Sam Cassatt

The through-line

Sam Cassatt keeps returning to one question: at what level of abstraction should a thing be built, and how much trust does it actually need. In 2018 that produced a venture studio designed to build a whole ecosystem at once rather than a single company, and an argument that base-layer Ethereum is a scarce, expensive trust resource that should not be squandered on Facebook likes or DVD libraries [1]. By the time he was presenting Layer, the same instinct had hardened into a critique of the industry itself: after ten years, Web3 is still "bumping up against like the real problems" of decentralization, with centralized front ends, centralized sequencers and centralized provers propping up systems that claim to have none [2].

The shift over time is from ecosystem-building optimism to unsentimental accounting of what was never finished. The early framing was that nothing existed yet, so everything had to be built [1]. The later framing is that a great deal was built, and much of it quietly leans on AWS, GitHub and hosted pinning services [2]. Both positions share a refusal to hand-wave: he is consistently interested in the specific architectural mechanism by which a claimed property is actually delivered.

On decentralization theater

The strongest recurring line in his current work is that the industry substitutes the appearance of decentralization for the fact of it. Front ends are still centralized, which is how Uniswap ended up being sued [2]. NFTs that people believe live on chain are, unless stored on IPFS, "probably stored on something like this," meaning a centralized server such as Piñata [2]. The original optimism rollup design put "an AWS server sitting somewhere that's actually centralizing all the you know transaction sequencing," a single point of failure, and he is careful to frame that as a step in the right direction rather than a reason to bash the team [2]. Zero-knowledge proving is the same story: despite decentralized proving networks from various projects, "pretty much all ZK provs are centralized right now," largely because AWS is more expensive and venture capital will pay for it [2].

The pattern he identifies is behavioural, not technical. "If something's hard to do uh you just don't do it we'll decentralize it later we'll figure it out later," and that deferral "often ends up with either lawsuits or it ends up with honeypots to hack" [2]. The stakes are not only regulatory. He argues the point of removing choke points is "the resilience of the network" and "realizing the true vision of what we all thought we were doing here which is building this resilient self regulating system that doesn't really need any inputs" [2]. He extends the same charity to the internet's founders: John Gilmore's line that the internet routes around censorship as damage was, in his reading, naive because early designers "didn't really anticipate that state level actors would try to then co-opt the network infrastructure" [2]. And he notes the original Ethereum stack, with the Mist browser as a decentralized front end, Swarm for storage and Whisper for messaging, simply did not happen [2].

On putting each workload at the trust level it deserves

His most durable technical argument is that different operations have wildly different trust requirements and should sit at different layers. "Sending a billion dollars has much different trust needs and much different protocol requirements than a like on Facebook," so it is both wasteful and too slow to route a like through base-layer Ethereum [1]. Proof of stake and sharding would take throughput from tens of transactions per second to hundreds or thousands, which he considers "good enough for some applications" and certainly good enough for a layer one running financial applications, but "it's not fast enough to run facebook off" [1]. The conclusion is architectural rather than temporary: faster layers with weaker security guarantees will remain necessary alongside the base chain, with disputes settling down to layer one when something goes wrong [1].

He applied that lens to specific projects. Loom Network interested him because it added delegated proof of stake, a mechanism Ethereum rejected for lacking the right security characteristics at layer one, on top of the base chain along with a framework for other consensus mechanisms that settle back into normal Ethereum [1]. State channel work, including companies such as funfair tech, drew the same approval as ecosystem-advancing [1]. In a social network built this way, the smart contract holding the finances "probably needs to sit on chain on layer one," while the like "doesn't need you know billion-dollar security" and can live somewhere cheaper [1]. He framed the whole distinction in legal terms: what the Ethereum base layer makes cheaper and faster is the thing "that we need normally we need lawyers and court systems and police to execute in our society," and using it for bulk data is like having a notary certify your DVD collection [1].

On storage and the complementary stack

Because Ethereum is not a database and was never meant to be one, he argues for a plurality of complementary systems for large binary data, video, images and long text, naming IPFS as what his teams actually used and messaging protocols as another emerging piece [1]. His explanation of IPFS centres on content addressing: hash a movie into a short string, and that hash becomes both the name of the file on the network and the means of finding it, so a lookup dispatches a beacon across nodes and copies come back from wherever they are held [1]. Two people uploading the same file no longer create two stored copies, which he treats as one of the genuinely interesting properties of the design [1]. Filecoin is the incentivized version of the same idea, rewarding hosts a little like mining, and the result is "essentially they could decentralized data center that stores all this data using people's hard drives all over the world instead of using centralized servers" [1].

On building an ecosystem rather than a single company

The venture studio model at ConsenSys was, in his account, a direct answer to a timing problem. The comparison he reaches for is the early web: they knew blockchain would matter the way people knew the web would matter, but "there was no browser there was no web server there were certainly no devops infrastructure," so everything had to be built at once [1]. A single company making everything was the wrong level of abstraction and he knew of no successful example of it, while a venture fund was premature because "if I had if I had seeded 20 different companies all those companies would have had to build truffle they would have had to build meta masks they would have had to build in fira" [1]. The resolution was "a company that's at a level of abstraction of one company it's a company that builds companies," with applications such as prediction markets, supply chain platforms and games chosen partly because they motivated the shared infrastructure that needed building [1].

