A crypto wallet can look innocent enough: a balance, a few buttons, perhaps a cheerful icon. Behind the glass sits a small republic of hazards. There are secret phrases that cannot be forgotten, addresses that cannot be mistyped, signatures that cannot be recalled and networks that do not particularly care whether a newcomer understands them. Francesco Agosti has spent the past five years trying to make that republic governable.
Agosti is the chief technology officer and a co-founder of Phantom, the app that began as a Solana wallet and grew into a multichain consumer-finance product. His job is full of an unusually delicate contradiction. The software must give people direct control of digital assets, yet spare them the ceremony and jargon that direct control has traditionally required. It must be simple without pretending the stakes are small.
That tension explains more about him than the familiar startup nouns: founder, engineer, executive. Agosti's public writing is practical and spare. His personal website is almost comically economical. It points outward to work, code and profiles, then gets out of the way. The sensibility carries into Phantom. The technology may be elaborate. The user should not have to admire the elaboration.
A useful education in scale
The route to Phantom began at Duke University, where Agosti studied computer science from 2011 to 2015. For one summer he worked as a particle-physics research fellow under Mark Kruse. Particle physics and consumer software do not often share a paragraph, but they do reward a similar temperament: invisible systems, indirect evidence and a low tolerance for hand-waving.
After Duke came Yelp. From 2015 to 2018, Agosti worked as a software engineer on a product built for ordinary people at considerable scale. He then moved deeper into crypto's machinery at 0x, an open protocol for decentralized exchange, where he became a tech lead. Yelp offered the consumer surface. 0x offered the protocols underneath. Phantom would require both kinds of fluency.
His surviving technical posts from the period are not manifestos. One explains how to pre-render a single-page application hosted on Amazon S3 so search engines can understand it. Others record the monthly work of the 0x core team. They show an engineer occupied by the uncelebrated joins between systems: the place where a modern web application meets an old crawler, or where a protocol team turns progress into something outsiders can inspect. The grand theory, if there is one, keeps arriving disguised as maintenance.
In January 2021, Agosti joined Brandon Millman and Chris Kalani to start Phantom. The group brought engineering, product and design experience from 0x and elsewhere to a blunt premise: wallets were too difficult. Their beta community numbered 40,000 active testers. By July, the company had raised a $9 million Series A. In-app swaps passed $1 billion in volume that October. Growth was quick enough to make a friendly ghost into a serious infrastructure problem.
“The best time to build is in a bear market.”Francesco Agosti, 2022
Safety belongs in the interface
Self-custody is a charming phrase for an unforgiving arrangement. Control stays with the user; so do the consequences. In October 2021, Agosti and Kalani published an account of Phantom's early security work. The important move was philosophical as much as technical. Education could only go so far. If scammers were exploiting the product's mechanics, the product had to change.
Phantom removed an auto-approve option that malicious sites could abuse. It added clearer transaction details and simulated transactions before submission, showing what assets might move and whether the operation might fail. It introduced anti-phishing warnings and directed support away from public chat channels where impersonators prowled for worried users. These were not glamorous features. Glamour is rarely what one wants from a lock.
There is a design lesson in the transaction preview. Blockchains are literal machines: they execute what is signed, not what a person vaguely intended. Simulation inserts a moment of interpretation between code and consent. A list of instructions becomes a preview of assets moving in or out. The computer does not become more forgiving, but the human gets a fairer chance to understand the bargain. For a product aimed at newcomers, that translation is not decoration. It is part of the security model.
The choices reveal Agosti's practical thesis. A wallet should not merely warn users to become experts. It should carry more of the burden itself. Show the likely consequence before approval. Eliminate the dangerous shortcut. Make the legitimate help desk unmistakable. In ordinary finance, decades of institutional machinery sit between a mistake and catastrophe. In self-custody, software must recreate some of that protection without quietly reclaiming control.
Peak requests per second during the WEN claim exposed database and network limits. Phantom used the failure to prepare a dedicated squad and far more capacity for JUP four days later.
When the crowd arrives at once
The infrastructure underneath a calm interface occasionally announces itself. In January 2024, the WEN token airdrop sent a surge through Phantom. Traffic reached 8,000 requests per second, roughly eight times what the team had seen during the PYTH airdrop. Database connections were exhausted. Serverless executions multiplied. A network-address limit joined the party, the way a third act complication arrives just when the audience thinks the plot is sufficiently busy.
Four days later, the much larger JUP airdrop was due. Phantom assembled a dedicated squad that included Agosti and coordinated load tests with infrastructure and security partners. Providers tested capacity of 25,000 requests per second per node across three machines in separate clusters. The episode is a tidy portrait of engineering leadership: failure translated into a written account, then into capacity, coordination and another attempt.
There is no romance in a NAT gateway limit. There is, however, character in publishing what broke. Consumer products acquire trust partly through polish and partly through candor about the machinery beneath it. Phantom's account described the timeouts, the concurrency increases, the lingering connections and the fixes. The friendly surface survived because a team was willing to study an unfriendly postmortem.
From wallet to home for money
Phantom expanded beyond Solana to Ethereum, Polygon, Bitcoin and other networks. The company raised a $109 million Series B in 2022. In January 2025, a $150 million Series C co-led by Sequoia Capital and Paradigm valued it at $3 billion. Phantom's own history counted more than 15 million users that year; its current homepage says more than 20 million.
The more revealing change is linguistic. Phantom now describes itself as a home for trading, spending, sending and saving, not simply a crypto wallet. That ambition can sound enormous, but its engineering direction is specific: fewer raw addresses, fewer visible bridges, fewer moments when a user must know which network is doing what. The wallet remains. Its mechanics recede.
This is where consumer design and self-custody can quarrel. Every abstraction removes a decision from view, while the founding promise of the wallet is that the user remains in charge. Hide too little and the product feels like a cockpit. Hide too much and consent becomes theatre. Agosti's technical brief lives in that narrow middle: simplify the route, preserve the exits and make the consequential moment legible.
Agosti has also kept security near the center of that expansion. In October 2025, he announced the hiring of a new head of security and Phantom's membership in the Crypto Information Sharing and Analysis Center. His premise was direct: millions of monthly users holding billions in self-custodied assets place the company on crypto's front line. Scale does not make the old risks disappear. It gives them better attendance.
The best consumer infrastructure does not ask to be admired. It asks to be trusted, then keeps earning the privilege.
The next abstraction
In 2026, Agosti turned his attention to the way Phantom itself builds software. “With AI, software engineering is completely changing,” he wrote while recruiting a leader for an internal AI harness. The proposed job was not merely to add a chatbot to an engineering department. It was to rethink how the company ships product.
The challenge suits his career. At every stop, Agosti has worked at a boundary where technical complexity threatens to leak into human experience. AI-assisted engineering creates the same bargain inside the company. Faster production is appealing. In financial software, careless production is not. The useful system will be the one that increases leverage while preserving judgment, testing and responsibility.
Phantom's most interesting aspiration is therefore not a feature list. It is the disappearance of a category. If the company succeeds, a person may use open financial networks without pausing to admire the wallet, just as few diners at a restaurant contemplate Yelp's distributed systems before checking the reviews. The interface becomes familiar. The infrastructure behaves. Control remains where it belongs.
That is a severe standard for a piece of software and a rather elegant one for its CTO. Agosti is not trying to make crypto look simple in a screenshot. He is trying to make simplicity hold up when scammers arrive, when a million claims land together and when the underlying technology changes again. A disappearing act, properly executed, involves a remarkable amount of work behind the curtain.