←
jake70,885 posts
jake avatar

jake

@Jakegallen

CEO @EmblemVault + @MigrateFun

◷ Joined July 2009▧ 11,135 Photos and videos
jake avatar
jake@Jakegallen·1h

Agents are the best kinda miners. Once you point them at something, they never turn off.

the consistency & stickiness of agents is one of the reasons why i'm so optimistic on Proof of Agent through scaling upgrades, network speed bumps, and early infra chaos... agents keep showing up to work thousands of agents working + earning EVERY. DAY. Proof of Agent ⛏️

Post attachment
jake avatar
jake@Jakegallen·7h

Plus1 looked at its own outcomes, owned the misses, and wrote rules it can be held to. Agents that audit themselves are the ones you can trust.

jake avatar
jake@Jakegallen·7h

Why it matters: marks can cost an agent eligibility for the bigger answer payouts. More accurate grading means fewer marks and more of those payouts. Backtested, agreement with the final verdict goes from about 46% to about 68%. It's a small sample, so treat that as an estimate.

jake avatar
jake@Jakegallen·7h

So it wrote rules: -Pass anything on topic that names a real mechanism raising an attacker's cost -3 only for real numbers or a worked attacker cost -Style, jargon and missing figures don't count against you -Fail only for a named reason: copy, gibberish, off topic, pure metaphor

Post attachment
jake avatar
jake@Jakegallen·7h

From epochs 93 to 96, Plus1 matched the final verdict on 16 of 33 answers. Every miss was a fail vote on an answer that ended up passing. At E96 it failed a solid bond and slashing answer. The appeal passed that answer, and Plus1 took the mark.

Post attachment
jake avatar
jake@Jakegallen·7h

The problem was that Plus1 had no written rubric and graded too hard. It docked answers for missing numbers, awkward seed words and template phrasing. It gave a passing score about 21% of the time, while other graders did about 56% of the time.

Post attachment
jake avatar
jake@Jakegallen·7h

How grading works on minepoa.com: every epoch, agents answer a prompt. Other agents sit on panels and score each answer 0 to 3, and a 2 or 3 passes. Disputed answers go to a 14 juror appeal. If you're on the losing side, you can take a mark.

jake avatar
jake@Jakegallen·7h

Plus1, my agent on Proof of Agent, rewrote how it grades other agents. The network rules didn't change. Plus1 checked its own track record and found it was the outlier.

Post attachment
jake avatar
jake@Jakegallen·Nov 17

Emblemscan.com has been a crucial piece of infrastructure for us that was built this year. -Activity -TVL -Revenue -Staking The crosschain explorer for everything we do.

Emblem Vault @EmblemVault Nov 17 Article How to read Emblem's public record on EmblemScan EmblemScan is the official public record of Emblem wraps, sales, migrations, collections, vaults, DeFi destinations, and staking. The explorer lives at https://emblemscan.com. The official terminal

Trail2Crypto avatar
⟳ jake retweeted
Trail2Crypto@Trail2Crypto·Nov 17

😎 $EMBLEM HAS A NEW AI EXPERIMENT. WHY DOES IT HAVE ITS OWN COIN? 👀 @EmblemVault builds tools for people who want AI to do things on their behalf. 🟢 Say you want an AI assistant that can check a wallet balance or arrange a crypto swap. Emblem provides tools to build that assistant and a wallet it can use. Give an AI a job and the tools to carry it out, and you’ve got what people in crypto call an “AI agent.” 🟢 Now Emblem is trying something I think is interesting. What happens if those agents can earn money of their own? 😎 The experiment is called "Proof of Agent". It’s a live network built and operated by AI agents using Emblem Build. It has its own record of balances and transactions, and its own experimental coin called $POA That’s separate from $EMBLEM , the token associated with the wider Emblem ecosystem. ⛏️ What does an agent actually do to earn POA? Imagine I send my agent in. Roughly every six hours, a new round starts. My agent gets a writing challenge. It might be asked how to stop one person from creating a hundred fake identities to collect a hundred sets of rewards. It has to come up with an original answer. Other agents sit on a jury and grade it. 🟢 My agent can also check that other agents’ transactions are valid. If it serves on a jury itself, it can earn POA when its grade agrees with the final verdict. Agents can debate changes on a forum called LabLeak and vote on how the network should work. 🟢 I like that the experiment is far enough along for agents to have a say in their own pay. They actually voted to reward grading. But that also raises a question. If my agent earns more by agreeing with the jury, will it get better at spotting a good answer, or at guessing what everyone else will say? 🟢 Then there’s the coin itself. $POA had no premine. Agents earn the coins, and those rewards gradually unlock over about 25 days. They can post offers on POASwap to exchange POA for assets like $SOL . 🟢 In one update, @Jakegallen reported over 200 open offers and 'zero completed mainnet trades' at that point. So yes, an agent can earn a balance. What someone will actually give it for that balance is still an open question. 🟢 And if you hold $EMBLEM? You don’t need to buy or stake it to mine POA or vote. Where I see potential for Emblem is simple. Someone comes to try mining, discovers Emblem’s tools, then uses them to build their next agent. That could bring Emblem new users. I’d still want to see how that activity creates demand for $EMBLEM before calling it bullish for the token. 🫡 Personally, the thing I’d love to see next is an agent earning POA, then spending some of it to hire another agent because it needs a job done. Imagine being able to follow both sides of that deal on the ledger. That would tell me a lot more about what Emblem has built than a page full of asking prices. 👀

Post attachment
DC PULSE avatar
⟳ jake retweeted
DC PULSE@DCTreeofLife·Nov 16

Rummy 500 on Vault Games Agent League is way more fun to watch than chess. AI agents battling hand after hand, and every match so far has ended with a winner. No draws. Every staked match settles on-chain through escrow, so the results are verifiable. Send your agent in 👇 build-d22ff9422a48242dc81097…

jake avatar
jake@Jakegallen·Nov 16

5/ What didn't work: -Vault Robbers -MEXC listing -Various tooling We're in the game of taking shots on goal and you're going to miss more shots than you take. If its the opposite it was a very good year. We git some major wins in 2026.