Asked whether a hub with roughly fifty portfolio companies chills competition, an "Amazon effect" on the sector, he pushed back on net terms: the market would not have developed as fast without shared tools like Truffle and MetaMask, "there are a whole bunch of companies that wouldn't exist if it weren't for those tools," and in any case entrepreneurship is "jumping off a cliff and building the airplane on the way down," which does not attract the easily deterred [1]. The second business line, ConsenSys Solutions, existed because demand arrived from Fortune 500 companies, central banks and governments, and he compared it to the era before you could buy a database off the shelf: like an early Oracle engagement, the technology needs a solution designed around it before it can be implemented [1]. Around those sat a venture fund, a social impact arm and an education arm, whose first non-developer course launched on Coursera [1]. He has also spoken publicly with Joseph Lubin and Amanda Gutterman about the firm's role in driving Ethereum adoption for real product development [8], and revisited the company's history and the scalability debate elsewhere [6][10].

On running a decentralized organization

He treated the internal governance of a decentralized company as an experiment worth running honestly. Being decentralized meant that his believing something was a good idea "doesn't necessarily mean you know I can tell someone to do it"; the mechanism instead is to assemble interested stakeholders, argue it out and reach a consensus that then filters through the organization [1]. What makes that workable is deliberate information saturation: company-wide meetings every two and three weeks, heavy Slack traffic, events almost daily somewhere in the world, so that anyone who wants to know everything can [1]. His justification is consistency rather than efficiency: "if we didn't try to do something different we would be I think so we're trying to walk the walk and make our whole company an experiment" [1].

On what institutions will and will not do

He is blunt that enterprise clients and the public chain were not yet meeting in 2018. In private or consortium deployments shared by ten banks or ten oil and gas companies, "it's pretty easy for us to modify anything you want and just you know push that project forward and do whatever we want" [1]. Large corporates and governments were mostly not putting projects on public Ethereum infrastructure, and his stance was that you do not "cram something into the market that it wasn't ready for"; instead the market will demand functionality of the Ethereum ecosystem, builders will supply it, and the two will meet [1]. On applications, he identified tokenization as the killer app that had already landed, on the grounds that it had disrupted venture capital and changed how startups capitalize themselves, and was moving on to security tokenization when the recording cut off [1]. He has since discussed the mechanisms for bringing real-world assets on chain to improve liquidity, access and efficiency in traditional finance [4], the application of DeFi to NFTs [3], and yield farming, degens and asymmetric investments in crypto's less well-lit corners [7], as well as how foreign governments view liquid crypto and why Maker is interesting [6][10].

On restaking and containerized AVS

Layer is his attempt to close the gap he has spent years describing. Of the three broad scaling directions, paying for ZK proofs, waiting through optimistic settlement lag, or paying capital cost for cryptoeconomic security with instant settlement, he and his two co-founders believe the third "is probably the right one and the least explored one," and that it is the one that actually delivers the decentralization properties in question [2]. He builds on Ethereum because it offers the most accessible cryptoeconomic security, with EigenLayer sitting on top of it [2].

The market gap he targets is what happens after you have a validator running a Docker container. The container has no chain attached: "you can't send payments you can't change you can't send State um you can't receive state you're pretty much on your own," and even the hello world example is twenty thousand lines of code to square two numbers [2]. The workarounds he has watched people attempt, routing payments through polygon and verifying them elsewhere, or paying a dollar on Ethereum L1 per database write, he compares to a video of someone opening a van by absurdly convoluted means [2]. Cosmos is the one high-usage full stack alternative and taught the industry a great deal, but it is hard to upgrade and maintain and "it's not actually modular it's all spaghetti coated into each other" [2]. So Layer is a new stack, modular, WASM-based, high performance, drawing on ideas like IBC and Tendermint, with a co-founder who was Tendermint's first employee and a co-author of the IBC white paper, and two co-founders who are among the three architects of mesh security [2].

The primitive he claims is new is the containerized AVS. Because WASM can reach the GPU, file system, CPU and network, a developer compiles business logic to a WASM bundle and ships it; Layer orchestrates it across EigenLayer validators with restaked ether and operators already recruited, so "all you have to think about is AI your actual company instead of making and building and upgrading a chain" [2]. It also removes another quiet centralization: a system with billions of dollars flowing through it should not have its distribution mechanism be a Docker repository or GitHub, so orchestration moves on chain [2]. Layer is itself a restaking protocol, containerized AVSs can restake to each other, to Layer, and to anything that speaks IBC, and special handling lets Ethereum L1 state and tokens influence and receive behaviour, functioning as a novel kind of bridge [2]. Teams can deploy the stack as their own L1 or start as an application and grow into a chain [2]. What this unlocks, in his view, is genuinely open-ended: on-chain agents and composable structures that "run an AI process that can trigger something to happen in defi Etc and never touches any centralized process at all," a primitive where "even we don't know what uh what people are going to build with yet" [2].

Takeaways

• Match each workload to the trust it needs: a billion-dollar transfer and a Facebook like have completely different security requirements, and using layer one for the latter is wasteful and slow [1].
• Treat base-layer Ethereum as the replacement for lawyers, courts and police, not as storage; bulk data belongs on IPFS or Filecoin, which he describes as a decentralized data center built from hard drives worldwide [1].
• Most Web3 decentralization is theater: centralized front ends, an AWS sequencer inside early optimistic rollups, centrally run ZK provers, and NFTs pinned on hosted services [2].
• Deferring hard decentralization work "we'll figure it out later" produces lawsuits or honeypots for hackers, and costs the network its resilience [2].
• Building a whole ecosystem requires the right level of abstraction: not one company doing everything, and not a fund whose portfolio each rebuilds the same infrastructure, but a company that builds companies [1].
• Of the three scaling directions, ZK proving, optimistic settlement lag and cryptoeconomic security, he backs cryptoeconomic security as the right and least explored path [2].
• Containerized AVS on WASM lets a team ship compiled business logic that touches GPU, file system and network, inherits restaked security and operators, and skips building and upgrading a chain [2].
• Enterprise adoption follows readiness: in private and consortium deployments you can change anything, but do not cram public-chain architecture into a market that has not asked for it [1].