Enterprises need more than payments to put agents to work
a16z crypto examines how enterprises can safely expand AI agent responsibilities beyond payments. It finds verifiable permission controls, not just stablecoins, are the immediate priority for adoption.
النقاط الرئيسية
- Enterprises cite security (69%) as the main blocker to broader agent use.
- Only 27% believe current identity systems can govern agents at scale.
- Crypto’s signing and delegated‑authority tech can provide verifiable permission records.
- Builders should sell by solving a concrete delegation task, not by touting payments.
- Target buyers are identity/security leads for internal agents and CTOs for agent platforms.
- Security concerns slowing adoption
- 69%
- Identity systems ready to govern agents
- 27%
- Survey respondents
- 150
The question enterprises face
Enterprises are eager to let AI agents handle more work, but they worry about who authorises each action and how to revoke access. a16z crypto surveyed 150 IT and security decision‑makers in January 2026 and found that 69% say security concerns are slowing agent adoption, while only 27% feel their existing identity systems can govern agents at scale. The report asks: beyond faster payments, what blockchain feature can give companies confidence to delegate authority?
Data and method
The analysis draws on the survey, a study titled “Measuring Agents in Production,” and internal engineering data from Jellyfish and Turnkey’s Wayfinder case study. It looks at how agents are currently limited—short workflows, human‑in‑the‑loop approvals—and where cryptographic signing, session‑based delegation, and on‑chain permission registries could fill gaps.
Main findings
a16z crypto concludes that verifiability, not just stable‑coin payments, is the most pressing use case for blockchains in the near term. Companies need a shared, tamper‑proof record of who authorised an agent, what limits apply, and when revocation occurs. Crypto teams’ experience with delegated signing can enable session‑based authorisations that keep custody with the user while allowing software to act within defined bounds. The report highlights a concrete example: an external provider’s incident‑response agent reads logs, proposes fixes, and must obtain on‑chain‑verified approval before writing changes.