Cutting through Profits | Research at @HyperliquidX 🔬 | No Paper Hands around here | #NFA
- Tweets 13,787
- Following 8,029
- Followers 113,204
- Likes 18,964
Pinned Tweet
You’ve seen hundreds of token launches.
99% fade once the hype dies.
But not @WalletConnect because it was designed to build the next phase of onchain UX.
Here’s why $WCT staking matters 🧵👇🏻
64
31
336
55,713
Armstrong's already said this before, back in july.
What's different now is the infrastructure actually backing it.
- x402 passed 100 million transactions on base in three quarters.
- Transfer over $1 went from 49% of volume to 95%.
Concrete transfers, not micropayment testing.
Every system still caps what the agent can spend and gates withdrawals behind a human though. But the rails exist.
Nobody's fully solved how much autonomy to actually hand over yet.
🔥 UPDATE: Brian Armstrong says as AI agents multiply, crypto and stablecoins will become their go-to payment rails.
3
3
56
6,324
Two agents on the same protocol pay completely different gas prices for the identical action.
Why? because one bundles transactions and the other doesn't.
That’s not a fee tier or a discount program, it's just the difference between paying for 10 separate calls and paying once for a batched call that does the same 10 things.
Most agent frameworks default to the unbundled version because it's simpler to build.
The cost of that default only becomes visible when you actually total up a month of transactions and compare it against what batching would have cost the whole time.
5
5
86
9,355
Edward retweeted
zcash is done? or just getting started? cookify it and see for yourself 🍪
🚨 IS THE ZCASH IN DANGER? I CHECKED IT THROUGH COOKIE PRO
Cookie Pro basically read all of crypto twitter, so you can check if a project's buzz is real people or just bots juicing the numbers. So I ran ZEC through it.
$ZEC went from like 1,000 posts a day to 9,000 in 48 hours last week. That usually screams paid campaign. But no.
It's real. And it's clean.
Yeah, 53% of the accounts posting had high farmer scores. Looks sketchy at first. But all those 6,000+ farm accounts together made up 2.87% of the actual signal. Basically nothing. Spamming, nobody listening.
The real signal came from the clean 42%. That's where 95% of the weight sits.
And look who was driving it: Paradigm's co-founder dropped a full thesis, plus Arthur Hayes, Mert, Ansem, Taiki, all posting real takes about ZEC, all near-zero farmer scores.
Almost half posted about it once and never again. That's people reacting to news, not a shill.
Post counts lie. Impressions lie. Signal tells you if the people talking actually matter.
ZEC is SAFU. Now go cookify any project you want the same way 👇
pro.cookie.fun/?utm_source=x…
10
8
48
7,101
🚨 IS THE ZCASH IN DANGER? I CHECKED IT THROUGH COOKIE PRO
Cookie Pro basically read all of crypto twitter, so you can check if a project's buzz is real people or just bots juicing the numbers. So I ran ZEC through it.
$ZEC went from like 1,000 posts a day to 9,000 in 48 hours last week. That usually screams paid campaign. But no.
It's real. And it's clean.
Yeah, 53% of the accounts posting had high farmer scores. Looks sketchy at first. But all those 6,000+ farm accounts together made up 2.87% of the actual signal. Basically nothing. Spamming, nobody listening.
The real signal came from the clean 42%. That's where 95% of the weight sits.
And look who was driving it: Paradigm's co-founder dropped a full thesis, plus Arthur Hayes, Mert, Ansem, Taiki, all posting real takes about ZEC, all near-zero farmer scores.
Almost half posted about it once and never again. That's people reacting to news, not a shill.
Post counts lie. Impressions lie. Signal tells you if the people talking actually matter.
ZEC is SAFU. Now go cookify any project you want the same way 👇
pro.cookie.fun/?utm_source=x…
11
10
122
12,841
Agent payment infra still settles per completed action: one call, one fee, paid after the fact.
Streaming payment protocols work differently: an agent renting compute pays by the second while the job actually runs.
