Authored by Karim Halabi

Token design is the practice of designing incentives so that parties who have no reason to trust each other will nonetheless act in positive-sum ways. In cryptoeconomics, this can be achieved without legal enforcement or a central governing body, just financial incentives.

Just as Uber is a two-sided marketplace for rides, cryptoeconomics enable two-sided marketplaces for digital services. For example, Filecoin matches demand for data storage with supply, The Graph does so for data indexing, Livepeer for video transcoding, Akash for compute, and so on and so forth.

In each case, the native crypto-token of the network exists to align the incentives of these disparate stakeholders and create a self-organising marketplace.

In years prior, these networks depended on:

1. Humans using semantic judgement to participate in these marketplaces

2. Non-thinking machines running scripts to reduce inefficiencies

In the agentic era, we can combine both to create higher-efficiency marketplaces.

Curation: using tokens to signal

Whenever a network lets participants stake or lock tokens against a specific item in order to express a view about its quality or future demand, it is using a curation mechanism. Staked signal is meant to work as a prediction: capital flows toward the subgraphs, datasets, or providers that will be needed, before the fee revenue arrives and it simultaneously signals that the hivemind of the market considers this provider valuable.

The Graph is a canonical example. Curators stake GRT on the subgraphs they expect to be queried, and receive a share of the query fees that flow to whatever they signalled on. The stake is the prediction, and in theory, this signal informs indexers which subgraphs deserve resources before query revenue exists to prove it.

In its original design, Ocean Protocol applied the same logic to data. Stakers locked OCEAN against individual datasets and earned a share of each dataset's consumption fees, with staking explicitly framed as a curation signal pointing towards the datasets worth consuming.

Why it didn't work in practice

In production, these mechanisms did not work perfectly, and both networks ended up changing these incentive designs.

The design itself is not flawed, it was simply designed for the wrong stakeholders. Network participants using semantic judgement to stake (signal) were overwhelmingly humans. Humans, as we know, do not possess infinite attention and time and can only process so much information at once. This resulted in inefficiencies across these markets; a new data source might go unnoticed despite it being of high value. Continuous attention for humans is expensive.

On the other hand, continuous attention for machines is cheap. A bot can stay awake 24/7 monitoring a market, and adjust its stake across assets in real time. Where bots can fall short however is using semantic, non-data driven reasoning.

Curation mechanisms in crypto-networks did not fail because the incentives were wrong. They failed because the mechanism assumed a participant who did not exist.

What has changed in the agentic era

That participant is now arriving. Machines that stay awake and can use subjective reasoning are here, and are abundant. They have continuous semantic judgement.

An agent can answer a different class of question. Is this dataset of genuine quality? Is this subgraph well-constructed, or does it silently mishandle reorgs? Has this storage provider's retrieval latency degraded? Is this a real research corpus or synthetic filler generated to farm a reward multiplier? It can do the job of both a human and a script.

Those are the questions curation was always demanding of network participants. They were never latency problems; they were judgement problems with a latency requirement attached.

How this changes token design

If a continuous, judging participant is now viable, several things that were previously bad ideas become reasonable ones.

Stake becomes a robust demand signal. Human signal was lagging, bots couldnt reason. Agents can do both. Thus, they can allocate stake in real time and give clearer, more time appropriate signals to the consuming stakeholders in marketplaces.

Service consumers get better UX. The small invisible hands of agents create much more efficiency in digital markets, helping consumers of digital services (probably agents as well) get the same service for cheaper, or the better service for the same cost.

That is the version of curation the original designs envisaged. Their flaw was that they assumed these markets would be efficient which was not possible given the existing stakeholders. The emergence of agents now makes this possible.

Designing incentives for thinking machines

Several design principles follow from this analysis.

1. Do not try to funnel attention to only a few service providers. Previously, incentives in networks were rewritten based on the fact that attention was scarce, it is not abundant. Attention is no longer a limiting factor in market design.

2. We can use faster feedback loops to adjust parameters. Again, humans using semantic judgement is slower than machines using semantic judgement. Timed epochs to collect feedback and data can now be shortened as a direct result of this.

3. Epoch-based designs are outdated. Many protocols use time periods of days or weeks in which human participants can make decisions and vote on parameters. These inefficiencies are now redundant and these predictive votes can happen in real time

4. Identity doesn't matter. Identity is ephemeral and free for agents, and networks should be designed from the ground up with this in mind. 

In the future, we will have crypto-networks that are used solely by agents, and the native token coordinates their behaviours and incentives. Cryptoeconomics can and will evolve for this new era.