From OneSky cofounder to AI translation believer.

I cofounded OneSky. We served Tencent, LINE, Airbnb, change.org. Managed thousands of human translators. Built workflows that handled millions of words monthly. And you know what? The entire industry was fundamentally broken. 3-week turnaround for a mobile app update. $50,000 invoices for translating release notes. Endless email chains about whether 'user' should be formal or informal in German.

The real kicker? I'm currently struggling to learn Estonian. My kids find this hilarious. 'Isa built a translation company but can't even order coffee in Tallinn.' They're not wrong. Turns out 'kohvi' isn't pronounced how you'd think, and no amount of translation management experience helps when you're standing at a cafe counter sweating through basic vocabulary.

But here's what 6 years in human translation taught me: The problem wasn't the translators. They were brilliant. The problem was the system. The overhead. The communication gaps. The context that got lost between a developer in San Francisco and a translator in Seoul.

At OneSky, we had developers and PMs. Quality assurance workflows. Vendor management. Account executives. You know what actually translated the content? Maybe 5% of that workforce. The rest? Managing the 5%.

Reality check from the trenches: I watched Tencent pay us six figures to translate patch notes that GPT-5 now handles better in 30 seconds. Airbnb waited weeks for property descriptions that Claude could perfect in minutes. Change.org's urgent campaign translations sat in queues while people debated terminology.

The translation industry's dirty secret? It's not about translation quality anymore. Hasn't been for years. It's about managing complexity that shouldn't exist. We built entire businesses around coordinating humans to do what AI now does instantly, consistently, and without needing a project manager to send seventeen follow-up emails.

Now? I have Claude, GPT-5, and a direct line to what actually matters: the output. No account managers. No project coordinators. No vendor disputes. When a customer needs something fixed, I fix it. When LINE needed 50,000 words translated overnight back in the day, it was a crisis. Today, it's a Tuesday afternoon.

"But you can't match human quality!" says everyone who hasn't actually tested modern AI against their 'premium' translation vendors.

Friend, I've seen both sides. I've reviewed millions of human translations. The best ones? Incredible. The average ones that actually ship? AI beats them. Every. Single. Time. And it never translates 'Save' as 'Rescue' because it's having a bad day.

function buildSoftware() {
  while (problemExists) {
    const solution = thinkDeeply();
    const code = writeCleanCode(solution);
    const result = ship(code);
    
    if (result.usersSatisfied) {
      celebrate.withCoffee();
    } else {
      iterate();
    }
  }
}

I went from co-founding a company that served Fortune 500s to debugging production issues while my kids ask why I'm 'yelling at the computer in Cantonese when it only speaks Estonian.' (They think all computers in Estonia speak Estonian. I haven't corrected them.)

At OneSky, we had incident response teams. Now? It's just me, explaining to my family why dinner's delayed because 'Daddy needs to fix the robot translators.' They've started setting a place for 'the bug' at dinner, assuming it's a regular guest.

From enterprise meetings to solo reality:

But also?

At OneSky, we had microservices for microservices. Kubernetes clusters managing other Kubernetes clusters. You know what actually mattered? None of it. The tech was never the bottleneck — human coordination was.

Hot take: Your tech stack doesn't matter nearly as much as understanding your problem domain. I've seen million-dollar Kubernetes clusters serve broken experiences while someone's PHP monolith prints money.