AI agent tokens vs memecoins: what is left of the category

Diminished, not dead. CoinGecko's AI Agents category held $3.05 billion in total market capitalisation on $296.2 million of 24-hour volume as of 31 August 2026 — roughly 85% below the category's early-2025 peak, and consolidated into a handful of infrastructure names. The speculative cohort that made the category famous is largely gone. What survives is mostly not memecoin-adjacent any more.

The obituary that prompted this page came from inside the category. On 5 August 2026, Shaw Walters — founder of the project that launched as ai16z and became ElizaOS — told CoinDesk: "The token is dead. Completely." That project peaked at a $2.4 billion market capitalisation in January 2025 and had fallen 97%, to roughly $2.3 million. The rebrand followed an objection from the venture firm a16z to the original name.

One founder writing off one token is not the category. But the arithmetic behind it is worth reading carefully, because the same arithmetic applies to most of what launched in that wave.

What the category actually looks like now

Source: CoinGecko AI Agents category, 31 August 2026, on $296.2 million of 24-hour volume across the category.

Four names account for around $1.9 billion of a $3.05 billion category — roughly 62% of it. The long tail that gave the sector its 2024–25 character has compressed into something much closer to a normal small-cap crypto sector.

Two caveats on that $3.05 billion figure, both ours rather than CoinGecko's. First, the category definition is loose: FET, for instance, predates the agent-launchpad wave by years and is better described as AI infrastructure than as an agent token, so the category total flatters the specific thing most people mean by "AI agent tokens." Second, category market caps are sums of individual market caps, most of which are thin and none of which could be realised at the quoted price simultaneously. Both points are inferences from how the data is constructed, not published caveats.

Where the line between an agent token and a memecoin actually sits

There is a real distinction, and it is narrower than the marketing suggests. The test is simple: does the token do something inside the system, or does it only refer to it?

A token does something when holding or spending it is required to use a service, when it pays for compute or inference, when it governs a treasury or a parameter that materially matters, or when protocol revenue flows to it through a mechanism written into the contract rather than promised in a thread. Those are checkable properties. You can read the contract, find the fee route, and see whether spending happens.

A token only refers to a system when the agent works exactly the same whether or not the token exists. An AI agent posting on X with a ticker in its bio, where the token confers no access, no payment function and no governance over anything consequential, is a memecoin with a technology narrative attached. That is not an insult — it is a description of the mechanics. The value is reflexive and attention-driven, which is precisely how a memecoin works, and it should be understood on those terms rather than as an investment in software.

For most tokens in the 2024–25 agent wave, there was never a line. They launched on the same bonding-curve venues, with the same supply structures, the same creator-fee economics and the same holder distributions as any other launch that week. The distinguishing feature was the description. Our page on what the words memecoin, shitcoin and altcoin actually mean makes the same argument about naming generally: the category label a token claims tells you far less than its contract does.

The clean case in the other direction is infrastructure that uses agents without pretending the token is an agent. Clanker, on Base, is an AI agent that deploys tokens on Uniswap v4 and takes a fixed 20% of the creator LP fee charged per swap, with the creator keeping 80%. That is a mechanism you can verify by reading the fee route. Whether the deployed tokens are any good is an entirely separate question, and mostly the answer is no.

Why the category deflated

Three things happened at once, and none of them was specific to AI.

The narrative rotated. Attention-priced assets fall when attention moves, and the agent narrative had absorbed an unusual amount of it. A 97% drawdown on a token whose demand was narrative demand is not a surprising outcome; it is the standard outcome, arriving faster than usual because the run-up was faster than usual.

The products mostly did not need the tokens. Where an agent framework was genuinely useful, the utility accrued to the software, which is usually open source and free. A token bolted onto a free framework has to invent a reason to be held. Several tried; the ones that could not are the ones trading at a fraction of a percent of their peaks.

Supply arrived. Agent tokens launched into the same venue economics as everything else, which means creator fees, unlocks and heavy early distribution. The mechanics of who is paid, and when, are set out in how memecoin creators actually make money, and they do not change because a token is described as an agent.

