Dmitry Kachmar, Founder of Playgama – Interview Series

Dmitry Kachmar is Founder of Playgama, bringing a background in advertising technology and digital distribution to browser gaming. Before launching the company, he spent sixteen years at Yandex, where his work in advertising and commercial leadership shaped his approach to connecting products with audiences and building sustainable monetization. At Playgama, he focuses on the infrastructure developers need to test games, reach players, and learn from how people actually play.

Playgama provides distribution, monetization, and analytics tools for web game developers, including its open-source Bridge SDK and services for gaming portals and other platforms. Its AI Game Index tracks developer-reported AI tools alongside player responses to the games built with them. In this interview, Kachmar discusses what player behavior reveals about AI-built games, the limits of model comparisons, and why onboarding, retention, discovery, and creative judgment matter as game production becomes more accessible.

Your background spans adtech and classifieds, businesses where distribution and monetization are critical. What opportunity did you see in browser gaming that led you to found Playgama, and how did that experience shape the company's approach to helping developers reach players?

I spent sixteen years at a company called Yandex, one of the few players globally capable of beating Google in a major local market. I ended up running Ads as Chief Commercial Officer and CEO, managing over $2 billion in ad revenue. At that scale, you learn one thing: a good product without distribution is invisible, and distribution without monetization is a hobby.

When I looked at web games, I saw a market with real demand and real supply, but almost no functioning infrastructure in between. It’s like an iceberg: what’s visible is just a tiny slice of what’s actually there. Hundreds of millions of people play browser games every month, and thousands of developers make them. Yet three parts of it were completely broken.

First, monetization. Most web games ran ads with a very basic setup, so good games earned much less than they could. Second, discovery. Players found games through endless catalogs and luck, rather than algorithms tailored to their tastes. And advertising of the games themselves: if you had a good browser game, there was no normal way to buy players for it and see if they paid back.

So we started from the money and the traffic, not from the content. We built our own ad stack for web games, connected it directly to big demand, and used the same data to see which games hold players and deserve more of them. To me, these are two sides of the same problem: a game that earns well can afford to buy players, and a game that keeps players earns well.

What have you learned from watching people play AI-built games that you would never have discovered by evaluating the code alone? Can you share a game that surprised you, either positively or negatively?

The code tells you if the game runs and players tell you if it is a game. AI-built games almost never fail on code. They fail in the first 30 seconds, right in that window where you need to teach the rules without losing players’ attention. The game works and it looks nice, but the player simply closes the tab. You can’t see this in code review, because the developer already knows how to play and is interested by default.

There's another great case where a dev couldn't win his own game. The agent made the difficulty so high that even the author lost every time, and he only noticed it when he saw that nobody passed level two.

Your AI Game Index separates how widely developers use a tool from how players respond to its games. What explains the gap between developer adoption and player satisfaction, and what should a small studio take from that when choosing its tools?

I want to be careful here. Our index is a thermometer, not a judge. It shows what is happening and doesn't explain why.

In late October 2026, Claude Opus was behind more than half of all new AI games, while OpenAI games took more likes per play. Two things can be true at the same time.

I’m guessing, but I have no proof. The most popular model is probably used by everyone, including people making their first game ever, while a less popular model is a conscious choice by more motivated creators. So maybe we’re comparing people, not models. Also, when one model makes half of all games, its patterns become recognizable, and players get hungry for novelty.

The index relies on developer-reported model labels and games receiving different amounts of traffic. How do you distinguish a model's contribution from the developer's skill, the genre, or the audience a game attracts?

Honestly, we don't fully separate them, and we say so on the page. We basically just filter out the obvious noise. A game only counts once it hits 20 plays, and a model doesn’t get a rating until it’s logged 15 games. We use the median game, not the average, so one viral hit can't carry a whole row. Every number has a date, and small rows are marked as small.

Skill and genre are still inside the numbers. A model that attracts experienced developers will look better. A model that is popular for simple arcades will look different from one used for simulators. We can't remove this yet, so we don't hide it. Later we want to add breakdowns by genre and by developer experience. Until then, the best way to read the index is to see how real players interact with these games.

Where do AI-built games most often fall apart as games: controls, difficulty progression, level design, originality, or something else? Which of those weaknesses still requires the most intervention from an experienced designer?

