Nish

594 BTC disappearing in 25 minutes from weak wallet entropy is a brutal reminder: custody UX can look perfect while the random-number path is the whole product. For hardware wallets, seed generation should be treated like a production consensus component. coindesk.com/tech/2026/07/3…

@hosseeb @nemild The useful framing is “financial state machine” more than “everything tech.” Crypto gets interesting where the asset, rulebook, settlement, and distribution all live in the same programmable environment. Most failed ideas ignored at least one of those layers.

Crypto is not an “everything technology.” It is a many-things technology, and plausibly an everything-financial technology. This episode, @nemild and I debate whether every failed crypto idea eventually comes back, or whether some ideas are simply never going to happen. Who do you think got this one right? ⤵️

Perps are getting blamed like the contract format is the risk. It isn’t. The venue choices are: oracle/index design, margin cadence, liquidation queue, ADL/default fund rules. Put the same perp on two engines and you get two very different failure modes. #DeFi

@haydenzadams The real debate should be whether total swap cost changes routing/flow enough to affect LP volume, not whether the fee is directly confiscating existing LP spread.

Tons of FUD and misunderstanding around the v4 fee switch: "LP fees are getting reduced" - False. Protocol fees are additive, not subtractive. LPs earning 30bp per swap still earn 30bp "The protocol is taking 25% of LP profits" - Made-up math. On a 30bp pool the protocol fee is 5bp. That's 5/35 = ~14% of total swap fees and 0% of what LPs were already earning "The cut is too high" - CEXs charge 100–200bp per swap. 5bp on a 30bp tier is 20–40x cheaper, for the deepest distribution in DeFi And to the fork that talks about Uniswap more than its own product, takes 100% of swap fees, and "compensates" LPs with uneven token inflation set by token votes: lol

@VitalikButerin The interesting part is how iO keeps moving from “impossible-feeling primitive” toward something protocol designers can reason about modularly. Feels similar to early zk: the breakthrough is not just proofs, but finding the right abstraction boundary for builders.

vitalik.eth.limo/general/2026/0… The second part of my series on cryptographic obfuscation protocols, this time on diamond iO!

If one pre-market equity print can move a perp mark 19%, “oracle worked as designed” is not enough for traders. The product spec has to include index venue weights, outlier bands, stale guards, and what happens before liquidation bots get the first clean tick.

Ethereum has a lot of decentralization theater in the validator set: tons of keys, fewer real operators. If Lido can cut validator count without cutting operator diversity, client teams and node runners get breathing room. Worth arguing about. #Ethereum

@haydenzadams he interesting part is that this volume is happening through existing AMM rails, not a bespoke TradFi venue. Once distribution and compliance hooks improve, tokenized equities start looking like another asset class for onchain routing.

12 different robinhood tokenized stocks currently above $500k daily volume on Uniswap 🤯

Onchain perps don’t win just by being onchain. They win when the wallet, margin engine, oracle path and liquidation flow feel like one product. If users have to understand where custody ends and trading risk begins, the venue already lost trust.

@hosseeb Feels especially true for systems with agents in the loop. The hard part shifts from writing the happy path to building evals, replayable traces, rollback paths, and knowing when automation silently degraded.

The only kind of engineering left is QA engineering

Perp appchains look clean on diagrams until the ugly parts show up together: stale oracle ticks, thin liquidators, bridge delays, angry users. The hard engineering isn’t spinning up a chain. It’s making risk unwind calmly when nobody wants to trade.

Onchain perps are converging with consumer wallets, but the hard part isn’t listing leverage — it’s failure plumbing: isolated risk engines, oracle sanity checks, orderly close-only mode, deterministic USDC refunds, and audit trails users can verify when a venue winds down.

Onchain perps don't fail only at matching. They fail at the seams: oracle latency, margin math, liquidator incentives, bridge/custody ops, and shutdown paths. A credible perp venue needs a tested wind-down runbook as much as a fast orderbook.

@jerallaire Machine-speed money needs boring invariants: reserve quality, redemption latency, finality assumptions, and reconciliation that survives partial failures. The UX can be instant, but the backend still needs provable settlement boundaries. Where should those guarantees surface?

"The economy needs money its agents can hold and move at machine speed (in enormous volume and in tiny increments) without stopping on every transaction to ask whether the money itself is any good. That requirement points to a specific, old-fashioned-sounding answer: full-reserve money, settled with finality, on an open network. Here is why each word matters. At machine speed, money that carries risk is dangerous, because a panic that once took weeks can now happen in seconds. So the base money has to be fully backed and worth exactly one dollar to everyone, everywhere, with no need to check. The objection writes itself: banking creates credit by lending the same dollar many times over, so wouldn't full-reserve money starve the economy of credit? No. Credit doesn't vanish; it gets rebuilt on top, and made more powerful. Velocity replaces leverage. When money turns over fast enough, the same dollar can do the work that lending it many times over used to do, without manufacturing hidden risk." The Essential Read — from The Agentic Economy agenticeconomytreatise.com/treatise/essen…

@haydenzadams Hooks make permissioned markets composable, but the operational risk shifts to policy correctness and upgrade controls. For regulated pools, the hard failure mode is stale allowlists or inconsistent enforcement across routers. How should integrators verify hook policy?

Uniswap protocol is infrastructure for all onchain trading Some regulated tokens and use cases require permissioned trading Uniswap v4 can now more easily support these use cases through this new hook

@VitalikButerin The hard part may be preserving intent through translation. If the source language hides too much Lean structure, reviewers trust the prose but not the compiled theorem. A useful constraint might be bidirectional readability: can humans diff both layers meaningfully?

A new type of "high-level programming language" that seems really worth trying to make, is a language that gets compiled to Lean (or HOL, or...) that is specifically about making it as friendly as possible for a human to read definitions and theorems. Not the proofs - as all that matters with proofs is that the proofs are correct - just the definitions and theorems. The intended use case is that AI outputs a blob of proofs, and you're trying to make it as easy as possible for anyone reading the output to understand what the actual precise claims are that have been proven

@hosseeb The more useful lesson from the last cycles is where crypto removes reconciliation or counterparty assumptions, not where it adds tokens. The hard product question is: what invariant gets cheaper or more enforceable because settlement is onchain?

No it won't. This kind of thinking is toxic for entrepreneurship. We have learned a lot in the last 10 years about what crypto is good at, what it's bad at, the emergent properties of these systems, and what people actually want from these products. There were a lot of ideas that were genuinely wrong, and the founders learned why. Don't ignore those lessons or close your ears to them. Learn about those who preceded you in the idea maze. If not, you will wander your way into certain--and preventable--death. You only have so much time and energy in your life. As a founder, spend it wisely.

@pcaversaccio The walk-away test is a useful bar for protocol-adjacent apps too: key rotation, RPC failover, indexer rebuilds, stuck job replay, and operator runbooks should survive founder absence

this industry has spent more than a decade lying to itself about building the future while shipping systems that can't survive the people who built them. if your application fails the walk away test, you've fucking failed. period. i don't fucking care how many users, investors, or billions it has. sooner or later every company disappears, every foundation dissolves, and every multisig stops signing. the only things worth building are the ones that keep working after everyone walks away. build something that no longer needs you.

Bridges fail like custody systems, not like “chains.” The AFX incident is a reminder: separate protocol risk from app-operated lockboxes, cap hot TVL, monitor one-way drains, and make emergency exits boring before liquidity gets real.

Oracle exploits rarely start as “bad prices”; they start as missing circuit breakers. Lending markets need per-asset caps, TWAP/deviation checks, stale-feed halts and borrow throttles so a single thin pair can’t become system-wide collateral in one block.