TechFlow Logo
Login/ Sign up
ETH Gas
Gwei
Fear
gas
If enterprises want AI agents to truly go live, they need far more than just "crypto payments".

If enterprises want AI agents to truly go live, they need far more than just "crypto payments".

2026.11.18
Share
a16z Frontier GTM Lead Engages Directly with Enterprise Frontlines: Compared to stablecoin payments, verifiability, delegated authorization, and permission auditing remain the core bottlenecks to scaling Agent deployments.

Author: Christian Crowley, a16z Frontier GTM

Translated by: TechFlow

TechFlow Summary: In the narrative of Crypto and AI, the industry generally views stablecoin payments as the clearest intersection. However, through in-depth discussions with numerous compute, inference, and Agent companies, the a16z Frontier team has found that enterprises' most pressing pain point is not how to spend funds, but how to control permissions.

When Agents are granted higher autonomy, data leakage risks and over-authorization become the primary concerns hindering their adoption. The deep accumulation of crypto-native technologies in multi-signature schemes, delegated authorization, and immutable audit trails provides exactly the key solution to break this enterprise trust deadlock.

In our daily work within the a16z Frontier Go-To-Market (GTM) team, we frequently engage in deep discussions with leading full-stack players in the AI infrastructure space, including compute providers, managed inference service providers, and startups actively building Agents at the forefront. As we deepen these partnerships, we have been exploring exactly which key nodes crypto technology can empower these companies.

I originally expected "payments" to be the clearest breakthrough point, given the inherent potential of stablecoins to make capital flows faster, cheaper, and easier to manage. Surprisingly, however, compute and inference service providers are currently entirely focused on business growth and core operations: securing compute allocations, pushing deployments online, and strictly controlling profit margins.

For teams building and deploying Agents, "reliability" and "controllability" are mentioned far more frequently than payment issues: Which internal data can the Agent access? What actions is it authorized to perform? And how can enterprises trace and verify exactly what happened?

Payments remain a highly promising direction for Crypto × AI, but these frontline conversations reveal another more fundamental possibility: At least in the short term, "Verifiability" is an even more critical application scenario for blockchain technology. As enterprises grant Agents more responsibilities, they urgently need a clear mechanism to establish: who authorized a specific operation? What boundary constraints apply? And whether the current authorization remains valid?

For startups, the more pressing commercial opportunity right now lies in helping enterprises firmly maintain control over behavioral boundaries as they delegate more tasks to Agents. The crypto industry's profound experience accumulated in cryptographic signatures and delegated authorization happens to provide a clear path into this enterprise-level market.

Granting Agents More Autonomy: A Dual Test of Reliability and Permission Boundaries

Granting Agents more responsibilities requires both stable execution performance and clearly defined permission boundaries. Solely relying on permission controls cannot save an error-prone Agent; but when an Agent possesses sufficient capability to complete tasks, enterprises must still decide: which systems can it access? Which actions require manual approval? And how to revoke its access in emergencies? This is precisely where crypto teams excel in signatures and permission delegation.

Before enterprises expand an Agent's scope of duties, they must clearly define its access privileges, specify allowed actions, and establish how to enforce these restrictions. This principle holds true regardless of whether the Agent is retrieving internal data on behalf of ordinary consumers or modifying production systems on behalf of engineers.

Consider this scenario: A company introduces an external software provider's Agent to troubleshoot live production incidents. The Agent has permissions to read logs and propose fixes, but must obtain manual approval before writing to or executing changes in the production system. The company grants the permissions, the provider runs the Agent, and the receiving target system must verify whether the operation complies with regulations.

However, a January 2026 survey of 150 IT and security decision-makers revealed: 69% of respondents stated that security concerns are slowing down Agent deployment; only 27% believe their existing identity and access management systems are sufficient to govern Agents at scale; among these, data leaks and over-authorization are listed as the biggest hazards.

