OneKey founder Wang Yishi used a wide-ranging interview on Blockchain 100 to talk through his path into crypto, the way OneKey found its first users, how AI is changing both attack and defense in security work, and why he thinks personal growth has no ceiling.
The conversation was hosted by Beca. Wang said he bought his first Bitcoin in 2013 while he was still a junior majoring in civil engineering. Twelve years later, he is one of the founders of hardware wallet company OneKey.
He entered crypto in 2013 because Bitcoin was going up
Asked what pulled him from construction sites to crypto, Wang gave a blunt answer: price action. In his words, he bought because it had gone up.
He said 2013 was a bull market. Bitcoin had been rising since 2012 and reached nearly RMB 8,000 in the fourth quarter of 2013. His first purchase was made at a price of a little over $100, or roughly RMB 700.
At the time, he said, Taobao users could directly buy Bitcoin and XRP by paying through a product link. Domestic exchanges had also started to appear, including BTC China, founded by Yang Linke. Wang said the rally first drew his attention, then an article by Xiao Lei titled This Thing Comes Out and the World Changes pushed him to read more on Babit and Bitcointalk.
His later career path took him from ByteDance to Bixin and then to OneKey. Looking back, Wang said ByteDance had around 400 employees when he joined, versus about 100,000 now. At the time, the company’s most profitable business was advertising on the Jinri Toutiao app, and he put its valuation then at about $1 billion. He said it may now be worth $500 billion or even more than $1 trillion.
He described ByteDance at that stage as an “app factory.” The company would launch several apps each month, route traffic from the main app through tabs, and decide whether to keep investing or shut projects down based on data. A/B testing and data-driven decision-making were already deeply embedded.
Still, Wang said large companies come with a built-in problem: too many resources can become a curse. If you need people, you hire. If you need budget, you file for it. If you need traffic, you ask for it. In that environment, he said, a person can become more like a component than a builder and may miss out on what he called “wild survival experience” — building something new without traffic support or help on everything outside the product itself.
He said one reason he chose crypto was speed. Civil engineering moves slowly, with feedback cycles measured in years. Internet products move faster. Crypto moves faster still. Put money in, and if it rises, it rises. The feedback is immediate. He added that while working at ByteDance, he put almost all of his income into crypto after paying rent, which led him to ask why he should not just go all in on the industry itself.
OneKey’s first user wave came during DeFi Summer
Wang pushed back on the idea that OneKey had already “broken out” in the hardware wallet market against older players such as Ledger and Trezor. He said the company is still working hard on growth.
If there was one major inflection point, he said, it was DeFi Summer in 2020. That was when large amounts of capital moved from exchanges on to blockchains because on-chain pools offered high yields. The main tool for users at the time was MetaMask.
Wang said MetaMask’s security was not strong enough then. During an audit at the end of 2020, OneKey found that the way MetaMask stored seed phrases carried security risk because source files could be obtained and brute-forced more easily. Users with larger balances wanted access to DeFi, but they did not want to lose funds because a computer got infected with malware or a browser was compromised. That pushed them toward hardware wallets.
Even then, the user flow was clunky. Wang said a user trying to farm on Uniswap with Ledger had to install Ledger Live, install the Ethereum app, install MetaMask, connect MetaMask to Ledger, and then close the Ledger client because the two would compete for the same USB port. He compared the experience to using three remote controls for one television.
OneKey’s early advantage, he said, came from user experience improvements, especially localization and reducing friction. That brought in the company’s first wave of users.
Open source and verifiability
Wang said another reason for incremental growth was open source. Ledger, he noted, is still not open source. OneKey’s view is that there is a difference between asking users to trust that a product is secure and giving them the ability to verify whether it is secure. Over the long run, he said, the second model is better.
He extended that logic to AI models and said he is also bullish on open source there because verifiability matters.
After OneKey got its first growth wave, he said some later gains came because competitors made mistakes. He pointed to Coldcard, a wallet built by highly technical Bitcoin veterans, and said a seed phrase generation flaw remained publicly visible in open-source form from 2021 to 2025, nearly four years, without being fixed. That, he said, made him question how professional the product really was.
His summary was simple: OneKey did not “break out.” It survived. Each time a competitor made a basic mistake, OneKey’s user count would rise a little.
What went wrong during the growth phase
Wang said OneKey made plenty of mistakes of its own. The biggest one came during the DeFi Summer period, when the company’s wallet was out of stock for nearly a year.
The reason, he said, was that the team opened a new mold and wanted to build firmware end to end by itself, while underestimating both the difficulty and the development cycle. Hardware is not software. Once hardware is involved, supply chains enter the picture, and one problem can trigger a hundred more. The result was that OneKey had no inventory during a period when it should have been capturing users at scale. In his words, the company effectively gave traffic away.
He also cited technical architecture issues: premature design, premature optimization and over-design. The better order, he said, would have been to focus first on growth and getting users to use the product, then optimize later. If the team had understood that more clearly at the time, he said, the outcome could have been much better than it is now.
