StoneX Group Inc’s CTO Boris Levine does not sound like someone waiting for the future to arrive. He sounds like someone who has already lived through five different versions of it and is now watching the rest of the world catch up.
Levine holds the number one spot on HackerEarth, with a focus on mathematics and algorithms, and that sets the tone. He codes. He thinks in systems. He sees patterns long before others feel the tremor.
He has been at StoneX for years, but his job description keeps expanding. He is trying to turn a decades-old financial firm into a full-scale tech company. He is pushing teams across continents to rethink everything from testing to customer support to the code they ship.
And he is doing all of this while flying between conferences, reviewing hackathon pitches, helping shape Gartner as the member of the CIO Research Board, and sitting in front of developers who keep asking him whether the world is shifting beneath their feet.
He believes it is. And he thinks we are early. “Impossible just takes longer,” he said. It is the closest thing he has to a motto, and much of his work is built on that line.
When he landed in Bengaluru at 3 am on November 14, the StoneX hackathon was in full swing with 30 teams across the city, as well as in Pune, and more than 130 participants. The problems ranged from HR to data engineering to operations.
But what really excites him is not the hackathon itself. It’s what sits underneath it—a shift in how developers think about creating things in a world full of AI tools. “Everyone is very excited about possibilities and opportunities,” he said, coming straight from the Gartner summit in Barcelona. As part of the CIO Research Board, he’s spent four consecutive summits in conversations dominated entirely by AI, and its effects on engineering, business processes and customers.
What Would Levine Do?
Levine sits right at the centre of two worlds. One is the traditional financial stack that demands caution, stability and regulatory clarity. The other is the speed at which new models are breaking every expectation about what is possible in software development. That tension shapes almost every decision he makes.
StoneX uses Copilot and other tools across teams. “We can see very good signs of development process improvement in terms of efficiency,” he said. The company has built proofs of concept for automated testing, automated script generation and faster transition work. But he refuses to hand over full product creation to AI.
“We need to be very careful about what we can put in front of the customers, how secure, how stable the solution is,” he said. The company is regulated by dozens of bodies around the world. Every line of code carries weight and liabilities. And the same question keeps returning: how do you push the edge of technical possibility while living inside a system that cannot afford mistakes?
Levine believes the industry is still stuck in the wrong place.
Enterprises are chasing productivity when they should be chasing process redesign. Studies back him up. “Only 5% of the projects deliver return on investment,” he said, pointing to an MIT figure. McKinsey & Company’s recent report shows most pilots remain stuck in the POC bucket. Everyone is chasing faster development, faster operations, faster monitoring.
But the real return, he argues, will come only when companies change how the process itself works, not when they plug AI into the old version of it.
He is blunt about where the bottleneck is. “We’re not yet there,” he said. Not just because of technology, but also because of regulations. Financial firms cannot let models make decisions that touch clients without rigour, which does not yet exist. That is the gap, and it slows everything else down.
He is equally blunt about the limitations of current models. “You cannot fit your entire code base in the context window,” he said. Increasing the context window only creates new problems: cost, loss of focus, and error rates that spike with scale. If the window gets to a million tokens, even a small margin of error compounds. The model loses coherence. It drifts.
The path forward, he says, is clear—composite models, knowledge graphs, better architecture. Systems that work more like people do. “Highly skilled developers don’t remember every single line of code,” he said. They remember ideas, concepts and patterns. Models need the same structure. Context windows cannot get us there. Knowledge systems can.
What About the Future?
Levine has been following Meta chief AI scientist Yann LeCun’s work and also believes in world models. He does not believe anything today can truly reason across steps. Current systems try to process entire problems at once instead of breaking them down. “Very complex problems usually need to be split in different stages of proof,” he said.
That, he argues, is the skill missing from all large models today.
He pointed to another shift he is watching closely: edge AI—models that run on phones. He explained it through the lens of search. Today, StoneX and other firms rely on Google ranking. They know what matters: mobile speed, layout stability, relevance and token match. These rules determine how clients discover trading products. But if users move from Google to local models on their phones, everything changes.
“Users will stop searching Google. They will start asking questions to the local LLM model,” he said. When that happens, the entire onboarding funnel changes.
This leads him into a deeper fear. Advertising and sponsored answers. A future where on-device models suggest trades, products or even medicines based on whoever paid for influence. “How can you trust your helper on your phone that it actually tells you something?” he asked.
This kind of thinking is what makes him hard to categorise. He moves from pure code to system behaviour to sociology with the same clarity. He has done this before. His years at Intel shaped how he sees the hardware problem. The gap between the human brain running on “about 40 watts” and current models consuming kilowatts or more is, to him, absurd.
He has seen early attempts at new chips, and he believes hardware will unlock the next leap. He believes these chips will make AI so cheap that it will sit inside kettles, irons, refrigerators and everything else. That is the future he sees.
What Happens Inside StoneX?
Inside StoneX, the work is more grounded. The company is using AI to simplify customer service, unify operational data and remove the need for employees to jump between systems. “They can just ask our internal systems,” he said. The system converts questions into API calls, collects data and retrieves answers.
It changes speed and quality. On the development side, AI tools reduce PR review time, security checks and migrations. They accelerate the boring parts. They free teams to think better.
But he knows this comes with risk. “Three months down the line, I need to change something. I have a bunch of code that nobody understands,” he said. Models evolve. Vendors change. Context changes. Maintenance becomes a nightmare. Technical debt balloons. This keeps him cautious.
When it comes to clients, he refuses to move fast. He knows regulators will not accept AI-driven advice without a foundation that cannot be tricked by prompts.
“We need AI that is guarded and controlled, not by an extra system prompt,” he said. It must be built into the model. Something that cannot be overridden. Something that refuses to act outside its role. That, he says, will open the door to serious enterprise adoption. He does not believe synthetic data solves this. Only responsible AI will.
He is equally firm about where India fits into the future. StoneX has long-standing operations in India, with offices in Pune and Bengaluru.
India has scale, talent and deep pools of engineering talent. Pune brings payments and banking strength. Bengaluru brings universities and a flood of young developers. StoneX has invested in both for a reason. The offices are growing fast. The company sees them as long-term investments, not cost centres.
He paints India as one of the new bright centres of technical innovation. While the hackathon is one signal, the pace of hiring is another. But the shift in developer mindset, he says, is the biggest one.
When he talks about art, he becomes almost reflective. He loves painting. His favourite artist is Rubens Santoro. He slows down. He describes light, mood and influence. He talks about creativity as a way to detach from day-to-day technical intensity.
He sees programming as creative work. In places like Poland, developers are even given tax benefits for working in creative professions. He says some of the best engineers have no artistic hobbies. The connection, he says, is personal, not universal.
He does not fear AI taking over art. “Are we going to see the next Gustav Klimt?” he asks. He does not think the current mathematical models can produce anything that breaks out of the mainstream. He says the same about code. He does not expect AI to produce a new way of organising computation. It can remix, but it cannot invent the next architecture. Not yet.
But what about AGI? Levine says motivation, purpose and the ability to verify intelligence sit at the centre of the question. He has opinions but no definitive answer. “I would need to think about it,” he says when asked how we would even test real intelligence. Current models can easily pass Turing-style checks, he noted. They still have no purpose. No internal drive. No reason to choose one action over another.
He believes that true intelligence needs motivation built into its core.
The post ‘Impossible Just Takes Longer,’ Says StoneX Group Inc’s Relentless Super Coder CTO appeared first on Analytics India Magazine.
