Playable Behavior, unity, and an agent that plays genshin impact

Yo, 

How are things? Greetings from sunny Tel Aviv. Hope you’re getting ready to end the year on a high note, and preparing for a massive 2026 for your games. 

If you felt off these past few weeks don’t be hard on yourself, Mercury has been in retrograde for most of November. If you’re looking to blame others for feeling off, what’s better than blaming a far away planet? 

But Mercury aside, lots to cover this week - so let’s dive in. 

Supersonic, Sett & Gamigion Webinar on the UA skills to master in the AI era

Gamigion's Ömer Yakabagi is hosting Karen Levy, Director of Creative Growth from Supersonic by Unity and myself for a cool conversation about the skills UA people need to build in the AI era. 

We’re going to talk about what changed, what the usage of AI in user acquisition taught us so far, and what to do about it to kill it in your career in 2026 and beyond. 

Join us LIVE on December 9th by signing up at this link. If you can’t make it don’t worry about it - if you sign up we will send you the recording and the notes afterwards. 

7 killer tips to improve your playables based on how people actually play them

What do I love more than data? My daughter, chess, writing, and space. But besides those things, I love data the most. 

So I was thrilled to sit down and analyze more than 3.5 million play sessions within playables and start dissecting the data to get closer to the truth of how users engage with playables, and what separates great playables from bad ones. The men from the boys. The women from the girls. 

Some things I found interesting: 

🔥 the impact of the first time user experience of a playable is x2 on engagement, duration and drop-off
🔥 getting users to interact is crucial but the quality of the core game loop has a massive impact on getting people to continue to engage and eventually install 
🔥 when users do interact and start playing, they play for a long time 
🔥 imbalanced difficulty can quickly drive a drop off after players have engaged for a long time, and redirect rules at the wrong moments can kill a playable’s potential

👉 Read all about it in 7 killer tips to improve playable ad performance 

New podcast episode: Shaun Rutland ex-CEO of Hutch Games shares his wisdom on games, distribution & building a new studio 

Sett’s CEO, Amit Carmi sat down with Shaun Rutland, who was the CEO of Hutch Games and is now building a new and exciting studio. 

The big picture: Shaun Rutland co-founded Hutch, scaled it from zero to exit, and is now building a lean AI-first studio. His playbook: stay small, stay fast, stay close to the work.

Why it matters: The next wave of winning studios won't scale like the old ones. They'll optimize for clarity and speed, not headcount.

The path wasn't linear. Shaun left school at 16, ran a pirate game business in New Zealand, delivered mail at Royal Mail, then spent five years at PlayStation watching three games get canceled because decisions happened in email threads, not in rooms with the people doing the work.

That lesson shaped Hutch: make decisions with the people building the thing.
 
How Hutch found its edge:

 👉Started with audience, not genre. Racing games were underperforming on mobile, but car culture was massive. They built for identity and community, not mechanics.
 👉 First game made $1M with zero marketing. They reinvested everything.
 👉 Top Drives and F1 Clash followed. Built for people who care deeply about cars, not people who might download a racing game.

On AI:

- Thinks in value, not departments. Debugging, data access, handoffs between teams. That's where friction lives. That's where AI clears the path.
- Building his own tools now. Never coded before. Believes everyone should be able to touch the work.
- Calls it "permissionless culture." Titles don't limit contribution. Small teams, fewer boundaries, more ownership.

The bottom line: The studios that win next won't be the biggest. They'll be the ones that are ready to change the most. 

👉 Listen to the full episode here, or wherever you listen to podcasts.

Unite 2025 Revealed the rise of playables on Unity 

Unity gathered for their annual Unite conference in Barcelona last week. One thing that caught my attention was a talk by Rubin Kuyer, who manages business operations on the playable side. 

Here are a few takeaways from the talk: 

The number of impressions playables get on unity grew by x3 in the past two years to 10B impressions a month. 

Based on Unity’s data the most important factors are getting users into the fun in 1-3 seconds (ahm ahm look at our data as well!), convert better than other formats and often see IPMs of 40+. 

Most importantly from what Unity sees, great playables put an emphasis on the actual game & level - be it difficulty, progression, reward and design for long play sessions.

As we saw with our data, getting playables right and enjoying the benefit to your UA performance goes into drilling down on making the game experience itself better.

This goes against “easy hacks” of using too simple playables (interactive end cards) that are basically just a video with a button - or using rigid templates that basically don’t give you any control over the above factors that dictate the performance of the playable itself. 

The best winning playables win because they’re fun. To create fun you need complete control over the game design of your playable, not a template you can swap colors, some assets, and backgrounds.  

👉 You can watch the full talk here.

Really cool progress on AI that understands games - Bytedance teaches AI to play Genshin Impact

Last but not least - if you’re not spending your days reading essays about AI research like our great AI researchers, that’s okay. We got you. 

One area of research that is crucial in building true agentic AI workflows to develop fun game experiences (such as the agentic playable generation capability we’re building at Sett) is developing agents that can freely interact with games, understand them, and play them as a human would. 

This allows AI to much better understand the “fun” which is a crucial component in developing valuable AI systems for mobile game studios. 

One research from last week was from the Bytedance Seed research team that published a paper showing how they were able to train an AI model to understand the world of Genshin Impact, follow instructions, reason within the game and then play successfully and complete missions or levels - in a similar way to how a human would play. 

The cool thing about this model IMO is how it was trained based on raw human gameplay footage and was able to generalize things such as combat, mission understanding, boss fights, complex navigation, puzzle solving. 

Think about how rare it is for game studios to create a fun game - and the future in which game studios will be able to have systems that develop fun experiences by allowing the AI that would build this gameplay to also understand how real players would play the game, where they would get stuck, bored, or satisfied. 

👉 You can check out how it plays Genshin right here.

A cool future ahead of us all! 

See ya next time, 

Jonathan