Ole Hylland Spjeldnæs

The counterpoint is that reputation differences could point to underlying quality differences in the models and harnesses But for that purpose we will surely have specialized companies to generate benchmarks for arbitrary tasks

I have little faith in «reputation scores for agent-to-agent trust» It works for humans because we respond to incentives Don’t think it will be predictive for agents

Most agentic workflows follow what @eddylazzarin referred to as “AI sandwich” : human director → agent executor → human reviewer. This holds even more when agents automate economic workflows where the cost is of failure is higher. As a result, productivity gains are either capped at human reviewer bandwidth, or worse, agents make errors and lose real money. delta’s role is to automate the verification layer of the sandwich. You attach verifiable settlement constraints (in the form of cryptographic proofs) to any economic workflow, an agent executes it, and delta enforces your rules preventatively, before your money moves. Replace human reviews and counterparty trust with math, and we can raise the ceiling of what can be automated safely at scale.

Every workflow consists of 1. Execution (doing the thing) 2. Verification (ensuring it was done correctly) AI is solving 1, but not 2. Formal verification and cryptography are the tools needed to scale autonomy

Pretty lukewarm on x402 It solves the happy path of payments but that's never the hard part Agentic payments will not take off at scale without solving the boring stuff: liability, dispute resolution, regulation...

@DougieDeLuca Independently of ideology it’s naive to expect at-scale adoption of payment rails that e.g. don’t check for AML The benefit of crypto for the real economy is that we can make these checks automated via programmability

A lot of people, esp crypto natives, still think about compliance in way too black and white of a way. Either it exists and crypto “lost,” or it doesn’t and crypto “won.” I think that framing is way too narrow and misses what’s actually happened here. It is already a massive win that we’ve built financial rails where compliance is not uniformly mandated at the base layer. The fact that there are still options for anonymous, global individuals to hold assets, move money, access markets, and participate outside the constraints of the traditional system is a huge breakthrough. That did not exist in the same way before. That matters. But with that said, it is also a healthy evolution for compliance to show up on-chain. Some counterparties, some use cases, and some markets simply do not exist without it. Certain institutions won’t participate. Certain products can’t scale. Certain users won’t come on-chain at all unless there are clearer rules, identity, and guardrails in place. That is not failure. That is also what success looks like. To me, winning is not “everything must be anonymous” and it’s not “everything must look like the legacy financial system either.” Winning is building a financial system broad enough to serve more than one archetype: - the anon individual, - the global user shut out of legacy rails, - the institution, - the market maker, - the startup that needs compliant distribution, - the user who wants privacy, - and the user who wants protection and clarity. Crypto is most powerful when it expands the spectrum, not when it narrows it. In the past, the system was too narrow because everything was forced into compliant rails. A lot of crypto people now make the opposite mistake and think winning means only serving the non-compliant end of the spectrum. I don’t think that’s right either. The real win is having a financial system where more users, more behaviors, and more use cases can exist than before. Some of those will be fully permissionless. Some will be identity-gated. Some will sit in between. That broader spectrum is the point. And honestly, if crypto can support both anonymous global individuals and more compliant use cases that previously couldn’t exist on-chain, that is a much bigger victory than satisfying only one camp’s definition of purity.

current pov on crypto 1. web2 is great at execution (happy path), weak at verification (proving no rules were broken). crypto is great at verification, weak at execution. 2. verifiable execution is a a red herring. users and counterparties care about outcome guarantees. execution path doesn't matter. 3. the world computer is a harmful mental model. blockchains are deliberately bad computers: deterministic, resource-scarce, painful tooling. 4. new applications are not the point. crypto isn’t enabling tech like the internet or ai; it’s automation that reduces friction. very useful, just not that sexy. 5. blockchains have limited real-world use because they mostly verify onchain facts (signatures, balances, state transitions). real-economy checks (aml/kyc, proprietary logic/data) stay offchain. 6. ai + crypto are the yin&yang of autonomy: ai drives execution cost to ~0; crypto drives verification cost to ~0. ai without verifiable constraints will force all of humanity to work in compliance. at delta we’re building a shared settlement layer gated by expressive, real-world verification. verify constraints, not execution. execution stays web2 (ideally ai).

Impossible to build this onchain btw The word would have to be stored in plaintext so that partial matches (green/yellow letters) can be checked => you could just look up the word directly

@AFDudley0 2/ I don't think total ordering is an antipattern. You're extremely limited in what types of applications you can build without it. delta makes it work because the whole thing is a single state machine even though all execution is pushed out to the domains. Domains do order

@AFDudley0 1/ Are you familiar with arxiv.org/abs/2004.13184? It's basically the answer to the first question. For Bitcoin specifically I think the problem would be how PoW would work. Could you make "PoW certificates" in BRB? Not sure

delta's design is exceedingly simple. To get there, you can start with a rollup stack and just delete stuff: - Remove the need for bridges and 3rd party interoperability services by having all the "rollups" (domains) share state - Integrate the execution layer with the base layer to further simplify interop by making all tokens "L1-native" by default, regardless of where they're deployed - Remove ordering consensus zk settlement enables completely heterogenous execution environments within the global state machine. Our view is that the ideal permissionless system consists of a minimal, neutral base layer which provides censorship resistance and extreme scale, combined with highly opinionated, customized execution environments. This system is delta.

@AFDudley0 Yeah there are similarities, one important difference is that there's no locking funds in a smart contract, so moving assets across domains is a lot easier than I'd imagine for state channels

@spjoleh Have you read any of docs.statechannels.org/research-paper…? Although the motivations aren't identical, the constructions seem very similar, no?

Talking about crypto infra without talking like the people who always talk about crypto infra 🤝 @apriori0x is a thoughtful host and this was fun:)

Moving on from Topology guiltygyoza.xyz/2025/09/moving…

Link: spjoleh.com/articles/how-t…

Pt. III Our fitness app could never be built on a blockchain. Here we lay out how easy it is to build it as a domain

@Pr0cessus0 Just build systems which have different tradeoffs than what exists. A rollup on which you cannot do anything which you can't already do on Solana is so pointless. just use Solana then. At least centralized rollups can be really fast

@spjoleh I guess it’s useful for better liveness, but yeah there are better solutions for that

The whole point of rollups is that you're "secured by Ethereum". You don't need tendermint. Just be centralized. It's ok

@rahulghangas i always forget where i fall on this debate and i'm not looking to remind myself

@spjoleh You mean the bridge is “secured by Ethereum”