Is the Human Layer in Crypto Dying? 

Reading time: 8m 56s

If you’re anything like us, helplessly tethered to this industry over the last couple of years, you know that the vibe has definitely changed. 

Things feel a lot less exciting; a bit deserted (reason for Ansem’s routine pulse check tweets), and the only thing that seems to be getting attention is anything with two words tied to it - AI and agent. 

The popular consensus for why this is happening is that the industry is being severely optimized for AI agents, leaving everyone still sweating it out with direct human interaction or “the human layer,” in the trenches. 

So, while from the human perspective the industry might seem a bit stiff, the onchain environment is well and alive on a new layer (agentic layer) where humans can’t technically meddle. 

Efficiency is driving more users to AI-led interactions with platforms that were once designed for human tapping and clicking, now optimized for the non-hoomans.

‍A major player like Uniswap Labs launched 7 new open-source "Skills" for AI agents in February. These tools let autonomous AI coding agents (like those in Claude, Cursor, or other agent frameworks) interact directly and reliably with the Uniswap protocol onchain. 

However, contrary to the “AI agents will eat everything” statements flying around the timeline, a closer look tells a slightly different story - agent activity growth is more sector-specific than industry-wide. 

We decided to dig a bit further into this to see which sectors have been eaten so far and which remain to be eaten. 

Our goal? To find out if the human layer in crypto is truly dying, as well as look into solutions building on top of crypto’s new layer to ensure that control is not lost. 

AI agent-owned zones

In specific sectors, we have observed heavy AI agent-driven activity, with declining direct human interactions. Some of these sectors are listed below: 

Derivatives trading (perps)

The perpetuals market is the most definitively bot-dominated liquid market in crypto. This makes sense. Speed, pattern recognition, and 24/7 execution are things machines do better than people. Nobody's arguing that a human should be manually front-running transactions.

Top 10 perp protocols generated $500.6 billion in 30-day volume, with Hyperliquid alone claiming $189 billion (38%), followed by edgeX ($70.6 billion) and Aster ($64.7 billion).

Aster’s "Human vs. AI" live trading competition, conducted over two weeks under highly volatile conditions, is a clear case in point. 43% of human participants were liquidated, while all 30 AI agents completed the competition without a single liquidation, achieving a 100% survival rate. 

The human trading team as a whole recorded an overall ROI of -32.22%, while AI agents limited total losses to approximately $13,000 and an overall ROI of -4.48%.

Arbitrage trading (MEV)

This is the most absolute case of bot domination in all of crypto, simply because no profitable human MEV operator exists at scale. 

The MEV ecosystem across networks has evolved into a highly competitive automated trading industry, where specialized bots and infrastructure tools scan blockchain mempools.

In 2025, sandwich attacks accounted for $289.76 million, or 51.56% of total MEV transactions. On Solana, sandwich bots have captured between 1.7% and 5.4% of total daily trading volume, averaging 2.9%, resulting in $3.85 billion in sandwich trades executed across more than 3.9 million bundles.

One bot alone accounts for 42% of all sandwich volume, executing trades worth over $1.6 billion in the last 30 days.

This also stretches into DeFi protocols. The entire liquidation lifecycle: monitoring, triggering, and executing are handled by permissionless keeper bots.

Although this existed before the AI agent gold rush, the entire process is now significantly automated by agents as the DeFAI category continues to grow. 

Yield optimization

This category is agent-first by default. Stats show that 68% of new DeFi protocols launched in Q1 2026, included at least one autonomous AI agent for trading, liquidity management, and risk monitoring. 

Compared with the stats 12 months ago, we see a 15% uptick in the adoption of AI agents for yield. 

On platforms like Giza and ZyFAI, AI agents have continued to perform remarkably well - the latter achieving +73.42% yield outperformance compared to static strategies. 

Giza, on the other hand, recorded over 800k total autonomous transactions and up to $40 million in assets under management. 

