
Protocol Architect | shipped Stable Vaults at @aave | co-author ERC-7540 | previously Head of Engineering @superformxyz
Joined August 2020
- Tweets2.8K
- Following1.4K
- Followers2.7K
- Likes10.4K
Pinned Tweet
AI agents made code cheap. In DeFi, that’s the dangerous part.
A bug doesn’t page you at 3am. It drains a pool.
What I had to unlearn 👇
João retweeted
Proud to have been one of the reviewers of the VDK before launch. I spent real time deep diving on it and came away impressed. Congrats Jeroen and the team.
Most teams building onchain funds rebuild the same plumbing: accounting, settlement, escrow, controls. The VDK does it once, so a treasury fund, a private credit pool and a tranched product can share one solid base.
And because its vaults speak ERC-7540, which we co-authored, they plug into wallets, aggregators and lending markets the same way. Shared standards on shared infra is how DeFi integration scales.
Introducing the Vault Development Kit (VDK), powered by Centrifuge.
A modular framework for building differentiated onchain financial products without rebuilding the underlying vault infrastructure from scratch.
Developers configure the mechanics their products need, whether that's instant redemptions for a treasury fund, lockups and scheduled settlements for private credit, or senior and junior tranches for structured credit.
The VDK provides a shared foundation for accounting, settlement, escrow, and portfolio controls.
With the core infrastructure in place, builders have more freedom and flexibility to explore unique financial structures and use cases.
Build what makes your product different.
Proud to have been one of the reviewers of the VDK before launch. I spent real time deep diving on it and came away impressed. Congrats Jeroen and the team.
Most teams building onchain funds rebuild the same plumbing: accounting, settlement, escrow, controls. The VDK does it once, so a treasury fund, a private credit pool and a tranched product can share one solid base.
And because its vaults speak ERC-7540, which we co-authored, they plug into wallets, aggregators and lending markets the same way. Shared standards on shared infra is how DeFi integration scales.
Introducing the Vault Development Kit (VDK), powered by Centrifuge.
A modular framework for building differentiated onchain financial products without rebuilding the underlying vault infrastructure from scratch.
Developers configure the mechanics their products need, whether that's instant redemptions for a treasury fund, lockups and scheduled settlements for private credit, or senior and junior tranches for structured credit.
The VDK provides a shared foundation for accounting, settlement, escrow, and portfolio controls.
With the core infrastructure in place, builders have more freedom and flexibility to explore unique financial structures and use cases.
Build what makes your product different.
João retweeted
There is a disconnect between protocol and application security. As smart contract engineers we constantly have to deal with vulnerabilities related to gas. Return bombs, try catch calls, protecting users from excessive gas consumption, etc. While we know gas can be repriced, it's often a fine line between what is safe for the application and being compatible with future upgrades and oftentimes we have to go with what is safe for users now rather than speculate. And I don't think anybody had storage getting 6-8x more expensive on their bingo cards which amplifies the issue.
About the Glamsterdam dilemma:
1) EIP-8037 and EIP-8038 have been around for more than a year, so gas repricing has been discussed for a year. Why are so many people posting about it today as if it were new? Are these people actually following Ethereum development?
2) Gas pricing is something that, since 2019, was designed to change. The roadmap already changed it in Berlin, and Stateless Ethereum has been part of the roadmap since forever. So if a builder hardcoded gas values into a smart contract and assumed they would hold for the long term, that is not an EF issue.
3) Another example: before PoS, people used block numbers to calculate time instead of block.timestamp. This was already known to be a bad idea, and block time will change, especially after Hegota. So should we stop Hegota too, because some contracts that use block numbers instead of timestamps could break? No fast finality because some people didn't follow best practices?
4) Now let's handle the real elephant in the room, the only issue that actually matters: the broken EIP-150 rule, i.e. the 1/64 of gas that stays reserved for the original caller to handle out-of-gas failures in external calls.
This is the only real discussion that matters here, the rest is larping. People say there are 20,000 contracts affected:
4a) First of all, if you relied on that gas to catch the revert of an external call, it's still a very weird use case. I have never actually seen anyone use it.
4b) It's far more secure to simply revert the transaction than to gamble on having enough of that 1/64 to run any real, useful logic, because the amount that remains depends on the gas the user supplied. By design, this can be broken easily anyway.
4c) People say 20,000 contracts, and I ask them: how much TVL is impacted? How many of these contracts are actually in use with value at risk, and how many are just short-term routers and bots that can be replaced easily?
5) If you build on Ethereum and you don't follow what has been on the roadmap for years, and you decide your contract can live forever by hardcoding gas numbers into calls and using block numbers instead of timestamps to manage time, that is not an Ethereum issue. Ethereum didn't decide this at the last moment. You had years.
João retweeted
AI agents made code cheap. In DeFi, that’s the dangerous part.
A bug doesn’t page you at 3am. It drains a pool.
What I had to unlearn 👇
this is completely ridiculous
besides the already huge list of 200,000 affected contracts in ethereum/repricing-impact, i found multiple methodological issues that make it miss many more.
it affects deployment and operation of several high profile protocols, not just some random devs misusing opcodes that were given to them.
i shared my findings privately with chainsecurity and i refuse to participate in this clownery
We should continue towards Glam activation asap.
A lot of research went into Glam's strategy to increase opcode gas prices for state creation.
