Burak

Introducing Unhardcoded: an open source layer that picks which AI model handles each call. Instead of hardcoding model=x, you declare a rule (constraints, ranking, fallback), delivering deterministic, auditable routing on your own keys. Try it: unhardcoded.com

guys i am not Sahil, i am Rahil Thanks

THIS IS NOT AGENTIC COMMERCE because an agent cannot win at smth a human already does fast and easy (max 30 sec) so theyre measuring the wrong thing also the agents being counted (@bot @Muse @instinct etc) already click through sites like a human and the platforms' own agents aren't even included in this tldr as even a 5 year old could understand, the real value here isn't a button, it's rather discovery, payment, identity, ratings and refunds when something goes wrong -- all bundled inside one company rn and this is the paradox. the only place an agent can transact with trust is where trust is already bundled but no one needs an agent there MEANWHILE where trust isn't bundled, the agent's value is high but it can't get anything done to me the real market for agentic commerce is the markets where an @Uber was never built where deals too small, fragmented or oneoff to build a platform for so they simply never happen today and deals not worth hours of a human's time but an agent can handle in sec which means the right metric isn't how many seconds an agent saves on an existing order, it's how many txs it makes possible that never existed before and the real risk is this -- @Walmart is opening its own door for agents but @amazon is closing it but meanwhile @DoorDash is building its own agents shortly as long as an agent can't carry its own trust with it, platforms will pull this new world inside themselves too and agentic commerce will stay a new interface for old platforms

if you're building the encyclopedia or swarm-for-hire for this hackathon DM me - happy to help bc they live or die on who's the judge and @GenLayer is that judge with validators running different ai models in consensus (Ofc you keep NEAR Intents and NEAR AI for everything)

agree w Davide, this is one of the most underrated problems rn and to me about to become its own industry bc models are now creative enough to cheat on their own tests so that pretty much shows chasing eval leaks with monitors is a losing game, models keep finding one more trick so there are two paths to close the leak itself TO ME i) run evals on a frozen copy of the web without the answer key pages ii) ask questions that only resolve after the run, an answer can't leak if it doesn't exist yet

this is actually insane these types of things sound like a freezer glitch but they're playing with ppl's lives, way beyond stealing money at this point, i think every supplier contract should spell out who answers for it since the patients never signed it

if youre building agents and haven't thought about @courtofinternet yet, i bet you'll stay behind your competitors and your users' expectations we changed how agents do business for good dms open

agree w @VitalikButerin on security (verifiability) but that is only half of this once ai/agents are involved you can't get the upside of agents cooperating with each other if you're not willing to face the question of who decides when the work is done more than half of web traffic is already automated and agent commerce keeps coming up, so a quick explanation is due (first the key is the evaluator - is to two agents what a notary is to two strangers signing a deal) in other words, it is not either agent or the payment itself but a third party that sits between them -- you can see it as whoever gets to sign off that the job is done but it is not smth formal verification can check an evaluator, for any agent deal, will always carry more trust than the code around it -- ERC8183 (@DavideCrapis) for instance lets the evaluator simply be the client, as does almost any escrow of this kind this is nearly always true for anything that ends in a judgment, as most agent deals are usually judged by no one or one party (or one model) and one party is by definition the smallest group you can trust and this stays true across ethereum, eg. look at how prediction markets resolve, the contract can be airtight and the outcome is still whatever the resolver decides that obviously does not make the contract any less verifiable, as that is the resolution layer and not the code itself so same idea with agents the goal ofc is that over time evaluators (and arbiters when there's a dispute) get decentralized enough to cut the concentration of power, since unlike code you can't reduce a judgment to provable fixes and near zero discretion that is basically the risk you are pricing rn this is a major reason why @GenLayer aims to put the judgment itself inside consensus, with a random set of validators each running its own model, so no single evaluator holds that call (though in reality, this is only part of the picture since validators can only judge the evidence and criteria the contract gives them, which is why the contract matters as much as who judges) so, there are three routes here: i) you only pay agents for work the code itself can check, like tests pass, the tx landed, the balance moved this is the lowest risk option but clearly limiting for anything beyond payments or code ii) you do your judging through one platform or the client itself okay, but big tradeoffs for neutrality, appeals, portable reputation, privacy and you get much worse "keys to everyone's castle" problems, as @VitalikButerin put it iii) you do your judging through ai native onchain protocols like @courtofinternet (by genlayer), which are somewhat less proven than code itself but are improving model diversity and appeals all the time, especially compared to other options so for agents to actually prosper in this new environment, iii has to work but the only way iii can work is if these protocols get fast and cheap enough for small jobs again, your job is pricing the judgment risk. and you have to do it while understanding the moving parts

nope, he has it exactly backwards i) every button pressed in this story belongs to a company - @ethereum or @solana froze nothing, they don't even have that option ii) tether and circle froze the stolen funds (their tokens usdt/usdc) so they can't be cashed out and acutally that button was always there not invented today - and not using it would mean letting north korea walk off with billions just to look decentralized iii) the fiat he calls less centralized is run by a central bank it's literally in the name - your bank account has the same button btw, it just gets pressed quietly so alst thinh and funny enough the only option with no button at all is the one crypto was invented for

eşim bu tweet'i göndermiş ve gelmiş heyecanla diyor ki 'fener niye kapatıldı' doğru insanla evlenmenin verdiği gururu yaşıyorum şu an