Below, we outline Draw It’s mechanics, empirical outcomes, limitations, and potential futures, reflecting on its role in our broader vision for decentralized gaming.

The Mechanism

Draw It is a protocol for creative competition. Two players, allocated 60 seconds, draw on a web-based canvas in response to prompts like “Vitalik” or “buns”. Drawings are serialized as image data, minted as NFTs on a blockchain, and judged by an on-chain governor—an AI agent powered by a large language model (LLM). The LLM evaluates sketches for creativity, prompt fidelity, and clarity, assigning a “Winner” badge to the superior NFT. Both NFTs persist in a public ledger, accessible to all.

The game leverages our Pockit Challenge Protocol, a smart contract we designed to enable trustless, asynchronous gameplay.

Player 1 initiates a match by specifying a wager and governor address; Player 2 verifies these terms before joining. The governor resolves the match, selecting a winner and collecting a 2% arbitration fee. This asynchronous design allows players to participate across time zones, unbound by real-time constraints. The LLM, accessed via an API, processes image inputs with a directive to deliver fair, witty, and mildly humorous critiques. The system operates fully on-chain, embodying the decentralized ethos we pursue across our projects, from Pockit Milady’s social experiments to Draw It’s creative contests.

What Happened

Over the past few months, players engaged with prompts ranging from “Ethereum whale” to “degen buns,” playing over 500 rounds with 1000+ drawings. The Drawing Governor demonstrated competence in assessing visual inputs but struggled with crypto-specific terms. A sketch intended to depict Vitalik Buterin was parsed as “a wizard with distinctive facial hair,” and “buns” (Sanko’s on-chain pet mascot) was misconstrued as “a hamburger with lofty aspirations.” These errors underscore the LLM’s limitations in parsing subcultural context, a challenge we’ve noted in other AI-driven systems.

The Governor’s critiques, often incisive, sometimes veered into eccentricity. One spaceship sketch lost points for “excessive aubergine enthusiasm” due to overuse of certain phrases in the judgement critique. A cactus drawing sparked a narrative about a “solitary desert nomad,” which players found entertaining but tangential. These quirks drove lively discussions on our community Discord, where players tested strategies—some crafted detailed scenes, others embraced minimalism to exploit the AI’s biases.

The Pockit Challenge Protocol performed reliably, ensuring trustless execution of asynchronous matches. Gas fees were significantly low on Sanko’s L3 blockchain. The 2% arbitration fee proved sustainable, balancing operational costs with accessibility, consistent with our design goals for Pockit Milady and other projects.

Lessons and Trade-offs

Draw It yielded several insights.

First, LLMs can serve as autonomous arbiters for subjective tasks, adding personality through feedback, but require precise tuning for niche domains.

Second, the Pockit Challenge Protocol’s asynchronous mechanics and transparent ledger for round history foster trust, aligning with our commitment to credible neutrality.

Third, players value the Governor’s commentary, which transforms losses into engaging moments, hinting at the social potential of AI-mediated gameplay.

Limitations were evident. The LLM’s missteps on terms like “Vitalik” or “buns” highlight the need for domain-specific fine-tuning or context injection. Image-based LLMs struggled with low-fidelity or abstract inputs, as seen in the remilio-box confusion.

Speculative Extensions: Games as Adaptive Ecosystems

What intrigues us most about Draw It is its potential to model games as adaptive, community-driven ecosystems, a vision we’re pursuing across Pockit Game Corp’s projects. The Drawing Governor could curate the word pool by including emerging popular phrases from social media or community Discord discussions. By prioritizing engaging prompts and retiring less popular ones, it would align the game with player preferences. This forms a feedback loop: players’ drawings and conversations shape the Governor’s decisions, which in turn enhance future matches, creating a more compelling experience over time.

This loop could expand. The Governor might adjust scoring heuristics—perhaps emphasizing bold colors if players favor vibrant sketches—or propose new mechanics, like timed bonuses, via community channels. By retaining a memory of past interactions, it could refine its grasp of terms like “Vitalik,” reducing errors. Such adaptability points to a future we find compelling: games that evolve organically, driven by AI agents and player inputs, much like decentralized protocols evolve through governance. We see parallels in Pockit Milady’s social dynamics, where community interactions shape outcomes.

We’re optimistic about this path. If systems like Draw It can scale—through shared game infrastructure as smart contracts, robust LLMs, or insights from Pockit Milady—they could give rise to a vibrant ecosystem of decentralized, AI-mediated games. These would be more than pastimes; they’d be experiments in emergent coordination, blending creativity, technology, and community in ways that feel alive and open. Draw It is an early step, but it suggests a future where our protocols enable dynamic, player-driven worlds that thrive on the blockchain’s permissionless foundation.

To explore Draw It, visit draw.pockit.world.
Inspect the Pockit Challenge Protocol smart contract.