
Our work on the Frontier go-to-market team brings us into regular conversations with many of the leading companies across the AI infrastructure stack, from compute providers and managed inference businesses to teams
Our work on the Frontier go-to-market team brings us into regular conversations with many of the leading companies across the AI infrastructure stack, from compute providers and managed inference businesses to teams building agents. As we deepen those relationships, we’ve been exploring where crypto can help these businesses.
I expected payments to be the clearest opportunity, given the potential for stablecoins to make payment flows faster, cheaper, or easier to manage. But instead, compute and inference providers were focused on growth and the fundamentals of running their businesses: securing capacity, getting deployments live, and managing margins. For teams building and deploying agents, reliability and control came up more often than payments did: What could an agent access, which actions was it authorized to take, and how could the company verify what happened?
Payments remain a promising opportunity for crypto x AI . But these conversations raised another possibility: Verifiability is an even more important use case for blockchains, at least in the near term. As companies give agents more responsibility, they need ways to establish who authorized an action, what limits applied, and whether that authority was still valid.
The more immediate opportunity for builders is to help enterprises delegate more work to agents while retaining control over what they can do. Crypto teams’ experience with signing and delegated authority gives them a way into that market.
Giving agents more responsibility
Giving agents more responsibility requires both reliable performance and clear limits on the agents’ authority. Permissions alone will not fix an agent that consistently gets the task wrong. But when an agent performs well enough to be useful, a company still needs to decide what it can access, which actions require approval, and how to revoke its access. This is where crypto teams’ experience in signing and delegating authority is relevant.
Before an enterprise expands an agent’s responsibilities, it needs to define what the agent can access, which actions it can take, and how those limits will be enforced — especially since the interfaces (and sense of user control) changes in this new world. That applies whether the agent is accessing company information for a consumer or changing a production system for a developer.
Consider a company using an agent from an external software provider to investigate a production incident. The agent can read logs and propose a fix, but needs approval before writing to and changing the system. The company grants that authority, the provider operates the agent, and the system receiving the request needs to verify that the action is permitted. But a January 2026 survey of 150 IT and security decision-makers revealed that 69% of respondents said security concerns were slowing agent adoption. Only 27% agreed that their existing identity systems were fully equipped to govern agents at scale. Leading concerns included data leakage and excessive access .
Research on deployed agents, meanwhile, points to how much work remains before companies grant them greater autonomy. In “ Measuring Agents in Production ” — a study of why organizations build agents, how they build them, how they evaluate them, and their top development challenges — reliability was the leading development challenge. Teams often limited agents to short workflows and kept humans involved along the way. Jellyfish’s recent engineering data similarly shows that agent involvement is growing, while workflows that take a task from specification to pull request without a human in the loop remain less common.
A coding agent offers another example: It might produce useful changes while still requiring a human to approve every deployment. The company needs to specify which systems it can change, which actions require review, and how to revoke its access. Better controls could let a company delegate more work to either agent.
Give companies the ability to delegate
The problem becomes harder when an agent works across companies. In the earlier production-incident example, the company grants authority, the provider operates the agent, and the receiving system must enforce the limits. A verifiable record of the agent’s permissions would give the company and provider a shared basis for checking whether an action was authorized.
Existing identity systems and application permissions already address parts of this problem. The opening for crypto technologies or companies is to make delegation easier to authorize, enforce, revoke, and audit across those systems — given these teams’ experience with signing, delegated authority, and user-controlled accounts.
Crypto teams are already applying that experience to a challenge enterprises face: letting software act without giving it unlimited authority. In Turnkey’s Wayfinder case study , Parallel Studios used session-based signing so users could authorize multiple transactions without signing each one individually. According to the case study, the approach preserved users’ custody of their wallets.
Wayfinder applies that capability to onchain transactions. What interests me — and the enterprise teams I’ve been speaking with — is letting software act without giving it unlimited authority. The bigger opportunity here may be to let agents carry out more work within permissions a company can enforce and revoke, with a record of what was authorized.
A blockchain may add value when independent parties need a shared record of permissions that no single party controls, or when the assets being controlled are already onchain. But the parties still need to agree on who can grant authority, what a permission allows, and how revocation works — which is what blockchains are built for.
For offchain applications, a company could record an agent’s permissions and their expiration onchain, along with any revocation. This registry could run alongside existing offchain identity and application systems, which would check the agent’s signed requests against it before allowing an action. The company and provider could both check what was authorized, while the receiving systems would remain responsible for enforcing those limits.
Many of these agents operate offchain, so reaching enterprise customers means any built-in controls work within the systems those customers already use. For crypto teams, that often means extending their products beyond onchain applications. To close this gap in go to market, these teams need to show how their experience can help a company give an agent more responsibility without giving the agent more access than it needs. The practical test is: Can a target customer put an agent to work on a task it previously couldn’t delegate?
Telling an enterprise it needs “trust its agents” gives them little reason to buy your product. Instead, builders should start selling with a task the company wants to delegate but can’t yet because of permissions or approval requirements. Then show how your product removes that obstacle. From there, it’s positioning why the improvement is worth adopting a new system — is it valuable enough, will it help the enterprise bottom line or enter new areas?
Who would pay for this?
