a16z: The Hardest-to-Use Enterprise Software Is AI’s Biggest Opportunity
a16z: The Hardest-to-Use Enterprise Software Is AI’s Biggest Opportunity
When most people talk about AI in crypto, they focus on net-new experiences: smarter trading terminals, better wallet assistants, faster security tooling, or “agentic” DeFi strategies.
But the biggest near-term opportunity may be far less glamorous: using AI to make the existing enterprise software stack (SAP, Salesforce, ServiceNow) finally usable—and in doing so, making blockchain deployments actually operable inside real companies.
a16z recently made this case directly in Why the World Still Runs on SAP: systems of record persist not because they are loved, but because they hold the canonical data model, approvals, permissions, and “institutional memory” that keeps a business compliant and running. AI’s wedge is not replacing these systems overnight, but wrapping them with a programmable interface that turns messy human intent into correct, auditable action.
In crypto, that interface layer is exactly what’s been missing.
The enterprise truth: crypto adoption is an ERP problem, not a chain problem
By 2025, the crypto market’s center of gravity continued shifting from speculative experimentation toward enterprise rails: stablecoin-based settlement, tokenized cash management, and real-world asset tokenization.
- McKinsey described 2025 as a potential inflection point for tokenized cash and stablecoins in payments and treasury workflows in The stable door opens: how tokenized cash enables next-gen payments.
- The IMF has been quantifying how stablecoins are used across borders—and what that implies for financial plumbing—in research like Decrypting Crypto: How to Estimate International Stablecoin Flows, and also in policy-facing discussion such as How Stablecoins Can Improve Payments and Global Finance.
- Global standard-setters have been mapping tokenization’s systemic implications, for example in the FSB’s The Financial Stability Implications of Tokenisation and the BIS discussion paper Leveraging tokenisation for payments and financial transactions.
And yet, many enterprise pilots still stall in the same place: not on-chain execution, but off-chain operations—procurement, invoicing, approvals, reconciliation, audit trails, permissions, exception handling, and regulatory controls.
In other words: the world still runs on SAP—and so will most enterprise crypto.
Why “unusable software” is the perfect AI entry point for on-chain finance
The a16z argument is structural: the hardest-to-use enterprise software has the highest switching costs, the largest budgets, and the most painful workflows. That pain creates a buyer with urgency, and a willingness to pay for measurable time and risk reduction.
Crypto’s enterprise pain points map cleanly onto that structure:
- Stablecoin payments are fast on-chain, but slow in the back office (invoice matching, vendor onboarding, ERP posting logic).
- Tokenized assets can settle atomically, but still require the old world’s controls (RBAC, segregation of duties, audit readiness, and policy enforcement).
- On-chain treasury management promises 24/7 cash movement, but CFO teams need deterministic reporting and close processes that fit existing accounting systems.
This is why the best AI opportunities in crypto are often not consumer chatbots. They are enterprise “systems of action” that sit between systems of record and blockchains.
Three phases where AI can make enterprise crypto real (implementation → usage → extensions)
a16z breaks the enterprise lifecycle into implementation, usage, and extensions. Here’s what that looks like when the “new rail” is blockchain.
1) Implementation: translating legacy process reality into on-chain workflows
Enterprises don’t implement stablecoins by adding a wallet and calling it a day. They implement them by answering questions like:
- Which entities can hold balances, and under what policy?
- What are the approval thresholds by region, subsidiary, and counterparty?
- How do we map vendor master data to on-chain addressing?
- What is the exception process when a payment fails, is reversed, or is disputed?
- How do we prove controls to auditors?
This “translation work” is why enterprise rollouts are slow—and why AI can create immediate ROI by turning messy documentation, tickets, and meeting notes into structured requirements, test plans, and migration playbooks (the same pattern a16z highlights for ERP transformations in Why the World Still Runs on SAP).
For crypto teams, that means AI can help generate and maintain:
- Control matrices for stablecoin payment flows
- Mappings between ERP objects (vendors, invoices, subsidiaries) and on-chain representations
- Repeatable runbooks for month-end close with on-chain activity
- Automated reconciliation rules tuned to your accounting policy
The key is not “AI writes a smart contract.” It’s “AI makes enterprise adoption legible, testable, and governable.”
2) Usage and maintenance: making on-chain activity explainable to finance teams
Once implemented, the real workload begins: daily operations.
