The software found 22 definitions of revenue. The people had to decide what to do with them. For Nas Radev, this is the sort of discovery that makes enterprise AI interesting: an apparently technical assignment arriving, rather inconveniently, at a question about how a company understands itself.
The client was an attractions and leisure operator. Radev describes a business with multiple sites and millions of visitors, whose systems had accumulated different ways of counting money. A ticket could enter the calculation when it was sold or when someone visited. An annual pass could be counted immediately or spread across the year. Different teams had sensible reasons for their choices.
Then an AI agent was asked for revenue. It had plenty of numbers and no settled instruction about which meaning to use. Anyone who has watched colleagues debate a spreadsheet will recognise the difficulty. Giving the spreadsheet a conversational interface does little to shorten the debate.
Radev, co-founder and CEO of Infinite Lambda, has built his current argument about AI around this unglamorous territory. The systems need rebuilding. The definitions need agreeing. The people doing the work need to understand the result. His route here began with engineers enjoying an assignment, before the assignment became a business.
Six engineers, before the company
In Radev’s account, the beginning in 2019 was an e-commerce data platform and a team of six. Cloud infrastructure, real-time analytics and machine learning gave them absorbing problems to solve. Starting a company was not yet the point. Other businesses, it turned out, wanted similar help.
He co-founded Infinite Lambda with Márton Hubay, now its CTO. That origin gives the company a useful organising detail: the work came before the corporate story. Radev’s later recollection emphasises the pleasure of solving difficult problems together. The commercial opportunity followed the recognition that those problems kept appearing elsewhere.
By January 2020, his engineering advice already carried a little scar tissue. He opened an essay with an admission: I have failed, many times.
He credited better results to experience and capable colleagues. It is a refreshingly economical introduction to a field where even a minor software improvement can acquire the vocabulary of a moon landing.
His first recommendation was to check the data before releasing it. A newly assembled platform might appear to offer everything in one place, yet disagree with the reports its users already trusted. Filters, changing identifiers and mismatched records could undo the promise. The consequence was human: disappointed users returned to their old routines. A platform could be technically present and practically abandoned.
“I have failed, many times.”
Nas Radev, on learning to deliver data engineering
A company without one centre of gravity
Radev’s approach to building teams also questioned a familiar convenience. People who sit together are easy to put on the same project. He described organising teams around their skills and personalities instead, even when the company had offices available.
In December 2020, he reported that two-thirds of Infinite Lambda’s people were in smaller locations, naming Miskolc, Plovdiv, Bordeaux and Kosice among them. His argument was about access to a career: talented people should be able to practise demanding technical work without having to move to a major technology capital.
At that point, he put the company at nearly 70 people. He described its culture as something sustained through communication across locations, rather than something stored inside an office. For a founder, that is a fairly concrete claim about what management is for. The job includes making collaboration work for colleagues whose lives do not share a postcode.
The Slovak expansion put this thinking into everyday arrangements. In November 2021, Infinite Lambda described ten people spread between Kosice and Bratislava, with an office planned in Kosice. Employees could choose home, office or a mixture without formally reporting their location. The detail matters more than another declaration of flexibility: it describes a decision an employee could actually make on an ordinary morning.
The awkward arithmetic of selling time
Success brings its own engineering problem. Once a consultancy has learned to perform an assignment repeatedly, how much of that assignment should still require human labour? For Radev, this question now reaches into the way the company charges for its work.
In April 2026, he set out an argument against the billable hour. Automation creates an uncomfortable incentive for a business that earns money by selling professionals’ time. Completing a job sooner can reduce the amount of time available to sell. A useful improvement for the customer can become a difficult adjustment for the supplier.
His proposed direction is payment for outcomes, with accumulated delivery experience increasingly expressed through software. The expertise still matters; its commercial packaging changes. A client wants a functioning capability, and the calendar is usually an expense on the way there. Few customers have ever asked a migration to take longer for the pleasure of the timesheets.
Flowline, Infinite Lambda’s modernisation offering, sits inside this argument. Radev describes it as a way to automate much of the detailed migration work that previously occupied project teams. He calls the resulting model services-as-software. That is also a description of a founder revising the business he helped create: experience gained through people becomes part of the machinery that changes their assignments.
What belongs underneath the agent
Radev’s enterprise AI writing arranges the work into three layers. At the bottom is a usable data platform. In the middle is business context: agreed definitions and relationships. At the top are the agents that act on requests. The upper layer attracts attention because people can interact with it. The supporting layers tend to attract a meeting about next year’s budget.
He argues that each layer constrains the one above it. Fast, capable agents need systems they can query reliably, and the ability to retrieve information is only part of the job. A request about customers or revenue also requires a definition of those words that fits the business asking.
His ambition is to change how work moves through an organisation, so that a business request can produce a useful outcome without passing through long queues of coordinated tasks. This remains an argument about organisational design as much as model capability. Buying access to an agent does not, by itself, decide who owns a metric or who approves an action.
There is continuity with the engineer warning about validation in 2020. Both arguments end with somebody needing to trust an answer enough to use it. A brighter interface raises expectations. It also gives old inconsistencies a much more fluent spokesperson.
Twenty-two answers, one human meeting
The revenue example exposes the limit Radev places on automation. At the leisure operator, discovering conflicting definitions was work software could accelerate. Choosing an agreed definition required people with enough authority to make a decision and enough understanding to accept the trade-offs.
His account distinguishes the volume of technical investigation from the politics of agreement. Those are different demands on a team. Searching a large estate for competing calculations is laborious. Persuading departments to accept a shared rule involves responsibility, negotiation and judgement about people.
The final step, in his approach, is to put the agreed meaning into executable, governed logic. Otherwise the settlement risks becoming another document that colleagues politely acknowledge before returning to their own calculations. The difficulty offers a useful corrective to the idea of effortless automation: some of the work gets faster, and some becomes more visibly a management decision.
More countries, the same practical question
The company’s expansion has given these ideas a wider stage. In March 2025, Infinite Lambda announced that it was adding teams in the Americas, with initial colleagues in Chile, Argentina and Brazil. Radev emphasised the talent available there and the adoption of modern data tools.
The US launch brought Jim Ternyila in as regional general manager. Radev connected that move to demand for better infrastructure and teams able to use AI. The combination is characteristic of his account of the business: deliver a system, then help the people expected to depend on it.
In August 2026, Infinite Lambda announced Snowflake Partner of the Year for ASEAN, following its regional Rising Star recognition in 2025. It reported more than 250 data and AI projects and more than 100 Snowflake migrations. These are company achievements, built by teams. They also show how far the setting has widened since the six-engineer assignment.
Another development carried a different emphasis. Infinite Lambda announced B Corp certification in April 2026, following a review involving colleagues across the company. Radev’s response returned to the people sharing the work and the responsibilities surrounding it. Alongside the argument for automation sits an ambition about how a growing business should behave.
There are still people in the picture
A photograph on Radev’s 2019 running page supplies a welcome change of scenery: two people outdoors, rocks and water behind them, no architecture diagram in sight. He described preparing for his first half-marathon, with Mai alongside him to help make sure he finished. Even here, the undertaking had a teammate.
It would be excessive to turn a running photograph into a theory of leadership. His working choices offer enough evidence of their own: teams assembled across cities, a company built with Hubay, expansion led with regional colleagues, and software intended to take over more of the repetitive work.
The question that keeps returning is what people should do with the time and responsibility that remain. Radev’s 22 definitions make it wonderfully specific. Someone still has to sit down with the people who own those numbers, understand why they disagree, and leave with a decision they will use tomorrow.