AI isn’t just changing the nature of companies, but it’s also changing the nature of employees.

That’s the argument Kunal Shah made in a recent conversation, drawing on what he’s seen firsthand at CRED. Shah, who stepped down as CRED CEO this week to take over as global head of WhatsApp at Meta, described a split forming inside tech startups — one that goes well beyond the usual gaps in skill or seniority.

“Ten percent of every tech startup’s employees are becoming a completely different species to the ninety percent. Because their productivity is now at this exponential scale and they are actually finding every single person around them and every other process to be slow,” he said.

His evidence: “Ninety percent of code at CRED is written by AI now. This number would have been maybe five percent a year ago.” Five to ninety percent in a year is a dramatic shift by any measure, and Shah made clear the company was still in the middle of figuring out what building agentically at scale actually looks like.

The specific thing Shah is pointing to is a productivity divergence, not a skills gap. The employees who’ve learned to operate with AI as a genuine multiplier aren’t just faster at their existing jobs — they’re working at a pace that makes the rest of the organisation feel like friction. That’s a different kind of problem. It’s not about upskilling the 90%; it’s about what happens to team dynamics, processes, and organisational structure when 10% of your people are operating in a fundamentally different mode.

This pattern has been showing up elsewhere too. Salesforce stopped hiring software engineers entirely last year, citing over 30% engineering productivity gains from AI tools. Uber burned through its entire 2026 AI coding budget in four months — with 70% of code commits already AI-driven — though its COO admitted the connection between those numbers and actual business value was still unclear. Microsoft CEO Satya Nadella has framed the core challenge as a management discipline problem: matching what you spend on AI tokens to what you actually get out of them.

What Shah is describing sits underneath all of that. The macro debate about AI’s ROI is real, but there’s something more structural happening inside the companies running these experiments. A minority of employees have crossed a threshold — they’ve internalised the tools, built new workflows around them, and are now compounding. Everyone else is still working roughly the way they were before.

The uncomfortable implication of the 10/90 split is that it doesn’t automatically resolve itself over time. The people who are ahead keep moving faster; the gap between them and their colleagues who haven’t adapted is widening, and it shows in every meeting, every review cycle, every handoff. Shah’s framing — “a completely different species” — sounds dramatic, but it’s precise. It’s not that the 10% are smarter or more experienced. They’re operating on a different productivity curve.

For founders and operators, the question this raises isn’t just about tools or training budgets. It’s about what a company looks like structurally when a small fraction of employees can do the work that previously required teams, and when those employees find most of the existing organisational apparatus — approvals, processes, dependencies — actively slowing them down.

Shah has spent years thinking about how organisations shape behaviour, and his observation here tracks with something broader that’s becoming visible across India’s tech ecosystem: the AI transition isn’t happening uniformly, and the companies that figure out how to build around the 10% rather than just train up the 90% will probably look very different from the ones that don’t.