John Quach

Learning Scale at Google, Then Choosing the Mess

Quach spent time inside Google and Facebook specifically to see how the giants do it, how hiring machines are built and scaled. But the environment he kept returning to was the opposite one: the early startup, where the problems are messy and, in his words, there is almost nothing in place. "I love putting that structure in place," he says. From there he supported companies funded by Facebook, moved into crypto and Web3, and then into AI and gaming.

He has a quarrel with how the industry talks about that work, and he is writing an article about it. Everyone celebrates the zero to one build. Few people define the one. In talent and recruiting, Quach argues, the standard version is thin: stand up an ATS, put an interview structure in place, add some scoring, declare victory. "But I really think that it's a lot more than that," he says. His standard is a third dimension. You have to define the one so you can scale to twenty, and you have to build the one for scale, so that what works at twenty holds at two hundred and at two thousand "without you having to rewrite the playbook every time." It is an engineer's instinct applied to hiring systems: design the primitive correctly and the growth curve takes care of itself.

The AI Realist

Quach was early to AI in hiring, and he is not romantic about it. Seven or eight years ago, before the category was mainstream, he was already in conversation with the founder of a startup building recommendation engines for candidate matching and automated outreach. His view then and his view now are largely unchanged. "I think there's a lot of fluff. There's a lot of noise in the space."

At his last company he made tool evaluation a team sport: everyone was expected to test something, bring it back, and argue it out. The verdict was often unflattering. Vendors claim X, Y and Z, he says, "but when it comes down to it, I can't trust it. The results aren't consistent." Models still hallucinate. Tighten the filters too far and the system returns nothing, so you loosen them again and spend the afternoon doing it. He also names a cost few leaders account for: AI fatigue, a phenomenon Sequoia has written about, in which the sheer surface area of things to try becomes its own drain.

His discipline is ROI, framed with unsentimental math. If a build consumes twelve or fifteen hours and the requirement is two hires in one area, could he have found those two people in those same twelve or fifteen hours by hand? Sometimes the answer is yes. "What makes the most sense?" That is the question. His governing principle is that AI should augment, not substitute: look first at your workflow, find where the hours actually go, and let the technology complement what you are building rather than replace it.

Advice for What Comes Next

Heading into 2027, Quach would edit the word everyone reaches for. Not change. Iteration. He expects contraction, with a smaller set of products gathering real momentum, and he tells peers to keep re-scoring the field. Any tool tested six months ago deserves another look. "Just 'cause something didn't work for you before or left a bad taste in your mouth before, keep up with them. It's not they're not the right fit, the right tool, it's they're not there yet."

He also expects less building. Right now, he says, it is the Wild Wild West, with everyone spending time constructing their own machinery. Quach can do it himself, standing up agents without writing code, thanks to his engineering background. He simply does not think the profession should reinvent the wheel a thousand times over when capable off the shelf options are arriving fast. His closing counsel is a single word rarely offered in this market: be patient. The tools are going to stand apart, and they are going to get better.