Moving on from crypto

Why I’m moving on

Since 2017, I have worked in crypto. In 2019, we started Superfluid with a grandiose vision: to change how money worked.

Young and naive, possibly. Definitely a bit too early.

I still believe money streaming is a logical evolution of recurring payments, but I'm now significantly more conscious about how hard it is to overcome entrenched behaviour, especially on something so fundamental.

Earlier this year I decided to step away from my role at Superfluid. Miao (@hellwolf) has taken over, and is steering the team through the rollout of an exciting new product, Superfluid Wallet, a new wallet dedicated to create the best UX for mediating your agent's interactions onchain.

I’m incredibly proud of what we built, and I remain deeply invested in Superfluid’s success. But my own attention is moving elsewhere.



What changed

Superfluid's money streams introduced a radical new UX for a radically entrenched system.

This is the kind of change that takes decades, requires significant alliance-building and probably needs top-down pressure, rather than crypto's bottom-up counterculture.

While we saw signs of early adoption as soon as 2020, the market was simply not mature enough, and unsolved issues across accounting, taxation, and security consumed most of the goodwill we had initially built.

In 2017-2021, I was very bullish on Crypto creating fundamental societal changes. This was one of the reasons I went all in. I thought the technical innovations in trust, financial access, and transparency would change how a majority of people related to money, investments, and more broadly, institutions.

In that idealistic view, Superfluid winning didn't feel too crazy!

I do think crypto has been incredibly successful at creating alternatives to existing financial institutions, and I'm very impressed by advancements in this front. I am also super impressed by some net-new behaviours becoming ubiquitous, mostly in the space of new asset issuance or new market creations (ex. AMMs/Prediction Markets).

But if I ask a narrower, more sobering question: "where has crypto consistently proven itself better enough to displace an existing mainstream behaviour?", the list still feels uncomfortably short.

The clearest cases are not especially flattering: ransom payments, sanctions evasion, and perhaps gambling deposits and payouts. Stablecoins are a more constructive success, but they have largely strengthened access to the US dollar rather than creating an alternative to it.

I once thought crypto would be a tailwind for a genuinely multipolar world. It helped people leave parts of the traditional financial system, reduced certain forms of government control and, at least in the early Bitcoin imagination, offered an escape from dependence on dollars. Contrary to my hopes and predictions, a mixture of institutional capture, revealed preference, and broader geopolitical trends have instead re-centralized Crypto around the US and US Dollar.

The larger societal change

Meanwhile, an enormous shift is happening elsewhere.

Creating Software, which was the most profitable activity skill in the past 30 years, has now been all but automated in the span of roughly one year. I chat more with my agent than my friends and family. My AI Assistant makes more phone calls on my behalf than I do myself.

The societal consequences are clear, immediate and hugely disruptive.

They have pushed me towards a broader set of questions:

Who can see our data?

Who can change the software acting for us?

Who holds its keys?

Can we leave one system without abandoning our history, relationships and ability to function?

These questions become more urgent as software becomes more capable. I have been an avid user of OpenClaw and Hermes since January, and it is obvious to me that the future includes a significantly more digitised experience for all of us.

But when software can read our messages, coordinate our work and act on our behalf, a new class of custody and control problems appears.

In crypto, we learned to ask whether we truly controlled our assets. In the agentic world, we need to ask whether we truly control the systems acting in our name.

Just as crypto created some optionality for leaving the financial system, it is extremely important to create alternatives to centralised AI.

Whether intelligence is "administered" by centralized providers, or broadly available and decentralized (or at least polycentric!) will be one of the decisive battles of the next 5 years.

Any single winner would be a disastrous outcome for human agency. I think this needs to be much more broadly discussed and understood.

A pragmatic path towards private infrastructure

Recently, I have been going deeper into trusted execution environments and confidential virtual machines: TEEs and CVMs.

These technologies allow software to run inside hardware-protected environments designed to prevent even the machine’s operator from freely inspecting its memory and data. Through remote attestation, those environments can also provide evidence about what software is actually running.

They are not magic. Attestation does not prove that an application is safe, and confidential hardware cannot prevent authorised software from leaking information through an allowed channel.

But I increasingly believe TEEs and CVMs are the most pragmatic short-term answer to many of the trust problems we are about to encounter.

More importantly, they offer a promising way to shift what people expect from infrastructure. Today, we generally accept that whoever operates the cloud has ultimate access to what runs inside it. That should stop feeling inevitable.

Similarly, we currently assume that, as software deployers, we can always inspect, fix, or update the software we deploy. CVMs challenge that assumption. Like smart contracts, they can be deployed as immutable and opaque systems: once running, even the deployer may have no way to inspect or modify them. The huge difference is that they run on a full Linux computer.

The new expectation should be infrastructure without standing plaintext access: systems where privacy rests less on the operator’s promises and more on boundaries the operator cannot casually cross.

Building an immune system against AI centralisation requires stronger data sovereignty. For businesses and individuals, data is what defines and differentiates us. Keeping it out of the hands of data gobblers is becoming a basic form of self-defence

That is the direction I want to work on now: private, verifiable infrastructure for software that needs access to the most sensitive parts of a person’s life or business.

I don't think my time in crypto was a detour. Quite the opposite. Crypto taught me to care about trust boundaries, adversarial systems and the roots of control and privacy. Its hardest lesson is that compromises on fundamental values compound in ways that are not immediately visible. What may seem like a small compromise today can grow and spread like a cancer.

AI needs what crypto had: systems that are more sovereign and private, and crucially, systems that remain useful even when their operator is not (or no longer) benevolent.

Those lessons are still the point, and I'm simply moving from one expression of that work to another.

The question is no longer only who owns an asset, but rather who owns and controls the computer that thinks, remembers and acts beside us (or instead of you!).

That is where I am currently spending my time, and I'm excited to reveal my first product in this space very soon.