Enterprise teams do not want to “use a chain.” They want to ask:
- “Which invoices were paid in the last 24 hours, and what’s the variance vs. expected FX?”
- “Show me every transfer above $250k and who approved it.”
- “Explain why this settlement did not match the ERP posting.”
- “Simulate this payment batch before it is signed.”
This is where AI becomes the missing interface—especially when paired with strict controls. It can provide semantic search across ERP records, payment instructions, and blockchain transactions, then generate auditable explanations.
Notably, the enterprise software incumbents are moving in this direction too. SAP has been expanding its AI copilot Joule across its portfolio, for example in release highlights such as SAP Business AI: Release Highlights Q4 2024 and SAP Business AI: Release Highlights Q4 2025.
Crypto builders should treat this as a signal: the UI and workflow layer is where budgets unlock.
3) Extensions: thin, governed apps that connect ERP intent to smart contract execution
Even if your ERP stays the system of record, enterprises still need “thin apps” that make specific jobs easier:
- Vendor onboarding that ends with an approved on-chain payment capability
- A treasury console that batches stablecoin settlement with policy checks
- A tokenization workflow that ties legal documentation to the on-chain asset lifecycle
- An exception-handling flow for disputes, reversals, and failed transfers
AI makes these extensions cheaper to ship and easier to adapt as policies change. But the winning architecture looks less like “one giant crypto dashboard” and more like:
- A governed action layer (what can be done)
- A policy engine (when it can be done)
- A signing workflow (who must approve)
- A traceable audit trail (how it was done)
This is also where “agentic” workflows become practical: not autonomous AI moving money freely, but AI proposing actions that are executed only when controls are satisfied.
The control plane is the product: AI that can touch money must be governable
As soon as AI systems can initiate or prepare transactions, the conversation becomes less about convenience and more about risk:
- Prompt injection and data poisoning
- Misrouted payments due to ambiguous vendor identity
- Policy drift (what changed, who approved it, when it took effect)
- Permission overreach (agents that can do too much)
- Audit gaps (actions without verifiable attribution)
This is why enterprises increasingly anchor AI usage in formal governance frameworks. A practical reference point is the NIST AI Risk Management Framework (AI RMF 1.0), which emphasizes mapping, measuring, managing, and governing AI risks across the system lifecycle.
And for crypto specifically, compliance expectations don’t disappear just because settlement is programmable. FATF’s Updated Guidance for a Risk-Based Approach to Virtual Assets and VASPs remains a key baseline for how regulated entities think about AML/CFT controls and information-sharing obligations.
In enterprise crypto, the “AI layer” must behave like a controlled enterprise system—not a consumer assistant.
Key management: the safest agent is the one that cannot sign
The most robust pattern for enterprises is simple:
- Let AI draft, simulate, reconcile, and explain.
- Let humans (or policy-approved committees) authorize.
- Let hardened signing infrastructure execute.
In practice, this means separating intent from execution. AI can produce a transaction “intent” (who, what, why, limits, metadata), but signing keys should remain isolated—ideally offline.
This is the point where hardware-based self-custody still matters, even in a world full of AI copilots. OneKey, for example, is designed around offline private key isolation with an open-source approach and secure-element based protection, which aligns well with enterprise needs like verifiability, approval workflows, and blast-radius reduction.
The strategic idea is not “buy a device.” It’s: build a signing boundary that AI cannot cross.
What to watch next: the winners will make blockchains “feel like SAP integrations”
If a16z is right that the interface layer becomes the new frontier for legacy systems, then crypto’s enterprise frontier is clear:
- Stablecoin rails will keep growing, but the differentiator will be operational integration (controls, reconciliation, reporting).
- Tokenized assets will expand, but only where workflows match how enterprises already run (approvals, audit, exception handling).
- AI will be the bridge—turning enterprise intent into safe, governed on-chain actions.
Crypto won’t replace the enterprise stack. It will thread through it.
And the teams that win will not be the ones who only ship new chains or new primitives, but the ones who make the hardest enterprise software finally usable—so finance teams can adopt blockchain without needing to become blockchain experts.
Optional: where OneKey fits for enterprise teams experimenting with on-chain settlement
If your organization is piloting stablecoin payments or tokenized treasury workflows, consider treating hardware-backed signing as part of your “AI control plane” from day one. OneKey can serve as a practical foundation for isolating signing authority, supporting policy-driven approvals, and keeping private keys out of always-online environments—exactly the kind of boundary that makes AI-enabled finance safer to scale.
