Francesco Renzi, co-founder of Superfluid, said he has shifted his attention away from the crypto protocol he helped build and toward AI sovereignty, private infrastructure, and systems designed to preserve user control.
In a signed essay, Renzi wrote that he has worked in crypto since 2017 and co-founded Superfluid in 2019 with what he described as a broad ambition to change how money works. He said he still sees money streaming as the logical evolution of recurring payments, but now has a clearer view of how hard it is to change deeply embedded behavior, especially around something as basic as payments.
Renzi steps back as Superfluid moves ahead with a new wallet
Renzi said he decided earlier this year to step down from his role at Superfluid. According to his account, Miao has taken over and is leading the team’s push to launch Superfluid Wallet, a new wallet aimed at creating a better user experience for coordinating agents as they interact on-chain.
He added that he remains proud of what the team built and is still deeply invested in Superfluid’s success, even as his own focus moves elsewhere.
Why the original thesis ran into limits
Renzi said Superfluid’s streaming model introduced a radically new user experience into a system that is already deeply entrenched. In his view, that kind of shift takes decades, requires extensive coalition-building, and may need top-down pressure rather than the bottom-up, countercultural energy that has often defined crypto.
He wrote that the team saw signs of early adoption as far back as 2020, but the market was not mature enough. He also pointed to unresolved accounting, tax, and security issues, which he said consumed much of the goodwill the project had built early on.
A narrower test for crypto’s real-world replacement power
Looking back on the years from 2017 to 2021, Renzi said he once had strong conviction that crypto could drive fundamental social change. He believed technical advances around trust, financial access, and transparency would reshape how most people relate to money, investing, and institutions more broadly. From that standpoint, he said, the case for Superfluid did not seem unreasonable.
Now he frames the question more narrowly: where has crypto repeatedly shown that it is good enough to replace an existing mainstream behavior? In his words, that list remains uncomfortably short.
He said the clearest examples are not flattering, citing ransom payments, sanctions evasion, and possibly gambling deposits and payouts. Stablecoins, he wrote, are a more constructive success story, but they have largely reinforced access to the U.S. dollar rather than created an alternative to it.
Renzi also said he once expected crypto to act as a tailwind for a genuinely multipolar world by helping people exit parts of the traditional financial system, reducing some forms of government control, and, in the early Bitcoin vision, offering a path away from dollar dependence. Instead, he wrote, institutional capture, revealed preferences, and broader geopolitical trends combined to re-center crypto around the United States and the dollar.
Attention turns to AI, software control, and data access
Renzi said a larger social shift is now unfolding elsewhere. He wrote that creating software had been among the most lucrative productive skills of the past 30 years, yet in roughly a year it has become almost automated. He added that he now spends more time chatting with agents than with friends or family, and that his AI assistant places more phone calls on his behalf than he does himself.
Those changes pushed him toward a wider set of questions: who can see our data, who can alter the software that acts for us, who holds the keys to those systems, and whether people can leave a system without giving up their history, relationships, or ability to function normally.
Since January, Renzi said, he has been an active user of OpenClaw and Hermes. To him, that made one point obvious: the future will involve a much more digital experience for everyone. But once software can read messages, coordinate work, and act on a person’s behalf, a new class of custody and control problems appears.
He linked that concern back to lessons learned in crypto. In crypto, he wrote, users learned to ask whether they truly control their assets. In an agentic world, the question becomes whether they truly control the systems acting in their name.
Renzi argued that just as crypto created at least some optionality for exiting parts of the financial system, alternatives to centralized AI are also essential. He said one of the defining battles of the next five years will be whether intelligence is "managed" by centralized providers or becomes broadly available in decentralized, or at least multi-centered, forms. Any single winner, he wrote, would be disastrous for human agency.
TEEs and CVMs as the more practical near-term path
Renzi said he has recently been studying trusted execution environments, or TEEs, and confidential virtual machines, known as CVMs, in greater depth.
He described these tools as technologies that allow software to run inside hardware-protected environments designed to prevent even the machine operator from freely inspecting memory and data. Through remote attestation, those environments can also provide evidence of what software is actually running.
At the same time, he cautioned that the technologies are not magic. Attestation does not prove an application is secure, and confidential hardware cannot stop authorized software from leaking information through channels it is allowed to use.
Even with those limits, Renzi said he has become increasingly convinced that TEEs and CVMs are the most practical short-term answer to many of the trust problems that are about to emerge. He also argued that they offer a promising way to change what users expect from infrastructure.
Today, he wrote, people broadly accept that whoever runs the cloud has final access to what runs inside it. That should no longer feel inevitable. In the same way, software deployers usually assume they can inspect, patch, or update the systems they deploy. CVMs challenge that assumption.
Renzi compared them to smart contracts in one respect: they can be deployed as immutable and opaque systems that, once running, may be inaccessible even to the party that deployed them. The major difference, he said, is that they run on a full Linux computer.
His preferred direction is infrastructure with no standing plaintext access. In that model, privacy depends less on operator promises and more on boundaries operators cannot casually cross.
Next step: products built around data sovereignty
Renzi closed by saying that building an immune system against AI centralization will require stronger data sovereignty. For both companies and individuals, he wrote, data defines and differentiates them, and keeping data away from what he called data devourers is becoming a basic form of self-defense.
That is where he now wants to spend his time: building private, verifiable infrastructure for software that needs access to the most sensitive parts of a person’s life or a company’s business.
He said he does not view his time in crypto as a detour. On the contrary, he wrote, crypto taught him to focus on trust boundaries, adversarial systems, and the roots of control and privacy. Its hardest lesson, in his telling, is that compromises on core values compound in ways that are not always immediately visible.
AI, he said, now needs what crypto once had: more sovereign and more private systems that remain useful even if their operators are not, or are no longer, acting in good faith.
Renzi said he is not abandoning the work so much as moving from one expression of it to another. The central question, he wrote, is no longer only who owns an asset, but who owns and controls the computer that thinks, remembers, and acts beside us, or on our behalf. He added that he expects to announce his first product in this area soon.