03 · Intent Architectures
Intent-Based Architectures
Users specify what they want, not how to do it. The system chooses the path. The UX must make that handoff trustworthy.
By John Wright-Nyingifa · Product Designer building infrastructure for DeFi, DePIN, and autonomous agents.
Live Signal · March 2026
CoW Protocol processed $87B in 2025 (up from $40.2B in 2024). 1inch Fusion+ cross-chain hit $2.42M daily median (+129% QoQ). ERC-7683 (Across + Uniswap Labs) adopted by Arbitrum, Optimism, Scroll, Polygon. 40M+ smart accounts deployed. Anoma and Essential moving intent-centric architectures toward production.
Instead of "bridge then swap then deposit," the user submits an intent: "End with 1 ETH on Chain B within 10 minutes, paying at most $8 in fees."
The product advantage is abstraction. The system risk is ambiguity. This page covers how intents are structured, how solvers execute them, where uncertainty enters, and what UX must expose so users can trust the outcome.
Intent Structure
An intent has two parts: the desired state (outcome constraints: asset, amount, destination, time) and the execution strategy (left open for the system to choose). The product advantage is abstraction. The system risk is ambiguity.
Minimum received, max slippage, max total fees, deadline/time window, trust tier (secure vs standard), allowed protocols/chains. A well-designed intent is a safe contract between user and solver.
"Buy up to 500 USDC worth of token X if price ≤ Y." Make partial fulfillment explicit in previews. Define whether leftovers refund, stay as input asset, or roll over.
Cross-chain intents introduce dependencies: message delivery, verification/finality, liquidity on destination. Model as phases rather than a single success state.
Routing adapts to changing prices, congestion, solver availability, bridge delays. Powerful, but must remain within constraints.
Execution Models
An actor that proposes an execution plan, commits to an outcome, gets paid if achieved. May be centralized (single router) or competitive (many solvers bidding). CoW runs batch auctions. 1inch runs Dutch auctions. Across uses relayers.
Solvers search over venues (DEXs, aggregators), bridges, sequencing strategies, route splitting. The search space is where "better execution" comes from.
Package steps to reduce risk: approvals + swaps + deposits. Across chains, bundling becomes phase-based. True atomicity is rare.
Solvers score by: expected output, risk (trust tier, bridge risk), time to finality, probability of failure, operational cost. Good product summary: "best price," "fastest," "most secure."
Best effort (tries within constraints), Quoted outcome (commits to minimum received), Insurance/compensation (refunds on specific failures). Guarantees should be a first-class label in UX.
Competitive routing introduces bidding, solver reputation, winner selection. New failure modes: solver dropout, stale quotes, censorship or collusion.
SOLVER COMPETITION (March 2026) CoW Protocol ████████████████████████ $87B annual (batch auctions) UniswapX ████████████████ Part of Uniswap's $1-2B/day 1inch Fusion ██████████ $67.6M daily avg (+60.6% QoQ) Across █████████ $28.6B cumulative (ERC-7683) GUARANTEE TIERS ├─ Best effort Tries within constraints, no commitment ├─ Quoted outcome Commits to minimum received └─ Insured Refunds on specific failures
UX Implications
Help users express constraints without being technical: "Max fee," "Finish by," "Prefer safer route." The form is the contract.
Show: expected received (and minimum), fee breakdown, time range by phase, trust tier, what happens on failure (refund, retry, reroute).
Users don't need the whole search space. They need: why the chosen plan wins, top alternatives, what tradeoff was made.
"May take longer during congestion." "Verification required before withdrawal." Don't hide uncertainty behind "processing."
Submitted → Executing → Waiting for verification → Finalized → Needs attention. Five states, not two.
UX Implications
→ Design the "searching" state. Between intent submission and solver selection, users see nothing. This needs its own animation, not a repurposed spinner.
→ Show competition results, not process. "Saved $3.20 vs next best route" after settlement. Not a live auction dashboard.
→ Partial fulfillment is not an error. Use a progress bar, not an error modal. "82% filled, remaining 18% estimated in ~30s."
→ Account abstraction changes the entry point. 40M+ smart accounts means intents without manual signing. "State your goal" not "approve & swap."
Glossary
Declarative goal + constraints.
Actor that finds and executes a plan.
Proposed outcome under conditions.
Packaged set of steps.
What the user is promised if execution fails.