AI Agent implementation encounters obstacles: controllable permissions are more urgent than payment.

CN
2 hours ago
When companies empower AI Agents, the most pressing concerns are controllable and verifiable permissions, not payments.

Written by: Christian

Translated by: AididiaoJP, Foresight News

AI Agents Stuck on Permissions: Opportunities Before the Cryptography Team

Our cryptography team has talked a lot with AI infrastructure companies, initially thinking payments would be the biggest entry point. Stablecoin settlements are faster, cheaper, and easier to manage, which seems obvious.

However, actual conversations have not been like that. Compute and inference providers are more concerned with growth and fundamentals: how to lock in capacity, how to get deployments running, and how to maintain gross margins. Teams developing Agents repeatedly mention reliability and control—what Agents can access, what actions must be pre-approved, and how companies verify what actually happened.

Payments are still a long-term direction for crypto and AI, but the clearer demand right now is verifiability. When companies delegate more work to Agents, they must be clear on who approved the operation, where the permission boundaries are, and whether the authorization is still valid. The cryptography team, with its long history of signatures and delegated authorizations, is well-positioned to address this.

Define Boundaries Before Delegating Authority

When delegating tasks to Agents, they must be reliable enough, and their permissions must be clearly limited. A permission system cannot solve the problem of Agents constantly making mistakes; however, once they perform well enough to be trusted, companies must decide which systems they can access, which actions require approval, and how to revoke permissions at any time.

This is especially important under new systems. Whether the Agent is helping users manage company information or assisting developers in modifying production systems, the company must first define the boundaries and ensure that enforcement is possible.

For example, a company uses an external software Agent to investigate production incidents. The Agent can read logs and suggest fixes, but approvals must be obtained before modifying systems. The company authorizes, the provider runs the Agent, and the receiving system checks permissions—a complete process that needs to be verifiable. A survey conducted in January 2026 among 150 IT and security decision-makers showed that 69% believe security concerns are slowing down Agent deployment, while only 27% think existing identity systems can effectively manage Agents at scale. Data breaches and excessive permissions are major worries.

Research on already deployed Agents points to the same issue. In "Measuring Agents in Production," reliability is the biggest development challenge. Teams often limit Agents to short workflows and retain human intervention. Jellyfish's engineering data also shows that while Agent participation is increasing, fully automated workflows—from demand through to pull requests—are still rare.

Coding an Agent is another example: it can produce useful changes, but each deployment still requires human approval. Companies must specify which systems it can modify, which actions require review, and how to revoke permissions. Better control means companies are more willing to delegate more work.

Cross-Company Collaboration Requires a Trusted Record

When Agents work across companies, the issues become more complex. Companies authorize, providers run the Agents, and the receiving systems enforce compliance—all three parties need a common basis to confirm "whether this operation has been authorized."

Existing identity systems and application permissions already cover part of the demand. The opportunity for cryptography lies in simplifying authorization, execution, revocation, and auditing across systems. The cryptography team's experience with signatures, delegated authorizations, and user-controlled accounts can be quite beneficial here.

Existing cases illustrate this point. In Turnkey's Wayfinder case, Parallel Studios used session-based signatures, allowing users to authorize multiple transactions at once while still retaining control of their wallets. Wayfinder itself targets on-chain transactions, but what is more noteworthy is empowering software to act without granting it unlimited permissions. The greater opportunity lies in enabling Agents to perform more tasks within scopes that can be executed and revoked by the company, while leaving verifiable records.

The value of blockchain is in having a shared permission record that no individual can control alone or the asset itself being on-chain. However, all parties must first agree on "who can authorize, what exactly permissions allow, and how to revoke"—this is precisely what blockchain excels at.

For off-chain applications, companies can store Agent permissions, expiration times, and revocation records on-chain. This registry can run parallel to existing off-chain identity and application systems, with the receiving system verifying signature requests before authorizing operations. Both the company and the provider can check the authorization details, while the receiving system is responsible for enforcement.

Most Agents currently operate off-chain. To enter enterprise clients, built-in controls must integrate with their existing systems. For the cryptography team, this means extending products beyond purely on-chain scenarios. The key test for entering the market is simple: can target customers delegate tasks previously stalled by permission or approval issues to the Agent?

