AI won't take your job. Someone who uses it will.

A random idea, a viral trailer, and a tool nobody knew was Pakistani

Muzamil Hasan opened the episode, recorded for Thought Behind Things’s collaboration series with Endeavor Pakistan, with a demonstration rather than a question. Two days before recording, his brother, a cybersecurity engineer in the US with no art background and no connection to Pakistan, had made an AI-generated trailer imagining a world where the Mughal Empire never fell. It had gone semi-viral on YouTube. When Muzamil mentioned he was about to interview a founder in Islamabad, his brother told him about the tool he had used to make it: Imagine Art. He had no idea it was built in Pakistan.

“Mere liye badi fascinating baat thi,” Muzamil said, describing the moment he realized someone entirely outside the industry, with only an idea and no capital or training, now had a real shot at building his own creative universe using a tool built a few hundred miles from where Muzamil was sitting. He called the guest waiting across the table the founder and CEO of vyro.ai, Ahmed Abubakar, and said that of the more than twenty episodes already recorded that season, this was, in his words, “hamari season ki sabse mazedaar aur sabse powerful story,” the most interesting and most powerful story of the season.

The boy who sold water tanks in Karachi

Ahmed Abubakar’s story does not start in a garage or a dorm room. It starts on a rooftop in a village near Sargodha, where he used to hide from a bus driver because he did not have the fare, and later on the back of his father’s Suzuki, filling small water tanks at Surjani Town to sell across Gulistan-e-Johar. He was thirteen.

His father drove a taxi and, when that stopped working, switched to selling water door to door. Ahmed asked to come along. “Main na Abu Paani wala aur mukhtalif flats mein ja ke naam janaab shuru hua hoon,” he said, describing how he wrote “Ahmed, the water seller” on flats around the neighborhood and started taking orders directly from residents. It became his first lesson in what he later called marketing: showing up, communicating well, and being reliable enough that people called him back.

His mother died when he was in his first year of college, shortly after a last conversation in which she asked him to find her a verse in the Quran and prayed for him to get into NED, Karachi’s oldest technical university. He failed that first year. He repeated it, studying in an empty neighborhood park from seven in the morning, and turned a 2.5 GPA into a 3.94 by his final year at NED, where he studied computer systems networking and telecommunications.

Muzamil Hasan, hosting the conversation, stopped to name what that arc actually represented. “Agar main bahane banane baithoon na is kahani mein har bahana hai,” he said, “you know, mere saath to itni tragedy ho gayi, mere paas paise nahi hain, main middle class se aaya hua hoon, mere connections nahi hain, mera exposure nahi hai. But aapne us saari cheez ko prove wrong kiya.” Every excuse a person could reach for was available to Ahmed Abubakar, Muzamil said, and none of them stuck.

Three brothers, one room, and a viral background changer

After NED, Ahmed Abubakar joined Cybernet, then Afiniti, then in 2016 started pushing his younger brothers, Zain and Abdullah, toward computer science at NUST. He funded their first tuition payments from his own salary; they later won a Fulbright-funded scholarship. By 2019, all three were experimenting nights and weekends out of a small NUST research lab, testing machine learning models on cheap edge devices.

Their first serious idea, a mirror that would change a customer’s hair color in real time inside a salon, went nowhere. So they pivoted to something simpler: a background-changer photo app, modeled on what a Vietnamese team was already doing well. “Agar aap innovate nahi kar sakte, bahut achha, aap please bahut achhe ko copy kar lo,” Ahmed Abubakar said. “Aap jab usko copy karoge, you will be able to make better ways by sometime.” The app went unexpectedly viral in India after they added a feature letting users swap in Paris backdrops, and it crossed $50,000 in monthly recurring revenue during Covid, while all three brothers were still crammed into one small house in Islamabad.

Ahmed Abubakar remembered the first real payday from years earlier, the one that made the later number feel possible at all: selling a water tank for 70 rupees, two tanks for 140. “Mujhe ghar yaad aaya tha jab ghar pohanche,” he said, “to I was super happy, ke hum ab ghar mein khush reh sakte hain.” Muzamil pressed him on the psychological jump between that memory and a five-figure monthly number landing in an account for the first time, calling it “a phenomenal story of breaking glass ceilings and not having mental barriers.”

From ad-funded apps to a Google-recognized brand

The background changer was followed by an object-remover app, then several more, all funded by Google AdSense rather than subscriptions. In 2021, Google emailed the vyro.ai team directly, inviting a partnership after noticing how well their apps were performing. By 2022, Ahmed Abubakar and his brother Abdullah were flown to Singapore as one of Google’s top 50 developers.

