Bonded checkers is a nice twist. It puts a cost on rubber-stamping, which is how most multisig setups actually fail.

Keeping heavy crypto off the main thread makes sense. The tricky part is giving devs a clean way to reason about what got verified where.

Verifiable agent logs feel inevitable once agents can move money. “Trust the agent” won’t survive the first incident.

Payroll is where this clicks for people outside crypto. Once salaries move onchain, public balances stop being a feature.

Planning for hacked collateral upfront instead of improvising mid-incident is the right instinct. The dusting case is one most lending designs never write down.

Hooks becoming an integration layer, not just pool customization, is a cool direction. Curious how routing deals with wrapped sources that have very different latency.

Matches what we see. The contracts are tighter, and incidents keep moving to keepers, frontends and whatever signs the transactions.

Splitting native Rust verification from the onchain verifier path is useful context. Most people saw “SP1 bug” and assumed the worst.

Korean securities onchain would be a big one given how active retail is there. Interested to see how the compliance layer ends up looking.

Looking forward to the report. Adoption outside blockchains is the hardest part to track, so that section should be useful.

Math-heavy code is exactly where AI tooling gets overconfident. Would love to see the slides after.

Small change, big UX win. Nobody wants to work out whether their renewal got pricier because ETH moved.

Good that it’s research first. The bridge and existing-account migration seem like the hardest part if it goes ahead.

Corporate actions are what get people. A market that prices a stock token like any ERC-20 is fine right up until a split or a trading halt.

Governance keeps being the softest layer. Doesn’t matter how clean the token contract is if treasury control can be taken in a couple of txs.

CTFs built from real findings are the best way to spread that knowledge. Very few people are looking at Daml compared to the EVM, so this is timely.

And unlike a key, you can’t rotate it

Centimeter positioning is the difference between a robot mower and a robot that eats the flower beds.

How do you handle someone filming a screen that’s playing an existing video of the place?

Self-serve API keys is a quiet but big one