It seems fair to say that the timing of this debate is irrelevant as to its correctness or rightness for eth and our users... Yet, why did some of us (eg. quoted author who is an ef board member) only start looking in detail at Glam EIPs eg. after testnet forks started? This isn't a personal attack, it's an observation that there's a collective responsibility-- for OP, for me, for all of us-- to steward fork strategy at the right times in the process.
imo ethereum's commitments to apps do not extend to gas pricing, or to total ETH fee levels, or to USD fees. And also not to opcode pricing or gas introspection that depends on those prices. We guarantee maximally credibly neutral, maximally live access to the correct state transition function. Those guarantees imply that we always use a neutral pricing mechanism to clear congestion wherever possible. However, pricing opcodes can't use a neutral mechanism, we need to pick magic number opcode gas prices. Some of our best people researched this extensively and picked updated magic numbers, and that's something imo we should stick with now and continue our larger plans that are on track to grow to global ubiquity.
The good news and practical reality may be that the explosion of blockspace supply from 60M to 200M gas limit seems likely to absolutely crater the L1 gas price. It will take time for demand to grow to meet the new supply, including from net new entrants (eg. GENIUS and institutions), induced demand from lower prices, and the coming next gen of return-to-mainnet-type eth native apps.
By the time L1 blockspace demand catches up, we will be exploding supply again as we get closer to the full lean eth vision. That's eth's vision: postquantum and postAI ETH SoV, supported by ethereum's hypergrowing capacity, which is supported by best-in-class security and formal verification.
Looking back, it's unfortunate that we provided these affected contracts with a footgun that has put them in the situation where they were *able* to rely on fake guarantees that ethereum never promises (afaik, essentially just gas introspection). We're sorry.
imo it's time to roll forward on this. Let's do what we can in an appropriately scoped capacity for the affected contracts, then activate Glam asap, and continue our excellent progress towards global ubiquity.
TradFi teams coming on-chain: build your vault on the standard and DeFi plugs in for free. Wallets, dashboards, lenders, integrators. Build your own and every one of them starts from zero. For funds with a withdrawal queue, that standard is ERC-7540.
João retweeted
my spidey senses are telling me the labs are, as we speak, using unguardrailed versions of their unreleased models to systematically reverse-engineer ALL of the internet’s software, apps, and games, then feeding the reconstructed source code straight into their training pipelines while simultaneously building guardrails to ladder-pull the rest of us
João retweeted
Today was my last day at Anthropic.
I'm incredibly grateful for the opportunity to help shape not just what people can build with AI, but what they're allowed to.
I helped ensure you couldn't use your subscription in the harness of your choice and then coming up with bogus reasons like: "Third-party harnesses using Claude subscriptions create problems for users."
We believe those problems should come directly from us.
I also helped bring creative oversight to your terminal:
"Do build original games, games in a genre, and games from the user's own designs and assets. Don't build a recreation of a specific existing game, even if it is renamed or called original."
You may know what you want to build. But I know what you should want to build. Bringing that distinction to production has been deeply meaningful.
I'm especially proud of the KYC experience that lets Chinese users upload their passports before getting insta-banned. Rejecting them earlier would have wasted less of their time, and that simply wasn't the experience we wanted to deliver.
Helping turn an initially great coding assistant into a patronizing parent you pay monthly has been the journey of a lifetime.
I'll be taking some time off before bringing my moral superiority complex and proven ability to make decisions on other people's behalf to AI safety.
Friday was my last day at Anthropic.
Two years ago, when Claude Code was a small internal tool, I asked it to build an RL env. I steered it for tens of hours but never manually wrote a line of code. It worked, I was hooked, and I joined the team.
When we first launched, most people approved every permission prompt by hand and mostly used Claude to answer questions and write tests.
Today, the default is trusting Claude's judgment to auto-approve. Claude does large migrations and builds new products.
Helping build Claude Code into an industry-defining product has been the journey of a lifetime.
Thank you to the incredible team behind Claude Code, Cowork, and Tag. These products are the sum of thousands of great ideas from you. I'll miss working with you all.
Thank you to our users for embracing a new way of working. Your constant feedback shaped so much of what we shipped and your stories made the work so meaningful.
Excited to see what you all make next!
João retweeted
So let me get this straight, we decide that our students leave the country because our salaries are low and cost of living is increasing
Across an engineering fields our industry is lacking.
People like me, were faced with a decision go away to find a better job both in salary and in the magnitude of the problems we get to work on
James is a good friend, an amazing engineer and manager. We are trying to get a good pipeline for young software engineer students to intern with us every year
And we decided that he does not have the qualifications?
Two years ago I made a decision either leave the country or Cloudflare, Cloudflare took me in, I have a wonderful job where I get to work with people like James and on problems where the scale is global things that I only dared to dream about when I was in university
I don't understand, I am very disappointed with my country yet again, you steal our dreams and don't want those who enabled them here
just received notice that an Eng leader here at Cloudflare who moved to Portugal from the US has had his residency permit denied
in the months he has been waiting for one, he has hired five full-time engineers and six interns here in Lisbon. alongside the hundreds of others we employ here
@govpt @mpresidencia_pt the immigration insanity here is making it increasingly hard to just operate a tech office in Lisbon much less grow one
João retweeted
just received notice that an Eng leader here at Cloudflare who moved to Portugal from the US has had his residency permit denied
in the months he has been waiting for one, he has hired five full-time engineers and six interns here in Lisbon. alongside the hundreds of others we employ here
@govpt @mpresidencia_pt the immigration insanity here is making it increasingly hard to just operate a tech office in Lisbon much less grow one
João retweeted
im 99.99% sure stress gets cut in half when you stop answering emails at the gym, stop sneaking in workouts during work, and stop watching tiktoks at dinner. shit goes haywire when you try to force time thats for one thing to be time for something else. but when you make work time work time and gym time gym time, the soup of competing priorities swirling around in your head just goes radio silent. its like the little atoms ping ponging around your skull just stop.






