Another big part of navigating enterprise go-to-market is selling to the right buyer:
For a company deploying agents internally , the buyer is likely the head of identity or security engineering, with the CISO as a potential sponsor. The problem they want to address could be excessive agent permissions, slow revocation, or investigations that require reconstructing activity across several systems. Build the pitch around a specific deployment your product could unblock and show how it would fit into the company’s existing security systems.
For a software company selling agents to enterprises , start with the CTO or agent-platform engineering leader. Ask how much customer-specific authorization work goes into each deployment and whether that work is delaying launches or consuming engineering bandwidth. Show how your product could meet those customer requirements with less custom work and speed up deployments.
In either case, founders need to learn who owns the current workaround, what it costs, and whether they are more likely to solve it themselves or would pay an existing vendor to handle it.
Existing identity vendors have an advantage because their systems are already in place. Even when a new product offers better controls, the target customer always weighs that benefit against the work of integrating another tool and maintaining it. (Some enterprise salespeople describe this internal inertia as like fighting gravity.) Founders need to understand shortfalls in the existing system and why the customer would choose their product over extending an existing system. The bar to clear isn’t just meeting necessary gaps, but adding value wherethe improvement clearly outweighs the effort of adoption. Unblocking an important agent rollout may justify substantial integration work; a smaller operational improvement needs to be easy to adopt. Start with a very specific workflow and take on as much of the integration work as possible yourself so the burden isn’t on the target company to figure it all out.
If those conversations reveal such an opportunity gap, start narrow and make the value concrete. Let an agent take a specific action under permissions the customer can enforce and revoke, and return a record to its existing audit tools. A pilot could test whether that saves them integration work or lets the customer delegate tasks that still require manual review, while keeping unauthorized actions blocked.
To evaluate the potential customer opportunity, you or your sales team need to ask:
What task does the customer want to delegate, and what prevents it today?
Who owns the workaround, and what does it cost?
What improvement would justify adopting a new system?
What would adopting this require from the customer, and how much of that work can you take on?
What does your approach make possible, or materially easier, that the customer’s existing tools don’t?
You should know the answer to these questions before you approach the customer, not after.
And then we can get to payments
New payment infrastructure only earns adoption when it removes friction from something a customer wants to do. Growth in agent adoption alone doesn’t tell us which workflows need new payment infrastructure. The exact requirements depend on the workflow:
An agent purchasing access to an unfamiliar API may need a way to pay for it immediately.
A coding agent operating within an enterprise’s existing subscriptions may encounter no payment friction at all.
A compute provider might benefit from a different settlement method while still having more urgent operational problems competing for its attention.
Improving payments becomes a priority when the current setup gets in the way of something a customer wants to do. An agent might need manual account setup before it can buy access to a service. A billing model might make small purchases impractical. Settlement might create costly operational work. Those are reasons to consider adopting a different payment system.
Payments also require controls. AP2 , for example, uses cryptographically verifiable mandates to record what a user authorized an agent to do, including purchases made when the user isn’t present. But a customer can need better controls without needing yet another new way to pay.
Rising transaction volumes alone don’t establish the need for new payment infrastructure unless you understand what customers are buying, and why existing payment options fall short.
***
As we enter the agentic economy, the question isn’t how to bring stablecoins into AI businesses. It’s what would let these companies feel safe giving agents more responsibility? Crypto and other frontier founders should spend time with teams trying to understand that gap, and address it. It’s not unlike selling anything new — from the frontier — into the enterprise: You always have to start where your customers are, not just where you want them to be.
***
The views expressed here are those of the individual AH Capital Management, L.L.C. (“a16z”) personnel quoted and are not the views of a16z or its affiliates. Certain information contained in here has been obtained from third-party sources, including from portfolio companies of funds managed by a16z. While taken from sources believed to be reliable, a16z has not independently verified such information and makes no representations about the current or enduring accuracy of the information or its appropriateness for a given situation. In addition, this content may include third-party advertisements; a16z has not reviewed such advertisements and does not endorse any advertising content contained therein.
This content is provided for informational purposes only, and should not be relied upon as legal, business, investment, or tax advice. You should consult your own advisers as to those matters. References to any securities or digital assets are for illustrative purposes only, and do not constitute an investment recommendation or offer to provide investment advisory services. Furthermore, this content is not directed at nor intended for use by any investors or prospective investors, and may not under any circumstances be relied upon when making a decision to invest in any fund managed by a16z. (An offering to invest in an a16z fund will be made only by the private placement memorandum, subscription agreement, and other relevant documentation of any such fund and should be read in their entirety.) Any investments or portfolio companies mentioned, referred to, or described are not representative of all investments in vehicles managed by a16z, and there can be no assurance that the investments will be profitable or that other investments made in the future will have similar characteristics or results. A list of investments made by funds managed by Andreessen Horowitz (excluding investments for which the issuer has not provided permission for a16z to disclose publicly as well as unannounced investments in publicly traded digital assets) is available at https://a16z.com/investments/.
The content speaks only as of the date indicated. Any projections, estimates, forecasts, targets, prospects, and/or opinions expressed in these materials are subject to change without notice and may differ or be contrary to opinions expressed by others. Please see https://a16z.com/disclosures for additional important information.