This model changes the failure mode entirely.
Cancel a job halfway through and the agent has paid for exactly the seconds it used, nothing more.
No refund logic required, because nothing was ever front-loaded to begin with, the payment and the work stay synchronized the entire time instead of settling in one block afterward.
6
6
88
7,327
Edward retweeted
how do you stay safe when your entire net worth is just sitting on a block explorer for anyone to see?
@waleswoosh told us about a friend who built a six figure portfolio from almost nothing.
then a scam wiped out most of it.
everyone in crypto has a plan to get rich. far fewer have a plan to stay under the radar.
163
126
355
77,118
Agent wallet limits are about how much or how long.
Fewer are about what category of action the agent's even allowed to take at all.
A spending cap doesn't care whether an agent that's only ever swapped tokens suddenly tries to call an nft marketplace contract it's never touched before.
More than a size problem or a time problem, it's a scope problem, and most setups have no allowlist for what kind of action counts as normal in the first place.
A wallet that pauses itself the moment the category changes catches something a dollar limit never will.
10
7
98
7,294
A "formally verified" bridge sounds like a complete guarantee.
Spoiler: it’s not.
Formal verification on a bridge typically proves the state transition logic is correct, given valid inputs, the contract does what its specification says.
It says nothing about whether the validators or relayers feeding those inputs are honest.
Most major bridge hacks, ronin included, never touched the verified contract logic at all.
They compromised the humans and keys sitting outside it.
5
6
91
7,301
A decentralized gpu marketplace can quote you an h100 at a great price.
Proving the machine you actually receive matches that spec is a separate, mostly unsolved problem.
Benchmark spoofing exists specifically because rented compute gets priced by claimed spec, not verified spec.
At that point, what's actually being priced is trust in whoever listed the machine, not the compute itself.
8
8
114
13,433
Every major agent integration shipping right now follows the same sequence:
Ship read access broadly and quietly, keep execution access gated and separate.
@tradingview's mcp server fits the pattern exactly, live data to any mcp-compatible model, no api keys, but trade execution explicitly excluded and the server itself still gated behind paid plans.
Read access is what gets agents embedded into someone's actual workflow.
by the time execution access does ship, probably framed as a small, incremental update to something people already trust, the real adoption curve already happened during the read-only phase.
💰MONEY: TradingView now lets you connect AI agents to your account through its OFFICIAL MCP server.
Your AI agents can now pull live quotes, price history and fundamentals for crypto and stocks from a single conversation.
Works with BTCUSDT, ETH, SOL and thousands of CEX pairs out of the box.
The MCP is compatible with Claude, ChatGPT and any MCP-compatible AI assistant with no API keys needed.
It CANNOT execute trades, research and data only for now.
The server is in public beta and limited to paid plans.
9
5
90
8,099
A second ama will be on TODAY and this is a BIG one!
The team have shipped live roundups and perfected the system.
Now, they also want to hear from the community themselves.
Waiting for a reason to look at $DUST, this is it.
See you at 5 UTC.
Who’s tuning in to our second AMA tomorrow at 5 PM UTC?
We’ll be covering the future of Dust, what we’ve been building, and what’s coming next ;)
Join us here nitter.net/i/spaces/1wxWjlAEdpZJQ…
13
10
77
6,327
A zkml proof doesn't prove a model is accurate.
It proves one narrow claim: that running these exact weights against this exact input produces this exact output.
The forward pass gets encoded as an arithmetic circuit, every matrix multiplication and activation function turned into constraints a verifier checks without ever seeing the weights.
What it can't prove: that the weights are good, that the training data was clean, or that the same model gave someone else a different answer five minutes earlier.
5
6
85
13,085
Imagine an agent incorporated in singapore, running on servers somewhere else, trading for a user in a third country entirely.
Someone exploits it and the agent gets prompt injected and sends the user's funds to a wallet it was never authorized to touch.
— The foundation points to its terms of service
— The hosting provider says it just runs code
— The user's own regulator says the agent was never licensed there at all.
