The problem Sobek solves

Crypto needs to solve two UX challenges in order to bring the next 1B people on chain.

Consumer onboarding fails because the interface demands fluency before you can do anything useful. Voice removes that. Consumers can describe what they want, the agent executes.

Institutional adoption fails because on-chain counterparty risk is opaque in a way off-chain risk isn't. Institutions aren't risk-averse, they're opacity-averse.

Sobek makes counterparty risk more legible on-chain than off. Reputation scores are computed from transaction volume, deal size, and success rate. They are weighted to reward consistent high-value activity and resist Sybil attacks by design.

Challenges I ran into

The hardest technical challenge was coordinating four different chains. I had a really ambitious scope for this, and wanted to do something crosschain.

I used Hedera, Base mainnet, Sepolia Base (after I realized ERC-8004 wasn't on mainnet yet), and ADI each with different SDKs, different tooling maturity, and different mental models for how transactions work.

The initial approach was to build abstractions that smoothed over the differences. That created more problems than it solved. I caught early that it was on a path to become a complexity nightmare.

The fix was to reduce moving parts. Each chain owns exactly one responsibility with hard boundaries between them. Hedera handles trust and audit trail. Sepolia Base handles tokenized ERC-8004 identity. Base handles payments. Once we stopped fighting the complexity and just contained it, things stabilized.

Use of AI tools and agents

Sobek is designed to be agentic from the ground up. For humans, a voice agent makes transactions and token swaps easy. For agents, exposing an agent first API makes actions effective.

For both, a sophisticated reputational system built on top of ERC 8004 makes is it easier to tell counter party risk on-chain than off-chain.