We've been building UniversalAI for the past year — essentially trying to replicate the operational backend of a record label using AI agents and blockchain infrastructure. The problem we're solving: Independent music creators have more distribution access than ever (Spotify, YouTube, etc.), but they lack the operational infrastructure that labels provide: marketing automation, treasury management, catalog organization, revenue tracking across multiple streams, and IP protection. Most artists end up duct-taping together 10+ SaaS tools that don't talk to each other, or they sign deals primarily to access infrastructure — not because they want a partner. What we built:

AI Agent System: Specialized agents for marketing (drop scheduling, social automation), treasury (revenue tracking, split management), and catalog management. Built on ElizaOS with a custom Agent TCP protocol for inter-agent communication. Agents use pgvector-backed knowledge stores to learn creator preferences over time. WZRD.Studio: AI-powered video production pipeline. Text-to-video, AI storyboarding with style consistency across frames, timeline editing, and one-click export with platform-specific formatting. We're using a multi-model architecture rather than relying on a single foundation model. On-chain IP layer: Solana for ownership verification, Crossmint for wallet infrastructure (so creators don't need to understand crypto), decentralized storage for metadata. The goal is tamper-proof provenance without the Web3 learning curve. Treasury dashboard: Unified view of revenue across streaming, merch, NFTs, sync licensing, with split tracking for collaborators.

What's working:

The agent orchestration layer is surprisingly effective for repetitive ops tasks. Treasury tracking has been the most immediately useful feature for beta users — turns out most creators have no idea how much they're actually making across platforms until they see it in one place.

What's still hard:

Style consistency in AI video generation across longer projects is improving but not solved Agent "personality" calibration — getting agents to match a creator's voice/brand takes iteration On-chain verification is seamless for new work, but ingesting existing catalogs with provenance is messier