Dmitry Trub has spent years helping financial businesses put technology to work. Then his clients asked a question that was awkward precisely because it was reasonable: how much more productive had AI made his engineers? He could offer an experienced judgment. He could not offer a measured figure. An impression had arrived at a meeting where a number was expected, wearing a rather good suit.
So the founder and CEO of Findev made his own company the experiment. In September 2026, he began publishing an account of its attempt to bring AI into the machinery of the business. The outcome remained open. The promise was to report results, costs and mistakes as the work unfolded. For someone whose trade is advising others about technology, opening the workshop door invites an unusually close inspection.
The appeal of the story lies in its scale. Findev, as Trub described it that September, had 180 people, entities in several countries, and work spanning currencies, languages and jurisdictions. There were enough moving parts to make a demonstration interesting, and enough responsibilities to make a careless demonstration expensive. His experiment would have to live inside a business with clients expecting their ordinary working day.
September 2026 account
October 2026 account
Before the experiment, the trading floor
Trub’s earlier career helps explain the attention to things that must keep working. Before Deutsche Bank, he held leadership roles in software businesses including Exigen and First Line Software. A conference biography describes him building a team of more than 200 IT specialists at First Line. His education included applied mathematics and computer science at Saint Petersburg State Polytechnic University. The combination put people, delivery and technical reasoning in the same working life.
He joined Deutsche Bank in 2011. By the following year, a software engineering conference identified him as the head of its Connectivity department in Russia, responsible for how the bank connected with the outside world. Connectivity sounds like a modest word for an immodest dependency: a financial institution is of limited use if its systems cannot speak to the systems beyond its walls.
Later roles included co-head of trading technologies at the bank’s Russia Technology Centre and director and head of global market equities in Russia. The responsibilities covered software development and production stability. These are careers in which an elegant proposal still has to answer for what happens on Tuesday morning.
In July 2019, BCS Global Markets appointed him chief information officer, based in Moscow and reporting to CEO Roman Lokhov. His remit included investment banking software, institutional client service, IT strategy, operational automation, regulatory reporting and connections to new markets. It was a brief with very little room for technology as decoration. The systems had customers and consequences attached.
The company clients asked for
Findev’s founding account begins with existing relationships. Its core team had worked in finance-specific software engineering for years; clients encouraged it to assemble a global team. The company dates that step to 2020. Trub became the founder leading a business organised around knowledge accumulated inside financial technology, including his experience building platforms with global engineering teams at Deutsche Bank.
That origin gives the enterprise a particular shape. It grew from work and client demand, rather than a consumer product seeking its first audience. Findev presents its work around the demands of capital markets: growing volumes, changing requirements and software that must remain dependable. Its stated consulting values include candid conversations about what customers need. The practical test of that principle is whether the advice remains useful after the presentation ends.
Trub’s views on management also extend beyond architecture. In a September 2023 interview, where he was introduced as CTO of EXANTE, he discussed coordinating engineering teams across regions. He described shared training, regular communication, documentation and demonstrations of recently delivered products. Data could help teams see project timing and bottlenecks. Occasional meetings in person still mattered. The organisation needed ways to understand its work and ways for people to understand each other.
The assistants had arrived. The invoices had other plans.
When Trub took stock of Findev in September 2026, individual use of AI was already part of working life. Colleagues used company-funded tools for documents, contract checks and small calculations. Enthusiasm varied. Some people had found a place for AI throughout their day; others had tried it and remained unconvinced. He treated that difference as something to understand.
His description of the company’s operations was more familiar. Invoice reconciliation still used Excel formulas. Accounts and access permissions still required manual provisioning. Routine administration continued through established processes. Personal adoption and organisational change were proceeding at different speeds, inside the same business.
“Adoption follows people, not org charts.”
Dmitry Trub, September 2026
There is a certain comedy in the scene, recognisable to anyone who has watched an office acquire the future while retaining its favourite spreadsheet. Yet Trub’s attention stayed on responsibility. Nobody had been assigned the work of changing those processes. A subscription had a purchaser; the transformation needed an owner. The distinction helps explain why capable tools can coexist with remarkably durable habits.
Four answers, before buying another tool
His proposed method gave each recurring task four possible destinations: leave it with a person, automate it with a deterministic script, use an existing outside service, or build something with AI. That range made the exercise more interesting than a tour of new software. The question started with the work, leaving the machinery to earn its place.
He also intended to publish costs, including token bills. A bill records consumption; deciding whether the consumption was worthwhile requires another calculation. The experiment therefore set itself a harder assignment than showing that an AI system could complete a task. It would have to ask whether the arrangement made business sense.
Trub brought a long memory of adoption problems to the exercise. He had seen technically successful automation projects fail because people did not want to use them. That experience puts a useful restraint on a founder’s enthusiasm. A working demonstration can be applauded by everyone in the room and quietly ignored by everyone the next morning. Daily usefulness is a less theatrical achievement.
Back to school, with the company still running
To learn from people further along, Trub and colleagues went back to school. His expectation was modest: “I expected to nod politely.” Instead, the practitioners’ working systems left the group feeling behind. Seeing tools carry real company work changed the conversation from possibility to preparation.
The preparation concerned context. A model would need information about clients, arrangements, invoices and the decisions made in recent conversations. A person supplies much of that background to a personal assistant almost without noticing. Extending assistance across an organisation requires the background to become accessible beyond any one person’s memory.
He also noticed the supporting tools around the models. Findev began using Granola to capture meetings. Wispr Flow let him dictate, and he wrote that a good portion of the series took shape during walks. It is an appealing detail: a founder considering business automation while letting his own sentences arrive at walking pace.
The course sharpened his concern about disruption. Existing operations had to continue while changes were introduced. His immediate conclusion was that building agents would be premature. That pause has the flavour of his earlier production responsibilities: enthusiasm can propose a new system, while the working business retains a vote.
The freshest fact is still in someone’s head
By October 16, his writing had focused on keeping context current. Findev already maintained centralised records of employees, clients, contracts and documents. But the latest conversation could change what those records meant before an update reached them. A neat archive could still be behind the business it described.
That creates an awkward failure mode. An assistant can answer fluently using information that has expired. The visible confidence survives the invisible change in circumstances. Trub’s concern was the resulting erosion of trust: people need the system to understand the company they are working in today.
Findev’s distributed habits gave it a useful starting point. Much of its communication already happened through text and online calls. Capturing those exchanges could help preserve information that otherwise remained with the participants. The problem reaches into ordinary management: what has been decided, who knows it, and how does everyone else catch up?
Nine functions, and a little corporate archaeology
On October 22, Trub described the next exercise: writing Findev down. The team sorted recurring responsibilities across nine functions, from corporate management and delivery to finance, legal, human resources, operations and IT. They separated roles from the people performing them. One person could keep several roles; each responsibility now had a place in the map.
The exercise revealed familiar improvisations. Duties had accumulated with whoever was convenient, including the CEO, sometimes crossing formal departmental boundaries. At Findev’s size, a separate job for every responsibility would have made little sense. The arrangement worked because the people involved understood it. A new colleague, or a machine, needed the explanation in writing.
Some named functions had no dedicated team at all. Naming them made the work visible anyway. That is a revealing place for an AI project to arrive: a business describing the arrangements that allow it to function, including the ones so ordinary that nobody had bothered to document them.
For Trub, the next answer will have to emerge from that work. His public experiment has supplied a baseline, a set of choices and a growing account of the company itself. The invoices still need to reconcile. The clients still need their systems. Somewhere between those obligations and the new tools, the improvement will have to become countable.