That trip changed the scale of their ambition. “Pehla, Covid time ke andar mujhe yaad hai ke hum log likh rahe hote the ke pehle din hum $10 per day pe kaise jayenge,” Ahmed Abubakar said. “Ab hum Singapore mein baith ke yeh soch rahe hain ke yaar hum kis tarah se lakh dollar per month pe kaise jayenge.” When ChatGPT’s release reshaped what customers expected from AI products in late 2022, the team built their own text-to-image model, Imagine 3.1, and launched Imagine Art. Within one quarter it was competing with Midjourney and Canva for a top-three spot on Android’s art and design charts, and the team shifted the whole business from ad revenue to subscriptions.

By Ahmed Abubakar’s account in this conversation, vyro.ai is on track to close this year near $25 million in annual recurring revenue, with roughly 150 people in Pakistan and another 10 to 12 working internationally, serving a product he says has crossed 150 million downloads.

Foundational models are the operating system. We are the apps.

Much of the middle of the conversation is a debate about where value actually sits in the AI stack. Two years earlier, Ahmed Abubakar and Muzamil Hasan had argued about whether “wrapper” companies built on top of foundational models would survive. This time, Ahmed Abubakar returned to the argument with a clearer frame: foundational model companies like OpenAI have effectively built the operating system layer, and the businesses built on top of them are the apps, the actual usable experience most customers want.

“Jo humne is industry ko start kiya, ab alhamdulillah hamare paanch saal ka itna massive experience ho gaya hai ke aapne sirf model ke upar nahi play out karna,” he said. “You have to give an application layer experience.” He pointed to Sora and other model providers gradually opening public APIs as evidence that even the biggest labs now depend on the application layer to reach customers, rather than trying to serve them directly.

Muzamil pushed on the harder question sitting underneath that comfort: whether any AI company, wrapper or foundational, will still exist in fifteen years, given how fast the underlying economics of the industry are shifting. Ahmed Abubakar’s answer leaned on proximity rather than prediction: staying so close to the frontier, watching every model release, every price change, every new capability, that the team can adapt before a shift becomes a threat. “Humne apne aap ko itna zyada penetrate kiya hua hai is culture mein,” he said, “although my team is in Islamabad, we work like hum valley ke andar hi baithe hain.”

Muzamil raised the industry’s mood swing directly: two years earlier, the dominant fear was that foundational models would eat every application built on top of them. Now the dominant narrative had flipped, with commentators calling the entire AI buildout a bubble, comparing it to 1998-era internet overvaluation, and citing OpenAI’s reported multi-dollar spend for every dollar of revenue it collects from customers. Muzamil framed his own read of the moment carefully: AI is probably in a bubble the same way the internet was in 1998, overvalued and due for a correction, but that does not mean the underlying technology is not the future. The real risk for any founder, he argued, is not the correction itself, it is getting knocked out of the game before it arrives. “Ek chhakka marna toh bada asaan hai,” he said, “lekin darmiyan mein jo sabse zyada badi cheez hai woh yeh hai ke is bubble ke andar out nahi hona. Khelte rehna hai, test match ki tarah khelna hai.”

Ahmed Abubakar’s answer leaned on speed and closeness to the frontier rather than a prediction about timing. He described spending recent weeks in San Francisco meeting ByteDance’s team, who told him their video and image models had so far trained on a fraction of the content available to them, meaning Chinese labs still have significant unrealized capability. His conclusion was that no one, including the best-funded labs, actually knows which direction the market moves next, and that the founders who survive are the ones checking constantly rather than committing to a fixed roadmap. “Kisi ke paas koi answer nahi hai,” he said, pointing to conversations with executives at companies like Klarna as evidence that even mature, well-resourced businesses are improvising in real time rather than following a known playbook.

Profitable without funding, and raising anyway

Vyro.ai has never taken outside money. Ahmed Abubakar said the company is EBITDA positive, built entirely on subscription revenue, and had to build its own systems to manage the gap between what a subscription costs a customer and what the underlying model inference actually costs the business, capping usage the way ChatGPT does rather than losing money per user.

That discipline is now colliding with a different kind of ambition. Vyro.ai opened a San Francisco office last month, and Ahmed Abubakar has made three trips to the US this year meeting venture capitalists, because the brand he wants to build, positioned against Canva and Figma rather than regional competitors, requires a kind of talent and credibility that bootstrapping alone will not buy. “Aapka jo brand hai, bahut bada part brand hai,” he said. “With fund, you will be able to make a great brand. Yeh brand local ecosystem ka nahi hai, yeh ek value-based brand banana hai.”