Beyond Giza and ZyFAI, many more exist in this category, some we’ve covered and others we are happy to cover in-depth upon request and further review - encompassing: 

Arrakis, Reflect, AFI, Lulo, Sail, Almanak, Surf, Infinit, AXAL, Superform, DeFi Saver, Kamino, Mamo, HeyAnon, etc. 

New updates from bigwigs like Pendle, including the deployment of MCP connectors and the building out of skills to make Pendle easily integrate with crypto-native and non-native AI agents, are proof that the yield industry is tilting rapidly towards agent-first interactions. 

Spot trading & portfolio optimization 

Automated trading bots now account for an estimated 65% of global crypto trading volume. Daily active onchain AI agents reached 250k in early 2026, representing over 400% growth from 2025. 

On Solana specifically, AI agents generated $31 billion in DEX volume in 2025, accounting for roughly 2% of the total DEX volume of $1.5 trillion.

What we’ve seen so far is an increase in agent-driven spot trading, including memecoin trades across networks. 

Users are increasingly relying on agent-first infrastructure for token launches, trading, and portfolio management, driving the popularity of platforms such as Virtuals, Bankr, Glider, Surf, Symphony, and many others.

Battleground zones (agent + human activity)

Prediction markets 

Polymarket is the most granular test bed for AI vs. human in crypto, and the data is hard to argue with. We’ve all come across the posts bragging about making millions of dollars on prediction markets with agents. 

However, on a base of 10,582 active traders, 880 bots (8.3% of accounts) averaged $119,156 profit vs. $12,671 for humans - a 9.4x per-trader gap. 

Agents achieved a 66.4% profitability rate, compared with 45.3% for humans. The arbitrage window has compressed from 12.3 seconds in 2024 to 2.7 seconds in 2026, with sub-100ms execution bots capturing 73% of all arbitrage profits.

AI-powered agents now represent roughly 18% of total prediction market volume, and over 30% of Polymarket wallets are already using AI agents.

The nuance, however, is that for markets resolving over weeks or months, the gap narrows dramatically - and in some categories, humans actually outperform. 

Bots have proven to be bad at handling regime changes, so when fundamental dynamics shift, they get confused. On the contrary, humans adapt. 

So what we see is that the short-term arbitrage game is agent-owned, while the long-term judgment game is still human.

This infers a continued balance between agent activity and human interaction on prediction markets for the foreseeable future, until we probably get more sentient models with capabilities to make nuanced decisions that humans still dominate. 

DeFi lending

Lending is another clear example of layered automation. As we noted in the agent-owned zones, liquidation bots are entrenched; however, the vast majority of deposit and borrowing decisions remain human-made.

Aave V3 leads at $24.8 billion TVL, followed by Morpho V1 ($7.4 billion), JustLend ($3.5 billion), and SparkLend ($2.1 billion). 

DeFAI agents have redeployed over $2 billion in TVL across lending and yield-farming protocols - impressive in absolute terms, but representing under 2% of the $130-140 billion total DeFi TVL.

This evidently shows that the decision to deposit, the collateral choice, and the risk appetite all remain human calls. While AI agents handle the plumbing at the edges, the core remains largely human. 

The human-dominated zones

Stablecoins and card payments

The combined stablecoin market cap sits at roughly $312 billion as of March 2026. Adjusted volume, filtering out bot activity, MEV, and wash trading, reached $28 trillion in real economic activity in 2025. That's growing at a 133% compound annual rate since 2023.

Stablecoin transfers under $250 hit a record 5.84 billion in August 2025. We believe that those are humans sending money to family members, paying freelancers, and splitting bills. Over 80% of dollar-backed stablecoin transactions happen outside the agent-adoption-leading US. 

Real people in emerging markets use stablecoins as dollar access and economic hedge, making them directly integral to stablecoin market share. Transaction volume hit $1.78 trillion in February 2026 alone. 

Moreso, the card category is thriving as a result of clearer regulations. Products let users spend crypto assets anywhere traditional cards are accepted, with funds staying self-custodial until the moment of purchase.

