Agents. Agents everywhere.
We spend a lot of time thinking about where this is all going. Here are three bets we're willing to put our name on.
LLM Tokens will become a commodity
Models are improving at a rapid pace - but also all starting to product similar outputs. We believe eventually tokens will look a lot like electricity. Priced to the floor, undifferentiated, abundant. We are planning for that token-abundant future.
Context engineering, caching, routing, eval-driven prompt design — that is where the return on a token budget will come from. The moat is how you use them, not which ones you buy.
Agents 🤖 become first-class members of organizations
Today every agent and every workflow belongs to a person. A laptop, a login, a credit card. That is a temporary arrangement.
Soon agents will be hired by organizations - pre trained with all the skills, tools, and know how to get their tasks done. They will have seats on the org chart, budgets, charters, and policies for governance.. They will be evaluated the way employees are evaluated: on whether they get the work done.
AI 🤝 Blockchain. Finally a use case that makes sense.
Blockchain has always held promise - without much real utility. AI is it's killer usecase. For agents to be useful they need:
- Identity. Agents shouldn't manage passwords - or create emails for themselves - they identify via wallet identity.
- Payments. Agents need to transact at machine speed, per request, without API keys and monthly invoices. Open rails like x402 now make that possible.
- Digital rights management. Provenance for model output, licensing for training data, royalties that actually flow. A ledger is the right shape for this. Pricing models that enable value for value - rather than advertizing.
- Own your data Your data should be encrypted with keys you control, not keys someone else holds. Take back control of your data. Wallets are the cleanest primitive we have for that.
Orca, the Agent Registry, and the x402 gateway are the pieces we're building for the world these bets point to. If any of this reads true to you, we'd love to hear from you.