At the same time, research on Agent performance in production environments also indicates that significant groundwork remains to be done before enterprises truly hand over control. In a recent study on Agent production deployments, "reliability" was ranked as the top engineering challenge. Most teams were forced to restrict Agents to brief workflows and maintain high-frequency human intervention. Engineering analytics data further shows that although Agent engagement is increasing, workflows where Agents independently convert requirements into code and submit merges without human intervention remain extremely rare.

Code Agents serve as another classic example: they may produce high-quality code modifications, but enterprises still require every release deployment to be reviewed by engineers. Companies must strictly define which code repositories it can modify, which actions require secondary verification, and how to immediately cut off its permissions when necessary. More robust control mechanisms are precisely the key premise for unlocking deep enterprise delegation.

Enabling Enterprises with Cross-System Secure Delegation Capabilities

When Agents need to collaborate across enterprise boundaries, challenges multiply exponentially. In the previously mentioned production troubleshooting case, the enterprise handles authorization, the external vendor runs the Agent, and the receiving system must enforce restrictions. A verifiable permission record provides a common trust foundation for both the enterprise and the external vendor to verify authorizations.

Existing traditional identity management systems and application permissions can already cover part of the problem. But the unique breakthrough for crypto technology lies in leveraging technological deposits in digital signatures, delegated authorization, and self-custody account systems, making cross-heterogeneous-system authorization, execution, revocation, and auditing unprecedentedly lightweight and standardized.

Crypto teams have already begun translating this experience into tangible product solutions. For instance, in Turnkey's case study, Parallel Studios adopted a session-based dynamic signing mechanism, allowing users to safely authorize multiple interactions without manual confirmation per transaction, while absolutely maintaining ultimate control over the underlying account.

In on-chain scenarios, this capability has been thoroughly validated. But what truly excites me and many enterprise teams is how to achieve "allowing software to act autonomously without granting infinite permissions" in off-chain traditional systems. The larger market opportunity lies in letting Agents confidently handle more complex business flows within verifiable permissions that enterprises can execute, revoke, and audit at any time.

When multiple independent entities require a shared permission record immune to unilateral monopolization, blockchain can provide a highly credible foundation. All parties can still reach tamper-proof consensus on foundational rules such as who has the authority to authorize, the specific scope of single permissions, and revocation mechanisms.

In off-chain enterprise applications, companies can anchor an Agent's authorization records, validity periods, and revocation statuses to a lightweight ledger. This credential registry can run parallel to existing identity systems, with the system verifying the signature credentials carried with the Agent's request before approving it. Both enterprises and service providers can verify authorization details, while the endpoint business system focuses on enforcing underlying boundaries.

Many cutting-edge Agents essentially operate in purely off-chain environments, meaning reaching enterprise clients requires built-in control mechanisms to seamlessly integrate into clients' existing tech stacks. For crypto teams, this means stepping out of the pure on-chain comfort zone. In business development, teams must prove to clients: your technology enables enterprises to safely empower Agents without overly exposing excess permissions. The metric for success is very practical: Are clients willing to formally hand over tasks they previously hesitated to delegate due to approval concerns to the Agents?

Selling simply to enterprises that "you should trust your Agent" lacks commercial persuasiveness. Instead, entrepreneurs should cut into the specific pain points enterprises urgently want to delegate but are stuck behind permission approval thresholds, proving how your tools clear these obstacles, thereby demonstrating the actual bottom-line value of cost reduction and efficiency gains brought by the new system.

Who Will Pay for This Permission System?

In the enterprise GTM path, accurately identifying the true buyer decision-maker is crucial:

If the target client is an enterprise developing and deploying Agents internally, the buyers are typically heads of Identity Engineering or Security Engineering, with the Chief Information Security Officer (CISO) as the core sponsor. Their pressing pain points often involve: excessive sprawl of Agent permissions, slow permission revocation, or having to piece together audit trails across multiple siloed systems during incident investigations. Your product pitch should tightly revolve around unlocking a specific blocked workflow and prove it integrates seamlessly into the enterprise's existing security framework.