Onboarding and difficulty matter. The player doesn't understand the goal in the first seconds, or the game is either too easy or impossible. Controls on mobile are the third one, because agents often build for a mouse and forget about a thumb.

The good news is that these problems are visible in data. If many players quit on the third tutorial step, an agent can see it in telemetry and simplify that step. Difficulty can be rebalanced the same way.

What still needs an experienced designer is the thing data can't tell you: what the game is about and why it should exist. Agents are good at fixing a game. They are not yet good at deciding which game is worth making. Originality, taste, the feeling of "I haven't seen this before" — this is still human work.

For a small studio, producing a playable prototype is only one step toward building a sustainable business. Where is AI materially improving the economics of web game development, and which costs remain stubbornly difficult to reduce?

AI made production almost free. The one thing that definitely didn't get cheaper is player attention. Real players, actual attention, good taste, none of that got cheaper. The cost of getting a game in front of people who actually want it is the same or even higher, because now there are many more games competing for the same evening.

So the economics moved. Before, the expensive part was making the game. Now the most expensive part is everything after. It’s testing on real people, distribution, retention, monetization. A small studio should spend its saved production budget there, not on making ten more prototypes.

As generating and publishing games becomes easier, how should platforms keep discovery useful for players? What standards does Playgama use to decide which games deserve visibility, particularly when polished-looking titles may offer very little depth?

You should make a decision after people play, not before. A pitch, a trailer or nice screenshots are very easy to make now. Ten minutes of real play time are not.

In our sandbox every new game gets a fair first test with real players. Then behavior decides: how long people stay, do they come back, do they like it. Games that hold players get shown to more people, and the best of them move to our main catalog and to external platforms. Roughly one in ten makes it.

Polished but empty games usually lose here by themselves. They get a good click, but people leave fast. Behavior is much harder to fake than looks. I don't think the answer is a jury or a "no AI" label. Players don't care how a game was made. They care if it's good, and their time is the most honest signal we have.

What makes a game work inside a messaging app, social network, or other non-gaming platform? How do you design for that audience without making the game feel like an interruption or a disposable engagement tactic?

On a non-gaming platform, the player didn't come to play. He came to chat or to watch videos. So the game has to respect that.

Three things work. First, instant start: no loading screen, no tutorial wall, the game is understandable in three seconds. Second, short sessions that can end at any moment without losing progress, because the person will be pulled back to the chat. Third, a reason to come back that feels like a habit, not a trap: a daily puzzle, a friend's score, a short streak.

What doesn't work is treating the player as traffic. If the game exists only to keep someone inside the app longer, people can smell the desperation.

Your adtech experience gives you a close view of the tension between advertising revenue and player enjoyment. How should web game developers decide when an ad placement is worth the interruption, and which metrics reveal when monetization is damaging long-term retention?

The rule is pretty simple, ads should only hit during a natural pause and never right in the middle of the action. Between levels, after a loss, when the player chooses to watch a rewarded video for a bonus. Rewarded ads are the healthiest format, because the player decides.

To see damage, don't look at revenue per session alone. Look at revenue together with retention. If you add an ad placement and revenue per day goes up, but day-1 and day-7 return drop, you are eating your future players. Also watch session length and where people quit: if many sessions end right after an ad, that ad is too aggressive or in the wrong place.

The best teams test ad changes like gameplay changes: one change, a clean comparison, look at retention after a week, not only at revenue the next day.

If AI makes competent game production widely accessible, where will an independent developer's competitive advantage come from? What would you advise a small team to invest in today to build a business that remains distinctive as the tools improve?

When everyone can make a competent game, competence stops being an advantage. Three things stay human.

Taste, because what is different from the default becomes valuable. A clear style, a strange idea, a game that feels like a person made it.

Speed of learning matters. Not speed of making, everyone has that now, but speed of finding out what you did wrong. Teams that put a version in front of real players every day and change it based on what they see will win against teams that build for months in silence.

And a relationship with your players. A community, a name people recognize, a reason for them to try your next game.

Focus on setting up the loop first. Your tools will keep changing every few months, but you don’t have to learn all of them. Listening to players is the one habit that never expires.

Thank you for the great interview, readers who wish to learn more about web game distribution, monetization, and AI-built games should visit Playgama.