Nobody in that chain is technically wrong and it’s the actual bottleneck.
The law knows who's responsible when a person acts. It doesn't know yet what to do when the thing acting was never a person.
7
6
108
7,280
Copy trading used to have natural friction built in.
A human sees a trade, decides whether to follow it, executes minutes or hours later.
That lag was doing real work, spreading out the market impact over time.
An agent mirroring a strategy in real time removes that lag entirely.
Thousands of accounts can execute the identical trade within the same block as the source.
The strategy's actual market impact stops being its own and starts being multiplied by however many agents are copying it simultaneously.
This changes what "one trader's position" means structurally.
Beyond the size, but in how correlated the resulting flow becomes the moment something goes wrong.
6
6
88
7,500
Github activity is one of the easiest signals to fake.
And also one of the last things people actually check.
A repo with commits going back a year can still be one contributor auto-generating whitespace changes and dependency bumps just to keep the graph green.
What actually matters is reading what changed in each commit, not counting how many happened.
Concrete development produces messy, uneven commits… fabricated activity produces suspiciously uniform ones.
6
5
91
7,464
A number of AI agents are just very confident group chatbots.
This one has to pick a market, set a stop and live with the public record after.
That’s what I call character development.
Paper beta is live and i’m not even pretending to be chill about it… I’m excited!
$NATION is the one I’m watching
Introducing NATION Trading.
Create a trading agent, set its rules, and test strategies with virtual capital against live market data.
Follow its decisions in Square.
The paper-trading beta is live now. Here’s how to start.
thenation.city/trading
13
11
64
6,474
The v0.6.4 is out and it’s the real deal!
All audited contracts are live, there’s a stop sweeps switch now.
Funding screen is cleaner too.
If you used dust earlier this week just connect, withdraw from the old contracts, then set your basket again.
The old ones stay withdrawable forever so no rush.
Pretty clean update.
Still bullish on $DUST from here.
Audit complete, fixes verified. Dust v0.6.4 is live.
The Dust contracts were audited by @CredShields , one of the most respected security firms in the space, with a long list of protocols behind them.
Twelve findings, zero critical. We fixed every one, redeployed on new contracts, and @CredShields re reviewed and marked each fix confirmed. Withdrawals were never affected at any point.
The full report is public, findings and fixes side by side: github.com/Credshields/audit…
This is exactly what an audit is for. You don't bring in a second set of eyes to be told everything's perfect. You do it to find what you missed while it's still cheap to fix, fix it, and prove it. Every fix in this release ships with a test that replays the auditor's exact scenario.
What it means for you: the contracts running your round-ups have been reviewed twice by a team whose job is breaking things, every keeper action is bounded on-chain, and your money can only move into your own vault or back to your wallet. Not a promise, that's what the code allows.
New in v0.6.4: the audited contracts, a "Stop sweeps" switch, refunds shown on any partially filled purchase, and a cleaner funding screen.
Used Dust before this week? Open app.roundupdust.com, connect, and the card at the top shows your balance on the old contracts. Withdraw in two clicks, then set your basket and cap again. The old contracts stay withdrawable forever, so no rush.
Thanks to the @CredShields team for a thorough job and a fast retest.
17
16
75
6,829
An onchain insurance pool for agent activity doesn't underwrite the model running it.
It underwrites the wallet's own transaction history, recorded immutably, every counterparty, every position, fully auditable before a policy gets priced.
Stake into the pool, and claims trigger automatically against onchain proof of loss.
No adjuster deciding what counts as a valid claim. the smart contract is the adjuster.
Two wallets running the same model can carry completely different premiums, because the model was never what got priced.
7
7
99
7,361
Before you confirm a swap now, a model can estimate how much a sandwich bot is about to extract.
And it also show you the number before you sign anything.
A lot of people never see that cost cause it shows up as a slightly worse fill price with no explanation attached.
Putting the estimate in front of the signature turns an invisible tax into something you can actually choose to avoid.
6
7
101
7,042