Skip to main content

Nansen's 30-Month Bet: Why Billions of AI Agents Will Run Crypto Portfolios by 2028

· 11 min read
Dora Noda
Dora Noda
Software Engineer

On May 2, 2026, the most-cited on-chain analytics firm in crypto published the kind of forecast that quietly resets an entire sector's planning horizon. Nansen — the platform that indexes more than $2 trillion in tracked wallets and whose smart-money labels show up in nearly every serious crypto research deck — argued that by 2028, billions of AI agents will be the default vehicle for crypto investing. Not a feature. Not a niche. The default.

That is a 30-month timeline. For comparison, the software industry's own shift from manual coding to CI/CD pipelines took roughly a decade. Nansen's bet is that LLM acceleration plus on-chain composability compresses the analogous "manual-to-agentic" investing migration into less than three years. If the firm is even directionally correct, the implications cascade through every layer of the crypto stack — from how liquidity gets quoted, to how token launches are designed, to how RPC infrastructure gets billed.

Why This Forecast Carries Unusual Weight​

Predictions are cheap in crypto. Almost every research firm publishes a bull case for the technology it sells against. What makes Nansen's 2028 call structurally different is the firm's role in the market.

Nansen sits at the data layer. Its wallet labels — the "smart money" tags that identify VC desks, market makers, and notable individual traders — are referenced in VC theses, ETF prospectuses, exchange product roadmaps, and competitor research notes. When Bernstein wrote its tokenization supercycle thesis, when a16z published "stablecoins as the breakout app," when ARK called Bitcoin to $2.4 million — each of those forecasts became a reference point that other allocators had to either adopt or explicitly argue against. Nansen's agent forecast plays the same role for the AI-agent infrastructure layer.

The credibility is also self-fulfilling. Nansen's own product roadmap now includes a conversational trading agent that interfaces with aggregators like Jupiter and OKX to finalize trades from natural-language prompts. The forecast doubles as positioning. CEO Alex Svanevik has been laying the groundwork since February 2026, when he publicly forecast that by 2030 the primary interface for investors would be an AI agent rather than a dashboard. The 2028 number is the institutional version of that thesis — early enough to matter for current capital allocation, late enough to be defensible.

The Number That Changes the Architecture​

Billions of agents — not millions — is the part of the forecast worth reading carefully. Today's market structure assumes one human per wallet, occasionally one trading bot per strategy. Nansen's vision is one investor represented by many agents, each holding distinct strategy parameters, monitoring different on-chain conditions, and executing autonomously in parallel.

The shift is already visible in the data. Recent April 2026 reporting suggests that 95% of hedge funds have moved from manual LLM prompting to agentic frameworks — autonomous multi-agent systems that don't just describe the market but actively transact within it. AI agents are now estimated to command roughly 58% of automated investment decisions across institutional desks. The agentic AI sector itself sits at a market capitalization north of $22 billion as of late Q1 2026, with the broader Web3 AI agent market valued near $7.81 billion and growing.

Capital is following. Roughly 40 cents of every venture dollar invested in crypto firms during 2025 went to companies combining AI and crypto — more than double the 18 cents of the prior year. Coinbase Ventures was the most active crypto investor in Q1 2026 with 12 deals; the firm has openly prioritized agent infrastructure plays in its public theses.

What "Agent" Actually Means in 2026​

The vocabulary has drifted, so it is worth being precise. The agents Nansen is describing are not the rule-based trading bots of the 2020s. They are goal-directed systems that reason across multiple data inputs and execute multi-step strategies across DeFi protocols, centralized exchanges, and on-chain positions simultaneously.

A typical "agent fleet" in 2026 specializes by role:

  • Macro agents ingest Fed signals, global liquidity prints, and ETF flow data
  • Narrative agents scan Farcaster, X, and Telegram for sentiment shifts and emerging meta
  • Execution agents optimize routing, gas, and slippage across venues
  • Risk and compliance agents police position limits and flag regulatory exposure

Research has shown that "three-layer multi-agent frameworks" — typically a bull agent, bear agent, and risk supervisor in adversarial debate — consistently outperform single-model LLMs on out-of-sample evaluation. The dominant pattern is no longer "one big model" but committees of smaller, specialized models routed by an orchestration layer.

This is the architectural insight behind Svanevik's "trust ladder" framing. He has been blunt that pushing investors straight to fully autonomous trading would be the equivalent of climbing into a Tesla and immediately moving to the back seat — a setup for losses, regulatory backlash, and security incidents. The phased model is co-pilot first (agent suggests, human confirms), then constrained autonomy (agent executes within hard guardrails), then full autonomy for a narrow set of strategies. Nansen claims its expert-mode agents reach an 85% quality score on internal evaluations, against roughly 20% for unaugmented general-purpose models — a gap built by injecting the firm's proprietary on-chain analytics into the agent context.