Investors pushing back on a consumer-only product have already redirected the roadmap: vyro.ai is launching team plans, single sign-on, and SOC 2 compliance this quarter, aimed at the marketing agencies and businesses that Ahmed Abubakar says already make up its highest-paying customer segment.

Talent is an open market, not loyalty

Asked directly about retention, Ahmed Abubakar rejected the framing that Pakistani founders often use when engineers leave for international jobs. “Talent open market hai na, capitalism hai,” he said. “Koi yeh thodi hai ki loyalty-shoyalty kuch nahi hota jab tak pet nahi bhara hota.” He argued the real question is not how to guilt people into staying, but what model gives an engineer a reason to believe that time spent at the company pays off later.

Vyro.ai has recently completed a company valuation, a prerequisite for setting up a formal stock plan, and is opening an employee stock ownership pool for the first time, alongside plans to bring senior engineers who have worked in the US onto an advisory structure. Ahmed Abubakar also described operating with the lean team structures he saw firsthand while raising in the US, arguing that smaller, faster teams grow companies faster than large ones.

He credited part of that thinking to conversations with the founders behind Cloudways and Gaditek, a Karachi-founded hosting and web-services group whose brothers Ahmed Abubakar described as informal mentors. “Attitude of learning right, aap apne aap ko hamesha is tarah portray karein ki aapko kuch nahi aata,” he said, describing what he took from those calls, and noted their teams now collaborate directly on shared AI and churn-reduction work through Mixpanel. Ahmed Abubakar said vyro.ai is weighing a second international office, likely in the UAE, partly because Gaditek already operates there and the time zone overlaps far more cleanly with Islamabad than San Francisco does.

AI won’t take your job. Someone who uses it will.

On the future of creative work, Ahmed Abubakar’s position was specific: content production volume is rising because AI is lowering the cost of making things, not because it is replacing the people who make them. A shopkeeper who once needed to hire a graphic designer for a logo can now do it directly. A creator who once needed a small team of visual designers can run a faceless channel alone.

“Jobs replace AI nahi kar raha,” he said. “Jobs replace ho rahi hain by someone who knows AI.” Muzamil built directly on the point: the person doing the jingle work is not disappearing, but they are increasingly using an AI tool to do it, meaning the replacement is not human-to-machine but human-to-more-capable-human. Ahmed Abubakar agreed and pushed it further, arguing that within five years, millions of Pakistanis could plug into global services work simply by learning one tool well enough to serve ten small businesses at once, rather than waiting for one employer to hire them.

The real bottleneck is process, not the model

The sharpest technical argument in the conversation came late, when Ahmed Abubakar turned his own recent detour into songwriting into a case study. After writing a song for his late mother and releasing it, he described the frustration of using Suno to generate music quickly, only to find every result technically clean and emotionally empty. “Woh jo soul nahi hai,” he said, describing what was missing. “Ab mujhe wohi masla ho gaya jo pehle tha. I need to know something about something to be able to create something powerful.”

He extended that into a broader theory of why most AI creative tools underdeliver: they map the technical steps of a craft while skipping the part that actually took a professional years to learn, extracting the right information from a client before ever touching the tool. He described the difference between a designer who asks “what do you want?” and one who interrogates a client’s philosophy and audience before designing anything. “Woh client management ka aspect abhi tak tools mein nahi aa pa raha,” he said. “Mujhe us consumer se genius tarike se information extract karni hai taake main usko relevant product nikaal ke de sakoon.” His prediction is that the next generation of AI tools will not just generate content, they will interview the user the way a skilled professional would, before ever producing an output.

Education is the greatest equalizer

Closing the conversation, Muzamil Hasan asked Ahmed Abubakar what he would tell a young Pakistani with no connections, no capital, and no clear direction, given everything he had seen change in the industry over five years. Ahmed Abubakar’s answer returned to where the conversation started: his father, who never had a formal education and worked nights setting up a cart outside wedding halls, and a family that had nothing to hand down except the insistence that his brothers finish their degrees.

“Sabse important, everyone knows ke education is the enabler, is the basic necessity of every individual,” he said. He laid out what he called three requirements in sequence: education first, then a vision worth working toward, then the discipline to put in the hours. “Pain is one,” he said. “Ke yaar, education lein, khwab rakhein aur effort karein usko achieve karne ke liye.”

Muzamil closed the episode by turning the message toward students and early-career professionals directly, arguing that the value of Ahmed Abubakar’s story is not the destination but the refusal, at every stage, to treat a lack of resources as a reason to stop.