a16z: AI Makes Everyone 10× More Efficient — But the Real Winners Haven’t Appeared Yet
a16z: AI Makes Everyone 10× More Efficient — But the Real Winners Haven’t Appeared Yet
AI has already delivered something that feels undeniable: individuals can ship faster, research faster, and automate routine work at a speed that would have looked absurd just a few years ago. Yet, as a16z contributor George Sivulka argues in Institutional AI vs Individual AI, the “10× productivity” moment has not translated into “10× company value.” The productivity is real — but it’s not landing where people expect.
For crypto builders and users, this framing is more than a clever observation. It explains why “AI + blockchain” hype cycles often feel underwhelming in practice, and it points to where durable value may actually accrue in 2025–2026: not in standalone AI tools, but in institution-grade workflows — and increasingly, on-chain institutions.
This piece connects Sivulka’s institutional vs individual AI lens to the realities of wallet security, DeFi, tokenized real-world assets (RWA), and the emerging era of AI agents with wallets.
The electricity lesson, rewritten for crypto
Sivulka uses a historical analogy: in the 1890s, factories replaced steam engines with electric motors — but productivity gains stayed flat for decades, because factories kept the old layout. Only after they redesigned the whole system (assembly lines, unit drives, new job roles) did electricity’s upside appear. That’s the core argument in Institutional AI vs Individual AI.
Crypto has a similar pattern:
- We “swap the motor” when we add an AI copilot to trading, coding, customer support, or risk dashboards.
- But we don’t “redesign the factory” until we rebuild coordination, permissions, auditing, incentives, and accountability — the things institutions are made of.
Blockchains are uniquely good at that second part. Smart contracts don’t just do tasks; they enforce rules, create shared state, and produce audit trails. In other words: if AI is electricity, crypto can be the assembly line.
Individual AI is already everywhere in crypto — and it’s mostly “noise”
In crypto, individual AI shows up as:
- retail traders asking models to summarize narratives,
- devs generating smart contract scaffolds,
- analysts using LLMs to query on-chain data,
- community teams mass-producing content.
This increases output, but it also increases slop — more tokens, more dashboards, more “alpha threads,” more copycat apps. The market gets more crowded, faster.
A concrete sign of where this is going: data firms are already packaging specialized AI into “research in your pocket.” For example, Axios reported Nansen launching an AI chatbot trained on blockchain and wallet data, with an explicit roadmap toward trading agents. Exclusive: Nansen launches new crypto trading chatbot
That’s helpful — but it doesn’t solve the bigger question: how does an organization (or a DAO, or a protocol) turn AI output into reliable decisions and executed actions without blowing itself up?
That is the gap between individual AI and institutional AI.
“Institutional AI” in crypto means more than enterprise software — it means governance + execution
Sivulka’s essay outlines “pillars” that separate institutional AI from individual AI — themes like coordination, signal extraction, objectivity, outcomes, and promptless action. Institutional AI vs Individual AI
In crypto terms, that translates into design questions like:
1) Coordination: who is allowed to do what, with which keys?
If an AI agent can propose a trade, deploy a contract, rebalance collateral, or rotate liquidity — what permissions does it have?
On-chain coordination primitives already exist (multi-sig, timelocks, role-based access control), but AI pushes them from “nice-to-have” into “load-bearing.”
2) Signal: can we prove the data and the execution path?
AI can summarize market conditions — but in DeFi, the most expensive failures come from:
- bad assumptions,
- stale or manipulated inputs,
- and unaudited execution.
Institution-grade crypto systems need:
- deterministic execution (smart contracts),
- transparent state (public chains),
- and increasingly, verifiable data sources and monitoring.
3) Outcomes: does it drive revenue / risk reduction, not just saved time?
In crypto, “saving time” is cheap. Every edge gets arbitraged quickly.
What compounds is:
- better risk controls,
- better capital efficiency,
- better distribution and trust,
- better compliance and reporting for institutions entering on-chain markets.
This is why many AI-augmented crypto products will look less like “chat” and more like autopilot systems with guardrails.
2025’s real “factory redesign” signal: smart accounts and programmable wallets
If AI is moving from suggestion to action, then the wallet becomes the control plane.
