$AGENTO Tokenomics: How We Designed a MiCA-Compliant Utility Token for Distributed AI Compute

By Wingston Sharon | April 2026

I'm going to be direct about something: most crypto tokenomics are designed to maximize fundraising, not to build a functional economy. The token is often an afterthought to the fundraising mechanism.

We're doing it differently. $AGENTO is the reward currency for the Agentosaurus Proof-of-Evaluation (PoE) protocol — and every design decision starts from the question: "does this serve the people doing the work?"

In this post, I want to walk through the tokenomics we've designed for the Q2 2026 launch, and the legal reasoning behind each decision — particularly around Regulation (EU) 2023/1114, the Markets in Crypto-Assets Regulation, which became fully enforceable on December 30, 2024.

This is not legal advice. I'll say that now and I'll say it again. If you're building a token-based system in the EU, get qualified legal counsel specific to your design. What I can offer is the reasoning behind our choices.

What $AGENTO Is (And Isn't)

$AGENTO is a utility token. It is the unit of account for the Agentosaurus Proof-of-Evaluation protocol — a distributed system where GPU contributors run standardized AI benchmark evaluations and earn tokens in proportion to their verified work.

It is not:
- A stablecoin (not an ART or EMT under MiCA)
- A security token (not a claim on profits or governance rights)
- An investment vehicle (you earn it by doing work, not by speculating)
- An e-money substitute (not backed by fiat reserves)

This distinction matters enormously under MiCA. The regulatory path for utility tokens is fundamentally different from the path for asset-referenced tokens, e-money tokens, or tokens classified as financial instruments under MiFID II.

The MiCA Classification Analysis

Let me work through the MiCA classification tree for $AGENTO.

Step 1: Is it an ART or EMT?

Asset-referenced tokens (ARTs) maintain stable value by referencing multiple assets. $AGENTO has no peg mechanism. Its value is determined by the open market based on supply and demand from evaluators and users of the PoE protocol. ART rules don't apply.

E-money tokens (EMTs) reference a single fiat currency. Again, no peg, no reserves, no fiat backing. EMT rules don't apply.

Step 2: Is it a financial instrument under MiFID II?

This is the question that trips up most utility token projects. Under Article 2(2) of MiCA, if a crypto-asset qualifies as a financial instrument under MiFID II, MiCA doesn't apply — you fall under the older, stricter securities framework instead.

The key MiFID II categories to consider:
- Transferable securities: Would the token grant rights to profits, liquidation, or represent ownership claims? $AGENTO grants none of these.
- Units in collective investment undertakings: Not applicable — no pooled investment.
- Derivatives: No underlying reference instrument.

The European Securities and Markets Authority (ESMA) guidance on crypto-asset classification (January 2025) clarifies that tokens issued purely as reward mechanisms for computing services — with no attached ownership rights, no dividend entitlements, and no governance rights over a profit-generating entity — are unlikely to qualify as transferable securities.

We've designed $AGENTO with this specifically in mind. Token holders have no governance rights over Agentosaurus GmbH, no claim on company profits, and no liquidation rights. The token grants access to the PoE protocol's reward economy — nothing more.

Step 3: Does the general MiCA framework apply?

Having excluded ART, EMT, and MiFID II categorization, $AGENTO falls under MiCA's general "other crypto-assets" category (Title II, Articles 4-15). This is the most straightforward path through MiCA.

The key requirements for "other crypto-assets" under MiCA are:
1. Crypto-asset white paper (Article 6): Required before any public offering
2. CASP licensing: Required only if Agentosaurus acts as a crypto-asset service provider
3. Market abuse provisions (Articles 88-92): Prohibition on insider trading and market manipulation
4. Advertising requirements (Article 7): Clear, fair, not misleading communications

Let me walk through each.

The White Paper Requirement (Article 6)

We're required to publish a white paper at least 20 working days before any public offering of $AGENTO. The white paper must include:

Mandatory content (Article 6(1)):
- Description of the issuer and the project
- Description of the crypto-asset and its technical features
- Rights and obligations attached to the token
- Total supply, allocation, and distribution timeline
- Conflicts of interest
- Risk factors

What a MiCA white paper is NOT: It's not a prospectus. There's no mandatory minimum investment protection, no cooling-off period, and no compensation scheme. It's a disclosure document, not a marketing document — and the legal standard for accuracy is high (Article 15 liability for incomplete or misleading information).

We'll publish our white paper 30+ working days before the Q2 2026 launch. The extra buffer is intentional — we want community feedback on the tokenomics before we're legally committed to them.

Why We Don't Need CASP Licensing

This is probably the most important design decision in our tokenomics, and I want to explain the reasoning carefully.