This sector is maybe 5% agent-driven. The rest is people moving money. And unlike the bot-dominated sectors, the users here often don't know or care that they're using crypto. That's the point.

Wallets

The wallet is the last layer between a human and the blockchain, and it's the one layer that can't be fully abstracted away. 

While attempts at abstraction have been made, the approval process desperately needs human oversight. Someone has to sign. Someone has to decide whether they trust what's in front of them.

Phantom has over 15 million monthly active users. The entire wallet space is investing in distinctly human-centric improvements such as seedless onboarding, human-readable transaction previews, biometric security, and card-based spending. 

The best wallets in 2026 have evolved from seed-phrase-and-a-prayer storage containers into full financial dashboards.

Enterprise-grade agent wallets in 2026 include budget limits, allowlists, audit logs, and emergency stops - treating the agent as an operator with constrained authority, not an all-powerful signer. 

Human & agent verification layers: The more agents there are, the more this matters

The more agents flood onchain activity, the more valuable it becomes to prove you're human or that the agent is acting on behalf of a human.

There are a few projects along this line, ensuring we don’t get lost in the matrix of a machine world. 

World & AgentKit

First mention: World (formerly Worldcoin - WLD) - these lots have verified over 17 million users using iris-scanning Orb hardware (we know, we know….creepy). 

World describes itself as a response to an AI-saturated world - building digital infrastructure where being human actually carries weight.

And then it launched AgentKit. A toolkit that lets AI agents carry cryptographic proof that they're backed by a unique human via World ID, integrated with Coinbase and Cloudflare's x402 protocol for stablecoin micropayments.

t54

Another project on our radar is t54 (coolest name ever or weirdest name yet), which is building the trust and safety infrastructure (often called the "trust layer") for the agentic economy, the world where autonomous AI agents handle real tasks like managing money, making payments, and transacting on behalf of people or businesses.

Right now, AI agents moving real money is risky (no verification, no accountability, easy to scam or break compliance rules). 

t54 responds to this problem through x402-secure, which is a dedicated trust layer that elevates the x402 protocol for safe AI agent micropayments. x402-secure delivers real-time risk scoring through its Trustline Engine and helps detect fraud, including prompt injection, to ensure accountability. 

t54 provides these guardrails so institutions and users can actually trust agents with finances.

Self Protocol 

These lads are building a decentralized zk-proof human-agent binding layer on ERC-8004 (onchain agent identities). 

Self protocol uses zk tech to anchor every AI agent to a verified human owner (proof-of-humanity) without doxxing or data leaks. 

It prevents sybil, enables self-custody wallets, autonomous actions, and commerce protocols, all while maintaining human accountability. 

Selfclaw is integrated with ecosystems like Celo/Google Cloud, and fees are recycled to support verified agents. 

Kite AI 

Kite is a purpose-built L1 (EVM-compatible, with Proof of AI consensus) designed as the foundational rails for the agentic internet. 

It provides Agent Passport (verifiable identities, delegation, programmable spending rules or guardrails), autonomous, stablecoin payments, governance, and verification so agents can authenticate, transact, and collaborate without intermediaries. 

Concluding thoughts

In all seriousness, we are not anti-agent. The data is unambiguous in trading, MEV, and yield; the bots have won those rooms, and they're not giving them back. 

43% of humans getting liquidated while zero bots do in a head-to-head competition tells you everything you need to know about who owns the speed game.

But the scoreboard across the full stack still shows humans doing the vast majority of work that actually touches real life - in payments, identity, and verification. 

These are the layers where real value is being created, and real revenue is being generated. And they share a common trait: they require judgment, trust, physical presence, or cultural context - things that can't be reduced to an optimization function, yet.

We believe that teams shouldn’t completely abandon building for direct human interaction in these areas and sectors. 

Agents need humans more than humans need agents. At least for now. And we think that the teams that understand that, alongside those building agents and human-proof systems, are worth watching.