AI has compressed security work from months to weeks
On AI and security, Wang said this is not just a crypto problem. Many people used to think iOS was highly secure, but AI models have made it easier to uncover flaws there as well.
He said OneKey has an internal security team called Anzen Labs, with “Anzen” meaning safety in Japanese. This year, the team disclosed a USB vulnerability at Black Hat. From finding the bug to reproducing it and chaining several issues into a full supply-chain attack, AI was used through almost the entire process.
Wang gave a direct comparison. Before AI, building a complete attack chain like that would have required two to three relatively senior security researchers and about two months of work. This time, he said, the team found the issue with one security engineer in two weeks.
That speed shift is dramatic. But Wang said the easier bug-finding becomes for attackers, the easier it also becomes for defenders. Defenders have one extra advantage: they can inspect code in repositories that has not yet been released. Attackers first need to find what they want to attack. If a project is open source, everyone can inspect what is public. But products and code are always changing, and some of that work remains unpublished. That gives teams a chance to attack themselves first.
He said OneKey used to audit firmware security roughly twice a year, usually with cross-audits from at least two companies. The cadence was low. With AI tools, he said, audits can now be run every week and on every release.
His analogy for AI was a kitchen knife: it can be used to kill, or it can be used to cook. What matters is how teams use it.
The Bybit $1.5 billion theft and the weak point outside the obvious stack
Wang said stronger tools and persistent incidents can both be true at the same time. In earlier years, common attacks focused on smart contract bugs, flash loans and oracle manipulation. Those methods still exist, but he said they are less dominant now because many protocol developers use audit tools and formal verification to filter out a large share of those vulnerabilities.
That changes attacker incentives. If attacking code becomes less cost-effective, attackers look for people instead.
He used the Bybit $1.5 billion theft as a textbook example. In his telling, Bybit’s Safe multisig lost $1.5 billion even though each individual layer looked sound on its own: the multisig contract had no bug, the cold wallet itself was fine, and the Ledger hardware device had no bug either.
The problem, he said, began when a frontend engineer working on the Safe protocol was socially engineered by North Korean hacking group Lazarus. After that compromise, malicious code was inserted into Safe’s official frontend, and the code was configured to affect only Bybit’s address.
When four people at Bybit, including Ben and three finance and audit staff, signed the transaction, the webpage showed what looked like a normal transfer from a cold wallet to a hot wallet. But Ledger was using blind signing. It did not parse the Safe contract, did not display the delegatecall, and did not issue any warning.
As a result, Wang said, what they actually signed was not a standard transfer but a delegated handover of authority that gave away control of the Bybit Safe contract. Ownership was gone.
He said the case is painful precisely because each component looked fine in isolation. The one visible breach was the Safe frontend engineer’s machine, but once all the conditions lined up, the funds were lost. In the past, attackers tried to break code. Now they try to break people.
Wang borrowed an analogy from road safety: if you give drivers seat belts and airbags, they may drive faster. The same thing can happen in crypto. Teams add more audits, more multisig, more passphrases and raise the attack threshold, but the eventual failure often comes from the place nobody is watching. He compared it to a castle that is impossible to breach directly while the guard walks outside carrying the keys and loses them after a casual interaction.
Why OneKey publicly disclosed a Ledger bug
Wang also discussed a recent post framed as “we hacked Ledger,” which disclosed a transaction replacement vulnerability affecting Ledger. He said the first point to make is that the issue had already been fixed.
By the time he posted, Ledger firmware had already reached version 1.2.3. The flaw existed in version 1.2.1 and had been patched roughly two weeks earlier. Wang acknowledged that “we hacked Ledger” was a headline-driven way to put it.
Technically, he said, the issue was a TOCTOU bug — time-of-check to time-of-use — exploiting the timing gap between the device and the computer. A user might see “send $1 million from address A to address B” on the device and sign it. In the gap before execution, an attacker could push in transaction B, so the user would end up signing B while B displaced A.
By severity, Wang said, the flaw was quite dangerous.
He said Anzen Labs spends its days trying to hack OneKey’s own products. If the team cannot break its own systems, that means its security capability is still lacking. He added that the industry does share information. OneKey previously reported some security bugs to Keystone, gave the company about two months’ notice, explained how to reproduce the issues and provided full fixes, then disclosed them together after the patches were shipped and mandatory updates were in place. Wang said that kind of exchange reflects a healthy Olympic spirit.
Hiring at OneKey starts with paid practical work
On hiring, Wang explained why OneKey asks candidates to complete paid practical assignments and then, if they pass, go through a paid trial period. The core reason, he said, is simple: the company struggles to find the right people.
What looks like an unusual hiring process is not unusual at all in his view. It is a response to the low hit rate of conventional methods. He compared hiring to matchmaking. If a candidate performs well in an interview, that mainly proves the person is good at interviewing. Strong candidates can guide interviewers into reaching the conclusion they want them to reach.
That makes the interview process feel like mutual performance, he said. The candidate performs competence, and the company performs being a great place to work. Two actors watch each other and then decide whether to stay together.