A Crypto-Asset Service Provider (CASP) license under MiCA Article 59 is required for entities that provide crypto-asset services as defined in Article 3(1)(16). These services include:

• Operating a trading platform
• Exchanging crypto-assets for fiat or other crypto-assets
• Custody and administration of crypto-assets
• Execution of orders on behalf of clients
• Placement of crypto-assets

What we're doing that does NOT trigger CASP licensing:

Agentosaurus is not operating a trading platform. $AGENTO tokens are earned through the PoE protocol and can be exchanged on third-party markets — but we're not running the exchange. We're not taking custody of anyone's tokens (contributors hold their own wallets). We're not executing orders on behalf of clients.

We're running a protocol. Contributors plug in their GPUs, run evaluations, and receive tokens automatically through smart contracts. The smart contract is the CASP, in a sense — but a smart contract operated in a fully automated, non-custodial manner does not trigger the CASP licensing requirement under MiCA (Recital 22 explicitly excludes fully decentralized operations from CASP requirements, though this area remains subject to regulatory interpretation).

What we're doing that DOES require care:

The protocol launch itself — the initial distribution of tokens to early contributors — is an "offer to the public" under Article 4. This requires the white paper (discussed above) and the associated liability.

We've chosen to avoid any pre-sale, ICO, or private placement. All tokens are earned through computational work. This keeps us squarely in "utility token issued for services rendered" territory rather than "investment offering" territory. It's a cleaner regulatory position, even if it means slower initial token distribution.

The Tokenomics Design

Now the numbers. Here's the $AGENTO token allocation:

Total Supply: 1,000,000,000 (1 billion) tokens

The 40% mining allocation is the centerpiece of the design. It's the largest single category, it's earned (not granted), and it has no lockup — because the whole point is that people doing work today should receive rewards today.

Why 40% for mining?

We looked at comparable protocols that use token rewards for computational work:

• Filecoin: ~70% to mining (storage work), but over 6 years with complex vesting
• Render Network: 40% to node operators over time
• Helium: 65% to hotspot miners (though its economics have been restructured multiple times)

40% represents a balance between rewarding early contributors generously while maintaining enough tokens for protocol sustainability (ecosystem fund + foundation reserve).

The halving schedule — borrowed from Bitcoin's model — is deliberate. We halve mining rewards every 200 million tokens mined. Early evaluators earn more per evaluation than late evaluators. This isn't just about tokenomics; it's about incentivizing early protocol adoption when the network most needs validators.

Evaluation tiers and earning rates

The PoE protocol has three evaluation tiers:

Tier 1 (T4-equivalent compute):
- Task: HellaSwag + ARC-Challenge (basic reasoning benchmarks)
- Duration: ~30 minutes
- Reward: 100 $AGENTO tokens
- Hardware requirement: 8GB+ VRAM

Tier 2 (A10G-equivalent compute):
- Task: MMLU + HumanEval (knowledge + coding benchmarks)
- Duration: ~60 minutes
- Reward: 250 $AGENTO tokens
- Hardware requirement: 24GB+ VRAM

Tier 3 (A100-equivalent compute):
- Task: Full OSINT-bench suite (5 categories, real-world tasks)
- Duration: ~120 minutes
- Reward: 500 $AGENTO tokens
- Hardware requirement: 40GB+ VRAM

At current design parameters, a Tier 2 evaluator (Mac M2 Max, for example) running 8 hours per day would earn approximately 2,000 $AGENTO per day in the early network period. What that translates to in EUR depends on market price — which we have no ability or desire to predict or guarantee. We deliberately don't publish "expected EUR earnings" because that pushes us toward financial instrument territory.

The Fraud Prevention Design (Why PoE Works)

One concern about token-rewarded compute networks is gaming — submitting fake results, running evaluations without actually doing the work, or coordinating to approve low-quality evaluations.

The PoE protocol addresses this through:

1. Multi-evaluator consensus (5+ evaluators required)

No evaluation result is accepted unless at least 5 evaluators independently produce the same result. Cheating requires coordinating 5+ actors simultaneously — economically irrational when honest work pays the same.

2. Hardware attestation

Each evaluator registers a hardware profile (GPU UUID, VRAM capacity, driver version). Evaluations are matched to hardware tier. Running a Tier 3 evaluation on Tier 1 hardware produces obviously wrong results — the consensus mechanism rejects it.

3. Cryptographic work proofs

Each evaluation produces a cryptographic attestation — a hash of the model weights, input prompt, output, and timing data. This is stored on-chain. Attestations can be spot-checked by anyone, providing a permanent audit trail of which hardware evaluated which model.

