the next project i'm taking on might just be the biggest thing i've worked on as a protocol engineer. building out the settlement layer for an entire exchange on Solana, from on-chain custody and batched settlement to accounting, reconciliation and all the infrastructure that makes it reliable. feels like everything i've learned about distributed systems, financial markets and blockchains over the years has been leading up to something like this. very, very excited.

this is everything we did to scale @zerithstudio from zero in the last 4 months 00:00 Intro 00:22 Why Zerith started 01:09 What "AI-native" means 02:35 How we operate: tools and content 04:15 Business problem → technical solution 08:41 The engineering: scope and MVP 11:07 Taking it to production 12:25 Domains we build across 13:13 Entering a new domain 14:40 The loop: distribution, execution, learning 20:03 The team 23:54 Biggest lessons 27:27 Wrap

months of work, countless little details, and a vision finally coming together. can't wait to share it. @zerithstudio

we do have fun here @zerithstudio client acquisition is getting interactive.

the stablecoin thesis gets much more interesting once you stop looking at stablecoins purely as payment rails. stablecoins put dollars onchain, but tokenized treasuries are beginning to put something arguably more important onchain: the risk-free rate. crypto historically hasn’t had a particularly clean reference price for capital. a protocol offering 8% and another offering 12% could reflect some combination of leverage demand, token incentives, liquidity risk, smart contract risk and temporary market conditions. “yield” existed everywhere, but there wasn’t always a coherent baseline underneath it. once capital can move between idle stablecoins, tokenized treasuries earning something close to the short-term government rate, lending markets, fixed-rate products and progressively riskier credit, that changes. DeFi starts developing an actual term and risk structure for dollar capital. if an onchain treasury yields 4%, lending at 4.5% against meaningfully riskier collateral suddenly has to justify that extra 50bps. an LP position has to compensate you for inventory and smart contract risk relative to simply holding treasuries. liquidity incentives stop being “8% APY” in isolation and start being decomposable into the baseline return on capital + compensation for the risks you’re taking. the second-order effect is that idle liquidity itself becomes expensive. protocols increasingly have an incentive to make yield-bearing assets usable directly as collateral, liquidity and settlement assets rather than forcing capital to become unproductive every time it moves through the system. stablecoins may end up owning transaction settlement while tokenized treasuries and other yield-bearing assets increasingly own capital storage, with DeFi sitting above both and pricing liquidity, duration and credit risk. stablecoins put the dollar onchain. tokenized treasuries put the price of money onchain. the latter might end up changing DeFi market structure even more than the former.

one of crypto’s strongest ideas is atomicity: a transaction either completes entirely or reverts. that property is incredible for swaps, liquidations and settlement because it removes entire classes of counterparty and settlement risk. but somewhere along the way, we started treating atomic execution as though it were the ideal architecture for every financial interaction. i don’t think it is. real financial markets routinely separate decision, execution, clearing and settlement. a large order can be worked over time, routed across venues, partially filled, netted against other obligations and settled later. it sounds less elegant than a single atomic transaction, but it gives the system something extremely valuable: time and flexibility to find liquidity. this becomes particularly important when liquidity is fragmented across chains, pools, order books and market makers. atomic execution doesn’t cause that fragmentation, but it makes fragmented liquidity harder to coordinate because all the state and liquidity required for an operation often needs to exist simultaneously. we’ve built increasingly sophisticated aggregators, solvers, routers and flash-liquidity mechanisms to make that coordination possible. intents are interesting through this lens. their biggest contribution may not ultimately be “users say what they want instead of constructing transactions.” it may be allowing execution itself to become asynchronous. instead of submitting “swap 500 ETH for USDC right now,” a user could authorize something closer to an execution mandate: sell 500 ETH, receive at least X USDC, complete within 20 minutes, use only these venues, never exceed Y slippage or Z intermediate exposure. now the trade doesn’t have to be a transaction. it can be a persistent state machine. orders can be partially filled. liquidity can be sourced when it appears. compatible flows can be netted. execution can survive failures. individual legs can be retried or compensated. the system can move through intermediate states while remaining inside the boundaries the user authorized. that starts looking much closer to how sophisticated execution works in traditional markets, except the mandate itself can be cryptographically constrained and final settlement can still happen onchain. trust minimization moves from “every operation must happen simultaneously” toward “the executor can do whatever is necessary as long as it provably stays inside my authorized state space.” and this becomes much more important as DeFi moves beyond simple swaps into cross-chain trades, basis strategies, structured products, prediction-market hedges and portfolio rebalancing. buying spot on one venue while opening a perp hedge somewhere else isn’t really a transaction. it’s a financial process with intermediate states. pretending otherwise just pushes that complexity into routers, bridges and executors. DeFi has already started abstracting transactions into intents. i think the bigger transition will be abstracting intents into persistent execution mandates. atomic settlement is one of crypto’s superpowers. atomic execution doesn’t have to be. don’t put the entire trade onchain. put the rules of the trade onchain.

prediction markets being reduced to “betting apps” is probably the least interesting way to understand them. at their core, they’re markets for pricing uncertainty. spot markets price assets, derivatives price exposure to future prices, volatility, rates and credit, while prediction markets let you price exposure to specific future states of the world. the underlying financial concept isn’t new either. state-contingent claims are a very real thing in financial markets: securities whose payoff depends on a particular state occurring. prediction markets just make that primitive extremely intuitive. a contract trades somewhere between $0 and $1 and settles at $1 if a defined outcome occurs and $0 otherwise. that’s also why i think calling them “truth engines” slightly misses the point. they’re closer to expectation oracles. a contract trading at 63¢ doesn’t mean reality has declared something 63% likely; it means that, given the information, models, positioning, incentives and capital of everyone willing to trade, the market currently clears at a price implying roughly that probability. as new information arrives, that expectation gets repriced continuously. and their utility doesn’t stop at speculation. prediction markets can become instruments for actually managing risk. if you own an asset whose value is exposed to a regulatory decision, macro event, election outcome or commodity threshold, you can potentially take the opposite side of that state through a prediction contract without necessarily removing your underlying exposure. traders can use these markets to hedge event risk, use their implied probabilities as signals for positions elsewhere, or trade discrepancies between what prediction markets imply and what spot, futures, options or rates markets appear to be pricing. i think that financial use case is where the much larger unlock eventually sits. getting there requires deeper liquidity, better market structure, reliable settlement and infrastructure designed around traders managing exposure rather than users browsing things to bet on. but once those pieces exist, prediction markets can start moving from a product people occasionally speculate on to financial infrastructure people actually use. that’s the much more interesting framing to me. the question isn’t “what else can we let people bet on?” it’s “what kinds of uncertainty can we make priceable, tradeable and hedgeable?” once you look at prediction markets as markets for state-contingent claims and continuously updating expectations, they start looking a lot less like sportsbooks and a lot more like another layer of financial market infrastructure.

you cannot build the future by averaging yourself against the present.