Dmitry Kushnikov joined Manychat as its first employee in 2016, became CTO and later COO, and today holds both roles. As the company grew to more than 400 people in 24 countries, he was at the center of each major change in how it works: he designed and then led the 2018 move to cross-functional teams under Large-Scale Scrum, has run the company’s OKR system since 2020, and, as COO and CTO, now leads its AI transformation. In this interview, he discusses how AI is reshaping hiring, productivity, and the role of managers.
You joined Manychat as its first employee in 2016 and went on to become CTO and COO, being critical in scaling the company to more than 400 employees. How has that experience shaped your approach to hiring and organizational growth, and how are AI agents changing the decisions you make about expanding teams?
It taught me that every stage needs a different company, not just a bigger one. Until autumn 2017 we all worked from home, with almost no system. Then we got an office, and in 2018 we moved to cross-functional teams. In 2020 COVID sent everyone home again, and later we came back as a hybrid company. In 2022 and 2023 we went from one city to a global company with several hubs. Each time, the way we worked had to change before the headcount did.
Every new role is a cost, and not only a financial one: it also costs everyone’s time. So before opening a role, the question is whether the work should exist at all, and whether it can be done differently: start, stop, continue, or automate. AI agents make the last option real for much more of the work, so the bar for opening a role is higher than it was.

In 2018, you led Manychat's transition from functional departments to cross-functional teams using Large-Scale Scrum, an experience you later shared at TeamLead Conf in 2020 and the LeSS Conference in 2024. What problem did this transformation solve, and why do you believe the model now needs to evolve for the AI era?
The problem was accountability. In our early years we worked in functional teams: backend, frontend, design, each doing its part. When something went wrong, it wasn’t clear who owned the result, and responsibility moved from one team to another. I optimize at the team level. A team should be able to act as an independent unit that nothing can stop from reaching its result. Then it’s clear what to hold it accountable for.
So in June 2018, at about 30 people, we decided to test Large-Scale Scrum and move to cross-functional teams, each owning a customer problem end to end. Later we grouped teams into product areas and built communities of practice, so engineers in different teams still share standards. I spoke about how this worked for us at TeamLead Conf in 2020 and at the LeSS Conference in 2024.
The idea we kept is that how work is organized is separate from who reports to whom. Our Product Group is a virtual organization, and our AI Lab is built the same way.
Now we’re rethinking it again. Most agile frameworks teach that less is more, and with AI agents a small team can now do far more than it could even two years ago. That changes how product and engineering should work together, and what worked in 2018 won’t fit as it is. We’re working on it, and we’d rather talk about it once it proves itself.
As a member of Manychat's executive team, you developed The Manychat Way, a set of ten operating principles published in 2026. How have these principles influenced management decisions, particularly the decision not to measure employees by their use of AI tools?
They change decisions more than they change what we believe. Nothing in The Manychat Way was invented in 2026. We’ve had five values for years, and the ten principles are their heritage and their interpretation: how those values work when a decision is hard. We wrote them as an executive team, starting from more than twenty drafts at an offsite in February, and published them in May.
The part that changes decisions is that each principle names what we’re willing to sacrifice for it. “We prioritize ruthlessly” comes with “let some fires burn.” That gives a manager the words to stop a decent project and put the team on a great one.
It has already made at least one company-wide call. When many companies started pushing people to use LLMs more and counting the usage, we went the other way: we don’t count it. We celebrate outcomes, not outputs, so LLM usage and tokens will never be a number we ask people to hit.
Having overseen technology and operations at Manychat as the company expanded, you've had to establish ways of measuring performance across a growing organization. How do you distinguish genuine productivity from activity, especially as employees adopt AI tools?
Productivity is about the value a team creates; the process is secondary. So we agree on the result before the work starts: every objective needs a clear answer to “how will we know we achieved it?”, and key results have to be obvious and measurable.
Then we’re careful with anything that’s easy to count. Goodhart’s law applies here: the moment a measure becomes the target, it stops measuring anything real. When we started our AI transformation, managers asked whether we would measure each person’s use of AI tools. We won’t, and I don’t remember us ever counting lines of code either.
