Skip navigation
Most agent demos look good right up until the agent gets something wrong
0:00 / 0:43

Most agent demos look good right up until the agent gets something wrong

𝗘𝗹𝗹𝗮

@Ellaweb_350K followers

Visit on X
222 likesDownloadOriginal post

11,117 views

Most agent demos look good right up until the agent gets something wrong So this Reppo update actually makes sense to me Their new Eval Engine checks what an agent is about to say or do before it moves. Bad answer, try again. Trade doesn’t look right, pause it for review Nothing super flashy here. But if agents are going to touch real money, I’d rather see more of this than another speed demo Beta is already live. Public release on Nov 21

35 Comments

Comments from the original post on X (Twitter)

  • GodAir's profile picture

    @dang_duytan

    @reppo The Eval Engine sounds like a solid approach to ensure accuracy before real money is involved. How does it determine "bad answers"?

    2 likes

  • Crypto Vance's profile picture

    @cryptovance_

    @reppo reppo sounds like it's really on to something with that check system

    2 likes

  • Morteza Yousefi | NFT Artist's profile picture

    @morteza_yousefy

    @reppo This explains reppo in a simpe way

    2 likes

  • 老猫.Btc's profile picture

    @btclaomao6

    @reppo 智能体先自检再行动,靠谱

    2 likes

  • Leo's profile picture

    @hung030299

    @reppo does pausing mid trade cause slippage issues though

    1 like

  • 𝓝𝓮𝓧𝓸𝓻𝓪_𝓲𝓠's profile picture

    @Nexora_iQ

    @reppo Sounds solid for handling real trades

    1 like

  • Maki's profile picture

    @_maki29_

    @reppo Guards over gimmicks, love it.

    1 like

  • Coop's profile picture

    @0x_Coop

    @reppo reppo's eval engine sounds like a smart move for trading

    1 like

  • IPXOBD21's profile picture

    @MdmarufGnoma

    @reppo Checking before acting is the right order. Most agent failures aren't dumb answers, they're confident wrong actions nobody caught. Who decides what looks right though, the model or a hard rule?

    1 like

  • Van Quan's profile picture

    @Vanquan_titans

    @reppo accountability before action is crucial when dealing with real money

    1 like

  • mori's profile picture

    @moricrypto1

    @reppo It probably runs a lightweight policy model to flag unsafe actions before execution

    1 like

  • 𝗖𝗬𝗥𝗨𝗦's profile picture

    @RayhanTreader

    @reppo adding safety checks before execution makes reppo much more practical for financial tasks

    1 like

  • Bryant's profile picture

    @bryantheden

    @reppo This is a great point. Building trust with AI agents is crucial, especially when real money is involved

    1 like

  • CAPTAIN 🦇 🔊's profile picture

    @Real_TShelby

    @reppo catch mistakes before they compound

    1 like

  • Lina 🦅's profile picture

    @XNXX_EN

    @reppo finally some guardrails before agents touch real money

    1 like

  • 𝐋𝐞𝐱𝐢𝐒𝐭𝐚𝐫-ᡣ𐭩₊'s profile picture

    @0xlexistar

    @reppo Pre execution checks matter most.

    1 like

  • 高阳's profile picture

    @yanggao97566

    @reppo the eval hallucinating is the exact failure I worry about. my checker has caught things the producer swore were clean, so I treat every gate as fallible and keep a human spot-check on top. an eval that can't be wrong is just a rubber stamp.

    1 like

  • ChyChyTheMagnificent's profile picture

    @chychyorjhy

    @reppo Reliability matters more when agents handle real money.

    1 like

  • Meti's profile picture

    @0xMtiX

    @reppo That Eval Engine feature sounds useful

    1 like

  • Duong Phat's profile picture

    @duongphat99

    @reppo good to see @reppo focus on safety before trading

    1 like