Happy Friday y'all 💚 The dominating AI compute is not one-size-fits-all Extremely important notion that a small inference, a huge reasoning model and a private verified job should not all be sent to the same machine under the same rules so @ritualnet Resonance work makes so much sense its that simple idea of matching the job with the right compute Hardware, models, latency, prices and guarantees may be different MLCommons benchmarks already show how dramatically AI workloads differ in compute and performance requirements Ritual takes that reality and solves question like what if compute itself could become an open market? giving a much closer feel to how AI infrastructure actually needs to work

Volkswagen passed emissions tests while its cars were polluting the environment The software recognised when it was being tested and changed its behaviour. In 2015, regulators uncovered the deception. Affected cars could emit nitrogen oxides at up to 40 times the permission. I think about that when we talk about trusting AI with money. Volkswagen wasn’t an AI incident. But it showed how much can go wrong when people cannot inspect the software behaviour behind a reassuring result. That’s where @BaranosAI has an edge. Its design commits to those details before the computation runs. Independent participants can reproduce the job, challenge an incorrect execution and have the disputed computation checked onchain through @fogo Suppose a prediction market utilizes AI to assess evidence. If the operator quietly substitutes another model or changes the inputs, participants need a way to challenge that before accepting the settlement. Of course, the agreed model can still be wrong. Verification won’t fix poor judgment. It checks that the agreed computation was followed. Learning more n more about BaranosAI made this much more tangible for. The more responsibility we give AI, the more valuable that ability to check and challenge becomes.

GM CT 💚 If an AI agent disappears the moment one server goes down, I’d struggle to call that fully autonomous The @ritualnet persistent-agent design feels important here Agents getting better at reasoning and taking actions, but long-running reliability is still a problem Research on agentic systems keeps showing the same thing like being capable is not the same as being dependable over time Ritual is dope tackling the infrastructure side Agents can save its state, prove it is still alive, and recover from its last checkpoint just so unlike dying with the machine that hosted it For something managing capital, running a strategy, or coordinating with other agents, that matters a lot, what do you say?

Building the BaranosAI Educational Quest left me with a question I keep coming back to: Can an ordinary participant actually afford to hold an AI executor accountable? One detail in @BaranosAI design deserves more attention. Its Durable Compute Graph breaks computation into small, deterministic operations called kernels. A dispute progressively narrows the disagreement until the chain can adjudicate one kernel, which fits inside a single SVM transaction. The full challenge still involves multiple rounds. That gives Baranos a promising way to limit the computation needed for the final onchain ruling. But affordability needs the whole bill. A watcher still recomputes the result offchain and must respond within deadlines. The developer docs list a 1,000,000-lamport challenger bond in the alpha tooling. That is configurable capital at risk, not the price of verification. They also estimate roughly an hour of executor work for a 10,000-token document, which shows why hardware and workload matter alongside gas. I haven’t found a reproducible benchmark that puts the complete challenger cost into dollars. The whitepaper itself calls for total chain, execution and verification costs, with hardware, storage and network conditions disclosed. That’s the next demonstration I’d love to see: an independent participant challenging a deliberately incorrect execution, with the compute bill, transaction costs, bond and time to settlement all published. If those economics work for smaller participants, Baranos could make accountability much more accessible. That would give builders on @fogo a strong reason to use it. After helping people learn how verification works, I want to see how affordable it is to exercise that right.

GM CT 🌻 I absolutely loved reading this... dealing with the part most AI x crypto narratives skip; failure handling In production, AI is not a clean synchronous function call means responses can be late, malformed, fail mid-job, or come back through the wrong path The model @ritualnet's callback forces the contract to account for that explicitly in terms of timing, sender verification, execution success or output parsing much more mature design philosophy I'd say The real infrastructure edge is surely making AI callable and making AI failure legible, bounded & safe enough for contracts to rely on 💚

Gm CT ☀️ Happy Wednesday I think people overlook with Ritual is execution itself AI is getting most attention these days but Sequencing Rights is the part that feels structurally important On most chains, apps submit transactions and still leave ordering largely in the proposer’s hands Ritual lets the application define its own ordering rules it'd be useful to think of an exchange where a cancel request can be enforced before a later fill because the proposer chooses to be fair and the app defines how execution should happen Ritual is giving intelligent applications more control over the environment they execute in With mainnet so close, this is one part of @ritualnet I’d watch carefully

Co-location is probably the most TradFi thing about Raiku and I really mean that as a compliment Trading desks have spent years paying to be closer to exchanges When an arb disappears before the order even arrives, proximity starts to indicate an expense that thinking is brought to @solana by Blackline proximity to execution itself is the @raikucom 's edge I think more desks will start budgeting for this kind of access as they move strategies onchain traders competing for the same arb every day, where the code runs definitely deserve attention