1inch (@1inch): AI + DeFi: there are three steps between you and safety

AI + DeFi: there are three steps between you and safety Most DeFi teams are still using AI to do small things faster. First drafts, research, shorter paths to a working code snippet. But marginal productivity gains aren’t enough to make it. The teams that pull through in the next cycle will be the ones who cross three thresholds in time. They’ll redesign their fundamental models around AI. They'll meet enterprise standards, not just developer convenience. And they'll have the full stack of capabilities in place to build use cases that don't exist yet, rather than running existing workflows with AI in the loop. Most teams have dealt with one or two of these. Very few are close to all three. Threshold 1: Redesign your process We all have some kind of AI workflows in place. Some of us are experimenting with agents. But who’s really ready to entrust them with business-critical actions? In DeFi, actions happen on-chain. They might include fetching live pricing data, routing an order, submitting a transaction, checking a position. Most teams currently have AI tooling that handles the first three stages and nothing for the fourth. The research layer is AI-enabled. The execution layer is still manual. That gap is where most of the theoretical value of AI in this industry is currently sitting, unrealised. The coordination required to close that gap is harder than installing better tools. It requires your AI infrastructure to speak to your execution infrastructure. It requires governance around how and when agents can act. It requires decisions about what falls inside an automated workflow and what needs a human gate. That's a process redesign question, not a tooling question. George Sivulka wrote recently about the historical parallel with electrification. Factories that bolted electric motors onto steam-era layouts saw no gains for decades, until they redesigned operations from the ground up around the technology. The teams treating AI as a better way to run existing workflows are at the motor-bolting stage. Threshold 2: Build for enterprise No-one is treating enterprise requirements as friction anymore. Auditability, governance, certification: these are all table stakes now Enterprise-ready tooling is what gives AI workflows organisational legitimacy at scale. Without it, AI stays at the individual layer: each team member using their own tools, in their own way, with no integration into how the organisation actually makes decisions or acts. That's the productivity illusion. Individually useful, institutionally invisible. The decisions product teams make carry financial and regulatory consequences that don't apply in the same way to a marketing team writing faster with AI. A workflow that can route an order needs governance. An agent that can act on a price signal needs audit trails. These aren't bureaucratic additions to an otherwise clean system. They're what make autonomous operation defensible, internally and externally. The teams building AI infrastructure with auditability and governance baked in aren't moving slower. They're building something that can actually be deployed at scale without requiring a senior decision-maker to sign off on every output. Threshold 3: Prepare for what you haven’t thought of yet The depth of your stack is still a moat. Go back and AI-optimize it all. You have no idea what someone might build with it. Autonomous agents that monitor on-chain conditions and execute intent-based orders within defined parameters. Multi-agent coordination across protocols that surfaces arbitrage opportunities, compliance signals, or liquidity changes without human prompting. Compliance-aware execution routing that adapts to regulatory jurisdiction in real time. These new workflows are achievable faster if your agent can work inside the execution infrastructure. Back to Sivulka, who talks about "unprompted" action as the most advanced pillar of institutional AI: systems that detect risks and opportunities humans never think to ask about. What's interesting from a DeFi perspective is that parts of this infrastructure already exist on-chain: intent-based execution, autonomous order routing, MEV-protected settlement. The gap, for most teams, isn't the on-chain execution layer. It's the AI layer that can connect to it intelligently and act within organisational constraints. The gap is already opening The divide between AI-enabled crypto teams and AI-integrated ones is in its early stages. The first group is larger and growing. The second is small. That won't last. The teams that use this window to build complete processes will have structural advantages that keep compounding over time. Network effects, data advantages, and institutional credibility all accrue faster to teams that have built the factory right than to teams that have simply upgraded the tools. That’s why we’ve optimized 1inch Business around our new MCP. You can use it to speed up your integrations, or run complete agentic trading flows. We think it’s the best combination of API diversity, agentic efficiency and enterprise-grade reliability you’ll find.