Articles · The token
In mid-October a summary spread across the crypto press: the BitMEX founder had named three marks of projects set to win from the AI-agent boom. We opened the source — an hour-long interview with the podcast The Rollup — and found Hayes saying something else. Below are the verbatim quotes, the reason the retellings drifted, and a check of RNBW against the one criterion he genuinely calls decisive.
16 October 2026 · verified at publication
On 13 October Arthur Hayes — founder of the BitMEX exchange and of the investment firm Maelstrom — gave an hour-long interview to the podcast The Rollup. A day later a short summary went round the outlets: AI agents will make billions of transactions, every transaction is a fee, therefore buy protocols with real revenue. Plus three marks for picking them: used despite a fallen price, a reasonable price-to-earnings multiple, composability with other products.
The summary reads like investment analysis. We opened the source, because quoting a retelling of a translation is not quoting — and the tone there turned out to be different.
The host had asked how he thinks about the flow of capital within his investment thesis. The answer opens like this:
"At the high level I want to be able to tell a very interesting story, and I think the best story I want to tell is: AI agents will run all of your financial services. How are they going to access these financial services?"
— Arthur Hayes, The Rollup, 13.10.2026, 06:45
He then unfolds the picture: trillions of agents, round-the-clock portfolio rebalancing, a multitude of small transactions — "creating all these fees". And then comes a line that appears in none of the retellings:
"I want to tell that story. It doesn't really matter what it is that they're actually doing, because if I can tell that story with a straight face and people can nod their heads and believe me, then that's just gonna pump."
— same, 07:05
This is not an exposé, it is a correction of genre. Hayes is not posing as an analyst and hides nothing about the mechanism: he is describing market reflexivity out loud — the belief of many that everyone else believes. He is talking about building a narrative and about going long whatever benefits from it. The complaint is not with him but with the retellings, which served a discussion of storytelling as a checklist for picking assets.
There is substance in the interview, and it sits elsewhere — where the host asks about belief returning to the market.
"What I think is belief grounded in cash flows. We've been burned so many times… So who's left? It's people with real clients spending real money, and they're handing back money to token holders. Those are the only projects that can survive, because it is expensive to run one of these things."
— same, 11:32
And on what the agents will use:
"The agents are going to use what's Lindy, right? What's been around since 2020, DeFi summer… It's beaten-down stuff that's really well used. Maybe they just started giving money back to token holders because their lawyers let them."
— same, 12:13
And above all, about which projects will grow. His answer is negative in form: it will not be the new ones.
"I don't think this is going to be a wave of people starting new projects and those doing really really well. I'm sure there's going to be lots of copycat stuff."
— same, 12:49
So his subject is not AI agents as such at all. The subject is where the next cycle's money goes, and the agents are merely the story that will carry it there. The winners, in his telling, are those already running for years, used while the price has fallen, and handing money back to holders.
A note on accuracy. The interview transcript is machine-generated, and its speaker-change markers are placed imprecisely. The widely quoted line "the PE ratios are very reasonable and they're composable" sits immediately after a speaker change — it may have been said by the host, not by Hayes. So we do not attribute it to him, even though two of the three marks in the retellings grew out of it.
What remains is one criterion, stated by Hayes unambiguously and twice: the project hands money back to token holders. Not "earns" — hands back. That is the test worth applying.
The three marks from the retellings are still worth checking — one just has to know whose list it is. Two of the three we can measure with our own data; the third needs an honest "no".
The token launched in February at $0.0367, sank to $0.0121 in May, and now trades at …. Over the same stretch staking grew without a break: zero in March, 13.7M in June, 16.7M in August and … now. Stakers: …. The growth continued through the months when the price sat at its lows, so the mark holds in fact and not merely in words.
The Rainbow app takes 0.85% on a swap made inside it, and per DeFiLlama that comes to … a year, with … all time. Market cap to revenue is …, and … on the full supply. For this industry that is low; the multiples are taken apart in "RNBW is called a gem".
Here the answer is negative, and it is the most honest finding of this piece. RNBW trades in Aerodrome and Uniswap pools, but nothing is built on top of it: it is not accepted as collateral, it does not sit in anyone's vaults, it is not a brick for other protocols. Liquidity and composability are different things. The "Lego blocks" the interview speaks of mean Aave and Curve; RNBW remains an app token.
This is the very mark Hayes names himself. And here the answer is yes — verifiable on chain rather than taken on trust.
Rainbow buys RNBW back off the market. It is done by the address holding the bulk of the supply not yet in circulation — a Safe multisig where a transfer needs two signatures out of three:
0xcaa9bFCbc9ff87428535bc4E163296c3A8ae308E
It bought … RNBW between February and October 2026 — with market orders through Uniswap and Aerodrome pools, paying in USDC. Nothing is sold back: what is bought stays there.
Counted by us from the token's transfer log as of 16 October 2026; the latest purchase was on 12 October. In total … RNBW arrived at the address from pools across … transfers, but the February arrival of … RNBW came with no USDC paid in the same transaction — we do not count it as a purchase and have subtracted it. The shape of the buying is what separates a buyback from a trading bot: the bots alongside it show 24,000 trades and a zero balance, this address shows twenty-nine trades and nothing sold.
One caveat, without which the picture is incomplete. No obligation to buy back is fixed on chain. There is no contract routing a share of fees into purchases of the token, and no document requiring it — the buyback remains a company decision, one it is free to stop. Hayes, as it happens, describes exactly that nature: "maybe they just started giving money back to token holders because their lawyers let them".
Beyond the buyback, the holder gets what the token's own mechanics provide: a refund of part of their own fee as cashback, a share in the redistribution of other people's early-exit penalties, and staking rewards. That is not a share of the app's revenue — cashback depends on how much you personally traded — but the market buyback works differently and touches every holder at once.
On Hayes's own test — money handed back to holders — RNBW answers yes: the buyback is running and measurable. Of the three marks from the retellings two hold: usage grows while the price has fallen, and the multiple is low. Composability fails, and it is the only negative answer in this piece.
The buyback exists, but it is voluntary: neither an obligation nor the share of revenue going into it is fixed on chain. A mechanism written into a contract, or an announced programme with a verifiable address and a stated regularity, would remove the caveat entirely. What happens can be watched with the same data we already gather: purchases by the treasury address show up in the chain the same day.
This piece began as an application of someone else's checklist and ended with the discovery that the source contains no checklist. Between an hour-long interview and a three-hundred-word note the tone is lost, and the meaning with it: a discussion of how to tell a convincing story turned into a method for picking assets.
The limits of this piece, so it is not read more broadly than it is.
We are not claiming that Hayes misleads anyone: he describes the mechanics of the market openly and speaks about his positions plainly. We are not claiming his AI-agent thesis is wrong — only showing that in the source it is presented as a narrative rather than as a conclusion drawn from data. We give no investment advice and do not suggest buying or selling RNBW: the figures above are measurements, not a forecast. Past growth in staking promises nothing about the future.
This portal is independent and not affiliated with the Rainbow team. Every number here updates automatically from the blockchain and public sources; the quotes are checked against the machine transcript of the interview, and the link to the original is in the first section.