If the target client is a software company delivering Agent solutions to enterprise customers, the entry point is their Chief Technology Officer (CTO) or Agent Platform Engineering Lead. You might ask directly: How much custom engineering has your team invested to meet personalized authorization requirements for different enterprise clients? Is this delaying delivery schedules and consuming core R&D bandwidth? Then show them how your product replaces heavy custom development with standardized solutions, significantly accelerating commercial deployment.

Regardless of the client type, founders must assess the current situation: Who is responsible for the stopgap alternatives clients currently use? How high are the maintenance costs? Do they prefer building in-house or are they willing to purchase mature services from external vendors?

Traditional identity software giants possess natural incumbent advantages, as their systems are deeply embedded in enterprise foundations. Even if new products offer more granular control, clients will inevitably weigh and reconsider "the benefits of new features" against "the costs of integrating and maintaining brand-new tools" (many seasoned enterprise salespeople describe this internal inertia as fighting gravity).

Therefore, a new product's value must overwhelmingly surpass its implementation cost. If your solution unlocks a strategic Agent workflow launch for the company, introducing the new tool becomes naturally easier. The smart move is to target an extremely clear vertical workflow, with your team taking on as much integration dirty work as possible, minimizing the trial-and-error burden on the client team.

When approaching potential clients, the sales team must have clear answers to five core questions beforehand: First, what specific task does the client want to delegate, and what barriers currently constrain it? Second, who is responsible for existing alternatives, and how much sunk cost has been incurred? Third, to what extent must efficiency improve to justify adopting a new system? Fourth, what collaboration costs does adopting the solution require from the client, and how much can you shoulder? Fifth, what critical breakthrough has your solution achieved that existing tools absolutely cannot?

Reimagining Payments: Why Payments Must Yield to Control

New payment infrastructure only wins organic adoption when it genuinely eliminates core business friction for clients. Simply observing rising Agent adoption rates does not logically imply clients urgently need new payment systems; the true demand depends entirely on specific workflow attributes:

When an Agent needs to autonomously call previously uncooperative paid APIs across institutions, it may require a lightweight payment channel capable of instant settlement;

However, when a Code Agent operates within an enterprise's existing monthly subscription software ecosystem, it experiences zero payment friction;

Compute providers might benefit from more efficient settlement cycles, but their attention is currently occupied by more urgent operational contradictions like data center expansion and compute delivery.

Only when existing payment mechanisms truly hinder business operations does retrofitting payments rise to a priority. Examples include: agents requiring cumbersome manual account opening before purchasing services, billing models making micro-purchases financially unviable, or settlement processes incurring high reconciliation costs. These are the genuine drivers compelling clients to embrace new payment tools.

At the same time, payments themselves inherently rely on strict control. For instance, in the Agent Payment Protocol (AP2), it relies on cryptographically verifiable mandates to record the scope of operations a user authorizes an Agent to execute, including autonomous purchasing permissions during offline states. This again reinforces: the demand for stronger control often precedes the demand for completely new payment methods.

Mere growth in transaction volume alone is insufficient justification for building an entirely new payment rails, unless you have thoroughly understood exactly what clients are procuring and precisely why existing commercial payment methods fail.

Conclusion: Start from Clients' Current Pain Points

As the Agentic Economy fully unfolds, the core proposition facing the industry is undoubtedly not "how to forcefully stuff stablecoins into AI enterprises' ledgers," but "how to enable these enterprises to confidently assign more substantive operational responsibilities to Agents."

Entrepreneurs operating at the crypto frontier or cutting-edge AI tracks should take time to engage deeply with enterprise teams, genuinely understanding and bridging this trust gap. The logic for introducing any disruptive frontier technology to enterprises has never changed: you must always start from the client's real-world predicaments they currently face, rather than unilaterally projecting from the position you wish them to reach.