The survivors, broadly, are the ones where the token buys or governs something specific. That is an observation about the current composition of the category, not a prediction about it, and we make no claim about where any of these assets go from here.

How to tell which one you are looking at

If you want to classify a token yourself rather than take a category listing's word for it, four checks do most of the work, and all four are answerable in a few minutes.

Find the payment. Is the token spent for anything — inference, API calls, access tiers, agent deployment? If so, find an on-chain transaction where that spending happened. A documented mechanism with no transactions using it is a plan, not a product.

Find the fee route. If revenue is said to reach holders, identify the contract that does it and the address it pays from. Buyback-and-burn programmes are frequently announced, frequently paused, and rarely amended in public when they stop.

Read the holder distribution. Concentration tells you what will happen when attention fades, and it does not care what the token is called.

Check whether the software needs the token at all. If the framework runs fine without it, the token is a claim on attention. That may still be what you want, but price it as such.

What this article does not tell you

It does not tell you whether any of these assets are worth owning, and it names none of them as anything other than data points in a category total. We express no view on price, direction or prospects for any token mentioned.

It does not resolve the category boundary. CoinGecko's AI Agents category is one publisher's taxonomy, and other publishers draw it differently. The $3.05 billion figure is only as meaningful as that definition, which mixes older AI infrastructure with post-2024 agent launches.

The "roughly 85% off peak" figure is a category-level characterisation, not a single published index return. Individual outcomes range from tokens that have held most of their value to the 97% drawdown described above. Do not read the category number as typical of any specific token.

And it is dated. Every figure here is as of 31 August 2026 unless stated otherwise. A category this volatile can look materially different a month later; check current data before relying on any of it.

Frequently asked questions

Are AI agent tokens dead?

No, but the category is heavily diminished. It held $3.05 billion in market capitalisation on $296.2 million of 24-hour volume as of 31 August 2026, roughly 85% below its early-2025 peak, with about 62% of the total in four tokens. What has largely gone is the speculative long tail; what remains is closer to conventional crypto infrastructure.

What is the difference between an AI agent token and a memecoin?

Whether the token does anything in the system. If holding or spending it is required for access, compute, payment or meaningful governance, and you can find transactions where that happens, it is doing something. If the agent works identically without the token, it is a memecoin with a technology narrative — priced by attention, like any other.

What happened to ai16z and ElizaOS?

The project peaked at a $2.4 billion market capitalisation in January 2025 and rebranded from ai16z to ElizaOS after the venture firm a16z objected to the name. The token fell roughly 97%, to about $2.3 million. On 5 August 2026, founder Shaw Walters told CoinDesk: "The token is dead. Completely."

Do AI agent tokens launch on memecoin launchpads?

Many did. Through 2024 and 2025 a large share of agent tokens used the same bonding-curve venues, supply structures and creator-fee models as any other launch. The base rates in why most memecoins go to zero applied to them in full, because mechanically they were the same product with a different description.

Does an AI agent make a token safer?

No. Contract risk, liquidity risk and holder concentration are unaffected by what the project does off-chain. An agent that posts convincingly can make a token feel more substantial while the underlying supply and liquidity remain exactly as fragile as any other new launch. Run the same checks you would run on anything else.

---

The checks that survive a narrative change

Whatever a token calls itself, a buyer can verify very little without trusting someone — which is why on-chain commitments carry weight that descriptions do not. a fixed-term lock on the LP through Team Finance, built by TrustSwap, which also builds Meme Central, holds LP tokens for a fixed term on Ethereum, Robinhood Chain, Polygon, Base and BNB, and surfaces as a verified badge on that token's page in the Meme Central feed regardless of where it launched. It does not stop a team selling its own allocation, and it has no bearing on whether the project behind the token does anything at all.

---

Nothing here is financial, legal or tax advice. Memecoins are extremely high-risk: most lose most of their value, and the majority of tokens launched never reach a decentralised exchange at all. Never spend money you cannot afford to lose entirely. Meme Central does not recommend any specific token. Data described as Meme Central's own reflects tokens indexed by Meme Central and is not whole-market data.