The Market Structure Reset​

If Nansen's 2028 horizon proves right, several pillars of current crypto market structure get rebuilt at the same time.

Liquidity microstructure compresses. When agents replace humans on the bid and ask, spreads on long-tail tokens narrow, and quote refresh rates accelerate by orders of magnitude. Front-running dynamics on intent-based DEXes shift as solvers themselves become agents racing other agents in microsecond windows. Market makers that already run AI on the inside of their stacks gain disproportionately; smaller bots become the prey rather than the predator.

CEX-vs-DEX share rebalances. Agents prefer programmable venues. Composability — the ability to chain swaps, lending, perps, and bridging into a single transaction — is a feature humans rarely use in practice but agents exploit constantly. Centralized exchanges respond by building agent-callable APIs, MCP-compatible endpoints, and SDKs that match the ergonomics of on-chain venues. Hyperliquid, Drift, and the Solana DEX cluster benefit by default because their architecture was already programmatic.

Token launches change shape. Pitch decks and Discord launches are tuned for human attention. Agent-mediated capital allocation requires machine-readable disclosures, structured tokenomics specs, and standardized risk schemas. TGEs in 2027–2028 may look more like API documentation drops than community announcements — and projects that fail to publish in agent-readable formats simply do not show up in agent-driven discovery.

Systemic risk concentrates. This is the underdiscussed flip side. Thousands of agents trained on overlapping datasets and reading the same on-chain signals can produce algorithmic resonance — synchronized sell-offs that move faster and deeper than any human-driven crash. The flash-crash regime of equity markets in the 2010s is a preview, not a warning that has been heeded. Risk teams at exchanges and lending protocols are already war-gaming agent-correlated liquidation cascades.

What This Means for Infrastructure​

The shape of demand on the underlying infrastructure changes in ways that most providers are not yet pricing for.

Traditional crypto infrastructure assumes a human-trader access pattern: bursty, large, and intermittent. A retail user opens a wallet, refreshes a dashboard, executes a trade, and disappears for hours. RPC providers, indexers, and data services built rate limits and pricing tiers around that shape.

Agent fleets invert it. The new pattern is high-frequency, low-payload polling — thousands of small calls per minute per agent, sustained continuously. An execution agent monitoring liquidity across five chains generates more requests in an hour than a human user does in a month. Multiply by the "billions of agents" figure and the load curve resembles industrial telemetry more than retail finance.

The implications are concrete:

  • Rate-limit architectures need rebuilding to distinguish agent traffic from human traffic and price each accordingly
  • Read throughput becomes the binding constraint before gas in many workflows, requiring providers to treat reads as seriously as writes
  • Flat predictable pricing beats percentage-based fees for agents executing 10,000 transactions a day; percentage-based pricing simply routes the agent elsewhere
  • Wallet infrastructure splits between reasoning agents that query data and wallet-as-service agents that hold custody — each consuming infrastructure differently

The numbers are no longer hypothetical. In a 14-week beta program running from October 2025 through January 2026, over 1,000 participants created more than 9,500 agents that executed 187,000 autonomous crypto transactions. The x402 protocol — built specifically for autonomous machine-to-machine payments and API paywalls — has already processed more than 50 million transactions. The agent economy is past the proof-of-concept stage and is now scaling through operational pain points that infrastructure providers have to solve in real time.

BlockEden.xyz operates RPC and indexing infrastructure across 27+ chains, with rate-limit tiers and predictable pricing designed for both human-trader and agent-fleet workloads. As agent traffic shifts from edge case to default, the infrastructure layer that serves both reasoning and execution patterns becomes the toll booth of the agent economy. Explore our API marketplace to build on foundations sized for the next traffic regime.

The 2028 Bet, Restated​

Nansen is not the only voice forecasting agentic dominance. MoonPay's Open Wallet Standard, Coinbase's Agentic Wallet, Virtuals Protocol's economic OS thesis, and Bittensor's subnet expansion all point in the same direction. What Nansen contributes is the timeline and the credibility math: a most-cited analytics firm publicly anchoring on a 30-month horizon forces every other allocator to position for or against that view.

History suggests these reference forecasts shape behavior even when they miss the date. Bernstein's tokenization supercycle reset RWA roadmap allocations even as the actual TVL ramp lagged the forecast. ARK's Bitcoin price targets shaped corporate treasury cases regardless of whether the number printed. Nansen's 2028 call will likely do the same for the agent infrastructure layer — moving capital and roadmaps now, on the assumption that the architecture will be in place when the volumes arrive.

The unresolved questions are not whether agents will dominate, but which architecture wins, who captures the toll on every agent transaction, and whether the systemic-risk profile of an agent-mediated market gets stress-tested by a regulator-friendly incident before it gets stress-tested by an unfriendly one. Those answers will be written between now and 2028. Nansen has just placed its marker on the calendar.

Sources​

Share on Twitter