Start Selling to Enterprises with Specific Tasks That Are Stuck

Directly telling companies "you need to trust your Agent" is unlikely to seal the deal. A better approach is to start from specific tasks they want to delegate but are blocked by permissions or approvals, demonstrating how your product can remove these obstacles, and then explaining whether this improvement is worth introducing a new system—whether it is valuable, and if it can enhance profits or open new business opportunities.

Buyers depend on the scenario.

When deploying Agents internally, decision-makers are often identity or security engineering heads, and the CISO might be the sponsor. Pain points typically include excessive permissions, slow revocation, or difficulty in recreating operations during cross-system investigations. When selling, focus on a specific deployment that unlocks something and describe how it can integrate into the existing security system.

Software companies selling Agents typically approach CTOs or engineering heads of Agent platforms. First, ask how much custom authorization each client needs for deployment, whether it slows down deployment or depletes engineering resources, then demonstrate how your product meets their needs with less customization and accelerates deployment.

In either case, founders must clarify: who is maintaining the current temporary solution, what the costs are, and whether they are more likely to solve it themselves or prefer to pay a vendor.

Existing identity vendors have a first-mover advantage since their systems are already in operation. Even if new products have better control, clients will weigh integration and maintenance costs—some salespeople describe this as battling against gravity. You must be clear about the shortcomings of the existing system and why clients would choose you over extending the old system. The threshold is not just filling gaps, but the improvements must clearly outweigh the costs of adoption. Unlocking significant Agent deployment may warrant significant integration efforts, while minor operational improvements must be very easy to implement.

Start with very specific workflows, and try to take on the integration work yourself, rather than offloading all the burdens to the client. If conversations confirm there is an opportunity, conduct a narrow pilot: allow the Agent to complete specific actions within permissions that the client can execute and revoke while returning records to existing auditing tools. Test whether it reduces integration work, or enables clients to delegate tasks that previously required manual review, while blocking unauthorized actions.

When assessing customer opportunities, clarify the following questions before discussing:

  • What tasks does the client want to delegate? What is currently blocking it?
  • Who is maintaining the temporary solution? What are the costs?
  • What kind of improvement is worth introducing a new system?
  • What does adoption require from the client? How much can you take on?
  • What does your solution make possible or significantly easier that existing tools cannot achieve?

These questions should be clear before engaging with clients, not afterward.

Payments Are Still Important, But Should Be Secondary to Control

New payment infrastructure will only be adopted when it removes the friction that clients actually want to address. The growth of Agent adoption itself does not tell you which processes need new payment solutions. Specific needs depend on the scenario:

  • Agents may need immediate payments when accessing unfamiliar APIs;
  • Coding Agents using existing enterprise subscriptions may not encounter any payment friction;
  • Compute providers may benefit from different settlement methods, but more urgent operational issues often take precedence.

Only when current setups block clients from doing what they want will payment improvements become a priority: needing manual account creation to purchase services, billing models making small purchases unattractive, and settlements resulting in high operational costs—these are the reasons for adopting new systems.

Payments also need control. For example, AP2 uses cryptographically verifiable authorizations to record what users allow Agents to do (including purchases made when users are absent). However, clients may only need better control without necessarily requiring yet another new payment method. Increased transaction volume alone cannot prove the need for new payment infrastructure unless you clearly understand what the client is purchasing and why existing options are inadequate.

Conclusion

Before entering the agency economy, the real question is not how to integrate stablecoins into AI businesses, but what will allow these companies to confidently delegate more responsibility to Agents. Cryptographers and other cutting-edge teams should spend time solving problems alongside teams filling this gap. This is like selling anything new to enterprises: it must start from where the customer currently is, not from where you hope they will arrive.

免责声明:本文章仅代表作者个人观点,不代表本平台的立场和观点。本文章仅供信息分享,不构成对任何人的任何投资建议。用户与作者之间的任何争议,与本平台无关。如网页中刊载的文章或图片涉及侵权,请提供相关的权利证明和身份证明发送邮件到support@aicoin.com,本平台相关工作人员将会进行核查。

Share To
APP

X

Telegram

Facebook

Reddit

CopyLink