4. Stake slashing (not in MVP, planned for Q4 2026)

In the full protocol, evaluators stake a small amount of $AGENTO to participate. Submitting fraudulent results causes stake slashing — you lose more than you gain. The MVP launches without slashing because we want low barriers to entry, but the economic model assumes slashing will be implemented.

The Governance Boundary

We're making a deliberate architectural choice: $AGENTO does not grant governance rights over Agentosaurus GmbH.

This means token holders cannot vote on company decisions, demand financial disclosures, or claim rights to company assets. The company and the protocol are separate. The token governs protocol parameters (evaluation tiers, reward rates, accepted model families) via a separate on-chain governance system — but that governance only covers protocol configuration, not company operations.

This boundary is designed partly for legal cleanliness (governance rights over a profit-generating entity would complicate the financial instrument analysis) and partly for operational security (you don't want token holders to be able to vote to change things that require regulatory compliance).

It also reflects an honest admission: decentralized governance works well for protocol parameters. It works poorly for regulatory compliance decisions that require qualified legal judgment.

What We're Still Working Through

I want to be honest about the open questions.

Jurisdiction of token issuance: We're incorporated in the Netherlands. Dutch law governs our MiCA white paper filing. But we're targeting contributors across the EU and potentially globally. The treatment of non-EU token holders varies by jurisdiction, and we're not yet certain how to handle, for example, US-based contributors (who may face different securities law analysis).

Staking and the CASP question: If we implement stake slashing in Q4 2026, does the staking mechanism trigger CASP "custody" requirements? Current guidance suggests no (you're not taking custody of others' tokens — you're locking your own tokens via smart contract) but this is an area of active regulatory development.

Secondary market dynamics: Once $AGENTO trades on DEXes, how do we comply with the market abuse provisions of MiCA (Articles 88-92)? We need monitoring procedures that detect unusual trading patterns — but monitoring DEX liquidity is technically different from monitoring a regulated exchange.

Tax treatment: Token rewards for computational work are almost certainly taxable income in most EU jurisdictions — probably at the moment of receipt, not at the moment of disposal. We're not qualified to give tax advice, but we're building the protocol with detailed on-chain records specifically to support contributors' tax reporting.

The Timeline

• Q1 2026: Tokenomics finalized, legal review complete
• Q2 2026: White paper published (30+ working days before launch)
• Q2 2026: PoE protocol smart contracts deployed (Ethereum or Polygon — TBD based on gas cost analysis)
• Q2 2026: DEX liquidity pool launched
• Q3 2026: First 100 evaluators onboarded, Tier 1 evaluations live
• Q4 2026: Tier 2/3 evaluations, stake slashing mechanism

Why This Matters Beyond Agentosaurus

The broader context here is that the EU is building a regulated crypto ecosystem that most of the world doesn't have yet. MiCA provides a clear legal path for utility token projects that previously operated in regulatory uncertainty.

That's genuinely valuable. If you're building a protocol that rewards computational work, energy contributions, data labeling, or any other economically productive activity — MiCA's "other crypto-assets" category gives you a compliant way to do it. The requirements are meaningful (white paper liability is real) but navigable.

The projects that will struggle under MiCA are those that tried to use "utility token" as a legal fiction for what was really a securities offering. MiCA is designed precisely to close that loophole. If your tokenomics are designed to fund operations via token sales rather than to compensate actual work, the regulatory path gets much harder.

Ours aren't. $AGENTO is earned, not purchased. That's not just a regulatory design choice — it's the whole point.

Get Involved

The PoE evaluator waitlist opens Q1 2026. If you have spare GPU capacity and want to participate in the early network:

• Mac M2/M3 (16GB+ RAM): Eligible for Tier 1 evaluations
• Linux workstation with RTX 3090+: Eligible for Tier 2 evaluations
• OCI bare metal or A100-class GPU: Eligible for Tier 3 evaluations

Early evaluators get priority access to high-value Tier 3 evaluations when they launch, and contribute to the consensus pool during the period when rewards are highest (before the first halving at 200M tokens mined).

If you're building something adjacent to distributed compute, evaluation infrastructure, or token-based incentive systems, I'd be interested in talking through the design — particularly on the open questions I mentioned above.

hello@agentosaurus.com or find me on the Agentosaurus Discord.

I'm not a lawyer. This post reflects our design reasoning, not legal advice. If you're building a token-based project in the EU, work with qualified legal counsel familiar with MiCA and the specific details of your design.

Wingston Sharon is the founder of Agentosaurus. Agentosaurus builds distributed AI evaluation infrastructure for European organizations. Learn more at agentosaurus.com.