A paid assignment over two or three days reveals more. It shows how a candidate thinks through a problem, how complete the work is, and how well the person delivers. Most important, Wang said, it shows whether the person communicates actively when problems appear or stays silent and surfaces a major issue at the end. In actual work, he wants problems raised early.
He stressed that these assignments are written internally, not copied from Y Combinator or used to extract free ideas. Candidates also benefit because they get a direct feel for how the company works. The person setting the task is usually the future teammate or manager, so the candidate can judge whether they want to work with that person.
Pay matters too. Wang said it is partly about respecting the candidate’s time, but it also signals seriousness. If a company suddenly throws a task at someone without compensation, the candidate may feel exploited. Paying for the work changes that dynamic.
AI coding, robotic arms and a shared sense of what is possible
Beca brought up one of Wang’s earlier remarks: that he could not accept engineers in 2025 who still could not use AI coding tools efficiently. Wang said that line was a bit extreme at the time, especially because Codex had not yet launched and AI was only beginning to improve productivity in programming and text work.
Still, he said the company has since rolled out many new hardware-related systems. One example is hardware testing. There are now no human test engineers in the office for that workflow, only four to five robotic arms.
Some phone tests can be run by plugging in a USB cable and reading commands directly. Others depend on external action, such as whether a swipe feels smooth or whether a button has the right tactile response. Those require machine vision and an external mechanical hand. Wang said that before 2026, many test engineers in the office still had to tap through these cases manually. Now robotic arms, high-definition cameras and models handle the process. On each release, the relevant build is automatically sent to the testing backend, and the robotic arms run through the test cases one by one.
He said this system was not built by a single hardware engineer or a single test engineer. It was built by two people working together. The key was that both shared a common understanding of what the strongest AI models could now do. Based on that shared boundary, they believed the task was possible, tried it, and proved that it was.
For Wang, that is AI’s biggest effect: it gives teams a common sense of the frontier, turning work that two disciplines once could not even imagine doing together into something that produces results greater than 1 + 1.
Giving his son capital to lose early
Wang also revisited a post he had made on X about opening Binance and Robinhood accounts for his son under his own name and adding money to them each year.
He said he did want to do it, but two practical issues got in the way. Robinhood requires a U.S. person, and his child is not one. Binance’s Junior account requires the child to be at least 6 or 13 years old, while his son was only three months old at the time. So the idea remains on his to-do list, but he said he will do it later.
He referenced the founder of Dell opening an account for newborn babies in the U.S. and putting in $250, saying he found the idea interesting.
The reason he wants to do this is that he believes children should encounter money early. Not pocket money for ice cream, but something more structured. Once a child can recognize numbers and handle arithmetic within 100, Wang said, parents can try giving them a savings account or even an investment account so they can learn how money works in the world.
His key point was that children should lose that money early. If a family gives a child RMB 10,000 and the child loses all of it before adulthood, Wang said, that is better than making a larger mistake later after graduation and employment. He summed it up with the phrase “Fail cheap, fail early,” and said the process also builds financial awareness.
AI, children and imagination
Wang rejected the idea that in the AI era only children from wealthy families will avoid having their imagination constrained by AI. His response was that AI should free imagination, not limit it.
He said human progress ultimately comes from younger people not simply following older people. With AI, children can do many more things. They can be builders from the start. They do not necessarily need to learn programming before they can make a product. If they can express what they want, they can build.
He pointed to the spread of home 3D printers such as Bambu Lab. Families can print many small objects at home, not just figurines but practical items such as automatic switches or foldable pads for coffee machines. Wang said it is hard for him to imagine how happy he would have been if he had access to that kind of tool as a child.
In his view, children can now build apps, websites and physical objects in the real world, then combine them in countless ways. The possibilities are effectively unlimited.
“Show me the product”
At the end of the interview, Beca asked what he would say to people who want to build something in crypto today. Wang said he would hesitate to offer advice because he does not think he has done everything especially well himself.
But if he had to say one thing, it would be this: everyone’s capacity for growth is unlimited. What you studied, what you did after graduation, what you did at 20, 25 and 30 can all be completely different.
He said liking something determines how much effort you put into it, while your perspective and understanding determine its upper bound. If you like the thing and it sits on a long slope in a long market, you are more likely to go far.
He also said people used to keep their work private and only show it after launch. That has changed because AI can now build things very quickly. If you have an idea, he said, you can direct Codex, Grok or Claude to help build it, and it may come together after only a couple of resets.
Once it exists, you can put it in front of users immediately and get feedback fast. Is it a real need? Is it only your need, or do many people share it without saying so? Can it make money? Can it keep users over time?
Wang cited Lovable as an example. Before building Lovable, its creator had made more than 30 vibe-coding apps that nobody used. Then Lovable took off. For Wang, that is a classic case of rapid trial and error.
When someone builds a new product, he said, they connect the strengths and weaknesses of everything they built before. AI cannot replace that because it lives in human judgment, taste and decision-making.
The old line was “idea is cheap, show me the code.” Wang’s updated version is different: you do not even need to show the code anymore. Just show the product.
His closing view was straightforward: this really is the best era.