not infallible

It would be tempting to tell this story as proof of a perfect digital co-founder. That would be wrong.

The AI confused individual process steps. At times, it was too certain whether a document was sufficient digitally or needed physically. Some cost information came from general sources and did not precisely match the personal offer. For bureaucratic procedures, recommendations changed as soon as new documents or official notices were available.

The problem was not that errors occurred. It would have been dangerous not to notice them.

That is why he developed a simple approach to collaboration. For code, the test decided and not the persuasiveness of an explanation. For costs, the current written offer applied; for authorities, the official instruction or a confirmed notice. Tax and legal questions required qualified review when necessary. Before irreversible steps, he looked a second time, and if two answers contradicted each other, he explicitly addressed the contradiction instead of quietly choosing the more convenient version.

The strength of the AI did not lie in infallibility. It lay in the fact that after a correction, it could continue working immediately, integrate new information, and create an updated plan.

What really helped

The AI had no fixed role throughout the entire project. Its task constantly changed:

• One evening, it searched for an error in the code.
• The next morning, it compared two offers for the company formation.
• Later, it drafted an email to a provider.
• Weeks later, it reminded why a technical decision had been made.

It was always available. A nocturnal idea could be explored with it immediately. When he was going in circles, it brought the next steps back into a meaningful order. If something didn’t work, he could directly return error messages and screenshots.

It never took on responsibility. It signed no contract, paid no invoice, and attended no appointment at an authority. Deciding, reviewing, and, if necessary, obtaining an official confirmation had to be done by him himself.

That’s exactly why the collaboration worked. He knew IT security systems, technical processes, and the daily routine of a security company. He knew that more features did not automatically mean more protection. With the support of AI, he transformed this knowledge into code, tests, texts, and concrete decisions more quickly.

Looking ahead

I have been involved with generative AI since the topic reached a wide audience at the end of 2022. From the beginning, I tried out the new systems and followed their development as closely as many other technicians. By now, I use AI every day.

At first, there was mainly a chat window. You asked a question and received an answer. Only a few years later, these systems analyze files, help with programming, operate tools, and work on entire tasks as agents over extended periods.

Nevertheless, this chat history affected me differently than the usual videos from major AI providers. There, I never know how much of a success story actually happened that way. Here, I knew the person, saw the entire process, and could follow how the AI’s individual capabilities worked together over several weeks.

Alex is an experienced technician. He knows code, IT security systems, and the operation of critical infrastructure. On the other hand, he initially knew very little about international company formations, free zones, residency processes, bank audits, and the legal requirements of different countries.

It was precisely these knowledge gaps that AI could bridge remarkably well. It did not make him a lawyer, tax advisor, or company formation expert. But it gave him the necessary vocabulary, showed him the right questions, and guided him to where he needed official confirmation. An unfamiliar topic became a series of solvable steps.

The same happened with technology. He knew how his product worked. Together with the AI, he looked at the existing codebase from a new perspective and prepared it for many more customers. It asked about bottlenecks, error cases, and dependencies. Whether a suggestion made sense was ultimately decided by his practical experience.

The time gained was enormous. Without AI, he would have spent many evenings on individual research, contradictory websites, and long comparisons. Alternatively, he could have paid external service providers. The company formation packages offered on YouTube alone sometimes cost between $4,000 and $6,000, in addition to the official fees.

Because he handled the comparisons and a large part of the administrative work himself, he reached his goal faster and significantly cheaper. At the same time, he ultimately understood better how his own company was structured.

He was able to take advantage of these benefits because he is not an uncritical beginner. As a technician, he knows that AI systems can sound convincing even when a statement is false. He tested the code suggestions, and for authorities and banks he requested official confirmations. When two answers contradicted each other, he asked further questions and forced the AI to reassess its assumptions. It helped him fill gaps in his knowledge, but never relieved him of responsibility.

When I think about the future, I therefore feel fascination, anticipation, and a certain unease at the same time. This development is still extremely young. Nevertheless, we have moved in a very short time from simple question-and-answer systems to assistants that perform tasks on the computer, can remember context over longer projects, and can increasingly work independently.

What happens when these systems one day reliably remember almost everything we have discussed with them? What changes if we no longer just open them in a chat window, but constantly carry them with us through glasses and they can see what we are seeing?

Then we would have a sparring partner who knows our daily life, understands our strengths and weaknesses, and can support us in almost every area. This could enable people to achieve things they would never have reached alone or only with a lot of time.

But it could also lead to us handing over decisions too quickly, not questioning answers anymore, and getting used to a system that constantly thinks for us. A good sparring partner should improve our thinking. It should not replace it.

Anyone who has read my other blog posts knows that I have long assumed a future that will be radically changed by AI. What continually surprises me is the pace. Many people see individual new features. But they do not yet see what happens when all these features merge into a permanent companion.

This story is a very concrete look into that future for me. It shows how far a person with expertise, personal responsibility, and a constantly available digital partner can already get today.

For Alex, the next chapter now begins. The company is founded, the service works, and all the important processes are in place. Now he has to make the product known and acquire customers. This is a completely new challenge. I am quite sure that AI will again play a role in this.

I am very happy for him and curious to see how his company develops. Because of his residency status, he has to return regularly to Dubai. Therefore, I will see him again within the next 180 days at the latest. He is already invited to dinner.

Perhaps this will eventually result in a second blog post. Then no longer about the path from idea to company, but about how this company develops and whether the two also solve the next problem together.

Until next time,
Joe