Two more rules. We commit people only to what they can actually control. The big outcome numbers live at the level of the whole company, because when every team optimizes its own number, the system breaks. And we don’t celebrate output. It works like Tetris: a filled line disappears, and the bar resets.
You’ve led engineering teams at Manychat since its early years, eventually becoming CTO and COO. How has your approach to hiring senior technical talent evolved, and why do you believe the expectations for a senior engineer in 2027 will differ from those in 2024?
Five years ago I wrote to the whole company that recruitment, usually head-hunting, was the most important of my tasks, and that was literally true. Today recruiting has its own team and leaders. The belief behind it hasn’t changed: the main skill of a leader is hiring people, ideally stronger than yourself.
What changed is the bar. For senior roles we look for validated and reflected experience: enough time in one place to live through the consequences of your own decisions and learn from them. With people from big companies, the question is what luggage they bring and what they will leave outside. And before we hire a senior person, we have to be able to say clearly what they will own in their first six months. If we can’t, the interview isn’t fair to the candidate.
The AI era is moving the bar again. A senior engineer of 2027 works with AI agents as partners, and that is a completely different way of thinking: you have to choose it. The job moves from writing every line to owning the result end to end, from what should be built to whether it works. That takes product judgment: I don’t believe an engineer without a vision of the product can do it well. We’re still writing the new criteria down, but that’s the direction.
Customer support became part of your scope at the end of 2023, and AI agents now handle its first line. What prompted this shift, and what results have you seen?
At the end of 2023 customer support became part of my scope, and I became the executive overseeing it. In early 2024 it was the first place where our growth hit hard: tickets grew faster than the team could handle, and for a while people across the company were answering them.
My role was strategic: shaping where support should go given what was happening in the company, and setting the direction. In 2025 that direction was that the first line goes to an AI agent. Our Head of Customer Experience and the CX team built the operation: they chose the tool against their own test thresholds and rolled it out to all customers.
We also measure it in a way that’s fair to the team. The team is accountable for the quality of its own service, and we track separately the complaints that come from product issues or third-party platforms.
Monthly ticket volume is about twice what it was in early 2024, and customer satisfaction rose from about 84% at the start of 2024 to about 90% through 2025.
Manychat operates across 24 countries and 16 hours of time zones. As the executive responsible for technology strategy and the AI Lab, how have you approached AI adoption across this distributed organization? What measurable results have you achieved so far?
We treat AI adoption as pull, not push. Most of the real practice started with a handful of people in different teams and countries who found what works and shared it openly, and our job is to make it easy for that to spread. In sensitive areas like security, payments and data access, we keep a person in the loop.
We measure outcomes, not usage. Support is where we measured first: the measure is the cost to serve a resolved conversation, paired with customer satisfaction. Since the end of 2024 ticket volume has roughly doubled, while the frontline team grew by about a third.
Outside support we’re still finding the right measure for each function, and until we have one we’d rather not count activity.
After a decade building Manychat, you've described the manager of the future as a "visionary operator." How do you expect AI to transform management and organizational structures over the next three to five years, and how are you preparing Manychat for those changes?
Managers’ work will change the most. AI agents will take over more of the routine, and what stays with people is judgment, vision and responsibility for other people. The profile we want to grow is the visionary operator: someone who can see where we’re going and also make it happen. AI agents help close the gaps on both sides. They complement people; they don’t replace them.
The way we prepare starts with one position: if you want to lead a transformation, start with yourself, not with cutting your people. A company only changes when each of us does, so leaders go first; I worked it through on myself before asking anyone else to.
Then we give everyone access to the tools, with budgets and licenses for the whole company, and we invest in training. And we’re changing how the company works with information: everything that can live as a policy or a process is written down and kept in a repository, so people and AI agents work from the same source.
You've risen from Manychat's first employee to CTO and COO, helping build an international technology company. What advice would you give to other engineers and technology leaders who aspire to achieve similar success?
Work on self-awareness before anything else. It’s a general leadership skill more than a people-management one: you can’t help your team until you understand where you are yourself, and people copy behavior, not instructions.
Then expect the bar to keep moving. When a company grows as fast as ours, staying at the same level means falling behind: you need to exceed to meet. Every stage asked for something I hadn’t done before, and the hardest part was usually letting go of what I was best at.
