CryptoFlix

I think the most interesting power dynamics are the ones that look like infrastructure until you trace where the decisions actually originate.
I spent time building recommendation systems for a consumer platform about three years ago. The system was framed internally as a tool for helping users discover content they would enjoy. Technically accurate. The system was also the mechanism that determined which content creators received visibility, which topics trended on the platform, and which user behaviors got reinforced through exposure. The users experienced it as personalization. The creators experienced it as an opaque gatekeeper. The platform experienced it as a growth lever.
All three framings were simultaneously correct. The infrastructure was neutral in its architecture and directional in its effects. Who it served most depended entirely on whose objectives had been encoded into the optimization function.
I thought about that recommendation system when I read through the technical documentation for Stacked's AI-powered offer engine. Because the framing as player-friendly personalization infrastructure is accurate, and the framing as an AI system making economic decisions about who receives what value and when is equally accurate. And the second framing is getting considerably less attention than the first.
What Stacked's AI is actually doing:
The Stacked documentation describes an AI-driven offer engine that tracks granular player events in real time and deploys personalized incentives based on behavior. Studio operators query the system in plain language to identify churn reasons, find where reward budgets are leaking, and deploy targeted re-engagement campaigns.
The internal campaign data is the most specific evidence of what this means in practice. Veteran players who had not spent in over 30 days were targeted with personalized re-engagement offers. The results: 178 percent lift in conversion to spend, 129 percent increase in active days, 131 percent return on reward spend.
Those numbers are real and they are impressive. They are also describing a system that identified a specific player behavioral profile, determined that a personalized incentive would convert that profile back to spending behavior, deployed the incentive, and measured the outcome. The players received what felt like a personalized offer. The system received behavioral confirmation that its targeting model had correctly predicted their response.
That confirmation becomes training data. The model gets better at predicting which players will respond to which incentives under which conditions. Over time, the system develops an increasingly precise map of how to convert specific behavioral profiles back to economic activity.
What bugs me:
The AI economist framing in Pixels' own communications is unusually honest about what the system is doing. Luke Barwikowski explicitly described Stacked as an embedded AI game economist. That description is accurate. It is also describing something that operates differently from how most players understand the reward systems they interact with.
A traditional reward system has visible rules. Complete this quest, receive this reward. The relationship between behavior and outcome is legible to the player. They can optimize, plan, and understand why they received what they received.
An AI-driven offer engine has optimized targeting. The player receives a personalized offer at a moment the system has determined is most likely to convert them. The player experiences this as a helpful reward. The system is executing a conversion campaign against a segmented behavioral profile.
Those two experiences feel identical from the inside. Their underlying nature is different in ways that matter for how players understand their relationship to the ecosystem.
When Stacked scales to external B2B studios, the AI economist embedded in those games will be optimizing for the studio's RORS, which the Stacked system guarantees will be positive before USDC rewards unlock. The players in those games will experience personalized engagement. The system will be running economic optimization campaigns against behavioral profiles those players did not knowingly consent to be part of.
My concern though:
The Stacked re-engagement campaign targeting veterans who had not spent in 30 days achieved a 131 percent RORS. That means the value generated by converting those players back to spending exceeded the cost of the rewards used to convert them.
What that metric does not capture is what happens at the 31st day when the re-engagement offer has been consumed and the player faces the same conditions that caused them to stop spending in the first place. A 131 percent RORS on a re-engagement campaign measures the conversion. It does not measure whether the underlying reason for disengagement was addressed or merely delayed by a well-timed incentive.
A system optimized for RORS can generate excellent short-term conversion metrics while systematically applying high-precision incentives to behavioral profiles at their most persuadable moments without addressing the underlying product or economic conditions that create disengagement in the first place.
The difference between fixing churn and profitably exploiting the moment before churn is measurable in the long-term retention data. RORS captures one. Cohort retention curves capture the other.
Still figuring out:
The recommendition system I built eventually received a mandate to optimize for long-term user satisfaction rather than short-term engagement. The metrics got harder to move. The outcomes got better for users. The two things were in tension in ways that took leadership commitment to hold.
Stacked's AI economist is currently optimizing for RORS. That is the right metric to get off the ground and prove the model works. The question that matters as the system scales is whether the optimization function will expand to include the behavioral outcomes that matter for players over longer time horizons or whether RORS remains the primary signal that shapes how the AI allocates value across the ecosystem.
An AI economist that optimizes only for return on reward spend is making decisions about who receives economic attention based entirely on who generates the best short-term return. That is a legitimate business objective. It is not the same as an AI economist that optimizes for ecosystem health over time.
The distinction is invisible in the current metrics. It shows up in whether the players the system re-engages at 131 percent RORS are still active six months later or whether they represent a profitable conversion that preceded a permanent exit.
Honestly still figuring out whether Stacked is building a game economy that gets smarter about serving players over time, or a targeting system that gets smarter about converting players at the optimal moment regardless of whether the conversion serves them.

$PIXEL @Pixels #pixel