Ethereum’s wallet UX and permissioning story has been evolving through account abstraction, starting with the ERC-4337 standard (ERC-4337: Account Abstraction Using Alt Mempool), and then accelerating with protocol-level upgrades. The Ethereum Foundation’s Pectra announcement explicitly highlighted new steps toward broader account abstraction functionality. Pectra Mainnet Announcement (Ethereum Foundation Blog)
Why this matters for AI + crypto:
- AI agents shouldn’t hold “god-mode” keys.
- They should operate under session permissions, spending limits, policy checks, and revocable authorizations.
- Smart accounts make those controls more native — which is exactly what “institutional intelligence” needs.
In other words: AI doesn’t just want a wallet. AI wants a wallet that behaves like an organization.
2025’s other “factory redesign” signal: tokenized Treasuries and on-chain finance becoming institutional
While memes grab attention, the most institution-shaped crypto trend has been RWA tokenization, especially tokenized U.S. Treasuries.
Public dashboards tracking tokenized Treasury products show steady growth and increasing institutional participation. RWA.xyz — Tokenized U.S. Treasuries dashboard
This is “institutional AI vs individual AI” in financial clothing:
- Individuals might use AI to find yield strategies.
- Institutions demand regulated wrappers, reporting, custody processes, and risk models.
- The winners aren’t the people generating more spreadsheets — they’re the systems that turn on-chain assets into repeatable treasury workflows.
AI will amplify this divergence. As tokenized assets become more numerous and more complex, institutions will rely on AI to monitor positions, risk limits, counterparties, and compliance — but only if the underlying infrastructure is auditable and controllable.
So where did the 10× productivity go — and what does crypto have to do with it?
A good mental model is: AI increases “raw output,” but competition converts that output into lower margins and higher expectations.
In crypto, that looks like:
- faster copycats → thinner app moats,
- more “AI-generated alpha” → less alpha,
- more tokens and narratives → harder discovery,
- more automated execution → more need for safety rails.
The value doesn’t disappear. It gets re-priced into the systems that can reliably coordinate action at scale.
That’s why, in the agentic era, the “real winners” are likely to be projects that can deliver:
- policy-driven execution (what is allowed, when, and why),
- verifiable operations (who did what; can we audit it),
- secure key custody and transaction approval flows,
- composable governance (human + agent + contract working together).
This is exactly where blockchain remains structurally differentiated: it can act as the institution layer that AI alone cannot provide.
Practical takeaway for users: AI assistants change how you sign — not whether you should verify
As AI tools become more agentic, the most common failure mode won’t be “bad prompts.” It will be over-delegation: letting automation act with insufficient constraints.
If you’re using AI for crypto operations in 2026, consider three rules:
-
Treat AI outputs as drafts, not authority.
If an AI suggests a contract interaction, verify the target address, approvals, and calldata intent. -
Separate “research” from “execution.”
Use AI to analyze — but route signing through explicit human confirmation. -
Adopt stronger self-custody practices as automation rises.
When the number of transactions increases (because agents can operate 24/7), the risk surface expands too.
A hardware wallet is a straightforward way to keep private keys offline while still benefiting from AI-driven workflows. If you’re leaning into an “AI co-pilot” lifestyle, it becomes even more important to have a signing device that is built for explicit verification, secure key isolation, and multi-chain usage.
OneKey focuses on self-custody with an offline-signing model and an open approach to security engineering — which fits the broader theme here: the future isn’t just smarter tools, it’s better institutions around those tools.
Closing: crypto’s AI moment won’t be a chatbot — it will be a new institution
Sivulka’s question — where did the productivity go? — is the right question for crypto too. The productivity is real. But the upside won’t accrue to whoever generates the most content, the most code, or the most trades.
It will accrue to whoever redesigns the factory:
- programmable wallets that support constrained autonomy,
- on-chain governance that coordinates humans and agents,
- auditable financial rails for tokenized assets,
- and security practices that assume automation is always one step away from catastrophic permissioning mistakes.
AI makes individuals faster. Crypto can make that speed safe, composable, and accountable. That’s where the real winners are still waiting to be built.
