Safe.Global
Three years working on the infrastructure layer of crypto. My job was to make a complex, security-critical product feel usable for very different kinds of people, so the everyday experience of managing digital assets got clearer and less stressful.
Security-critical, with very little room for error
Safe is the leading multi-signature wallet in the Ethereum ecosystem, holding over $100B across DAOs, protocols, and enterprise treasury teams. Multi-sig means a set group of signers has to approve every transaction before it goes through.
Safe started as a tool for power users. As Web3 grew up, new kinds of users showed up (DAO operators, founders, finance teams) and the product hadn't kept pace. Users ranged from first-timers to DAO treasurers looking after nine-figure holdings. Any design decision could become a failure point.
Users weren't confused. They were anxious.
Setting up a Safe meant understanding threshold logic and why transactions queue up instead of running right away. Experienced users were fine with it. For everyone else it was a wall.
I didn't push for a full overhaul at the start. The constraints were real and a big redesign wasn't the right first move. Instead I proposed a small test: does a bit of guidance at the right moment reduce drop-off?
Three changes: helpful tooltips at the key decision points, security tips that adjusted to the user's signer count, and live deployment status while the contract was being created. Watching a screen sit still while something important happened was making new users nervous.
Users weren't confused about what to do. They were unsure whether Safe was going to lose their money. The interface was competent but not reassuring.
The wording mattered as much as the mechanic
With data from the first round, I added a "Pay Now / Pay Later" deployment option so users could put off on-chain fees and look around the wallet before committing. What surprised me was that the wording itself was what gave people the confidence to keep going.
I also tried persona-based onboarding for different user types (individuals, DAOs, developers), then dropped it after testing. Splitting things up added design debt without much payoff. A modular flow that worked across user types without fragmenting the system turned out to be the better answer.
Nobody had defined what the flow actually was
The signing flow had a different kind of problem. Modal layouts varied, transaction states weren't documented, and engineers and outside contributors were building on patterns that didn't match.
Engineering suggested moving from a modal to a full screen. I saw it differently. The modal wasn't the problem. Nobody had actually defined what the transaction flow was.
I built a master list of every transaction type and state: queued, signed, failed, batched. Then I restructured the modal so metadata, the action, and signer information were clearly separated, with bold step indicators so users could see right away what was being asked of them.
That documentation became the reference point for engineers, PMs, and ecosystem contributors, which cut down a lot of the back-and-forth about design intent.
Getting people to understand what they're signing before signing it
Batch transactions needed a different mental model: grouping several on-chain actions into one signing event. The design problem was helping people understand before they committed.
I added an entry point that stayed visible at all times and a reorder function so users could tune gas usage and execution order. I put guidance in the empty states too, since a lot of users didn't realize batching worked for anything beyond simple token transfers.
Built for contributors I'd never meet
When Safe spun off from Gnosis and became independent, it wasn't just a product anymore. It was a platform, with outside developers building apps on top of it. Whatever patterns I set would get copied by contributors I'd never talk to.
I adapted MUI to Safe's visual standards and built it on design tokens so it could support white-labeling and future themes. Each component in Storybook stood for a settled design decision, not just how it looked but why.
The test I ended up using: could a stranger build this correctly without talking to me? If not, the work wasn't finished.
Every screen had to be intuitive, gas-aware, and launch-ready
When Safe launched its own token, I designed the apps for claiming, delegation, and community engagement. Airdrop claiming, delegation mechanics, incentive flows, all of it under real time pressure.
Safe Pass was the token-locking experience. The hard part was showing the boost logic, which was lock amount multiplied by time, in a way that felt clear instead of intimidating. I set the art direction, built components for the time-based rewards, and worked with the Tokenomics Lead to make sure the UI matched the real mechanics.