There are two AI economies right now.

And they're not speaking the same language.

When I sit down with our portfolio founders, we nerd out. Which workflows they've automated this month. Which agent setups broke and why. What they ripped out and rebuilt. How a five-person team is shipping like a twenty-person team.

Then I talk to someone in corporate.

They open ChatGPT. They type a question into the box. They copy the answer into an email.

That's the workflow. That's the whole thing.

It feels like talking to someone from 2023. Same tools, but a completely different relationship with them. And most of them have no idea there's a gap.

The Text Box Problem

Here's the reality: "using AI" and "being AI-native" are not the same skill.

Using AI means asking a chatbot to rewrite your paragraph.

Being AI-native means your first instinct on any task is "how do I never do this manually again?" You chain tools. You build the workflow. You treat the model like a junior team of ten, not a smarter Google.

One is a vitamin. The other rewires how the company runs.

And corporate is stuck firmly on the vitamin.

The Data Backs It Up

This isn't just a vibe from my calendar.

  • Gallup (Q2 2026): Only 15% of U.S. workers use AI daily. Among those who use it at all, just 16% use it for automation or coding. Writing and search dominate. The text box, in other words.
  • McKinsey (State of AI 2026): Only 37% of companies say AI has contributed any EBIT, essentially unchanged from 2025. 80% say it made them individually more productive. Personal productivity, zero P&L. That's what you get from a text box.
  • MIT NANDA's "GenAI Divide": Only about 40% of companies bought an official LLM subscription. But workers at 90%+ of them use personal AI tools anyway. The employees are ahead of the institution, and the institution is pretending not to notice.
  • Microsoft Work Trend Index 2026: 45% of AI users say it feels "safer to focus on current goals than to redesign work with AI." Only 13% say they're rewarded for reinventing how work gets done.

Now look at the other side.

  • Y Combinator: A quarter of the W25 batch had codebases that were ~95% AI-generated. That was early 2025, which is ancient history at this pace.
  • Stripe's 2025 Annual Letter: The number of companies hitting $10M ARR within three months of launch doubled year over year. Stripe's best guess as to why? LLMs.

One world is debating whether to allow it.

The other world is compounding on it.

The Hiring Tax Nobody Prices In

This is where it stops being an interesting observation and starts costing you money.

You're running an AI-native team. You hire a strong operator from a big company. Great résumé. Great interview. Says all the right things: "I'm very AI-forward. I use it every day."

Then they start.

They don't know the stack. They don't know how to scope a task for an agent. They don't instinctively ask "can this be automated?" before doing it by hand. They ask for the process doc instead of building the process.

It takes months to get them up to speed.

Months.

And it's not just their ramp. It's everyone's. Your best people stop shipping to hand-hold. Workflows get re-explained. Decisions slow down, because one person on the team is still working at 2023 speed.

On a five-person startup, that's not a minor drag. That's 20% of your company running on a different operating system.

The scariest part? They believe they're AI-native. The Dunning-Kruger is the tax. Someone who knows they're behind will sprint to catch up. Someone who thinks they're ahead won't even start.

Stop Everything vs. Try Everything

Why is the gap this big?

It's not the tools. Everyone has access to the same models.

It's the inertia.

Corporate inertia is to stop everything. New tool? Pause. Legal review. Security review. Pilot program. Steering committee. Policy memo. Six months later, approved for one use case, on one team, with guardrails.

Startup inertia is to try everything. New tool drops on Tuesday. Someone's tested it by Wednesday. It's in the workflow by Friday, or it's in the trash.

Run that loop every week for three years and you don't get a gap. You get two different species.

Let's be honest: corporate guardrails exist for real reasons. Data privacy is real. Compliance is real. Breaking a bank's systems isn't the same as breaking a seed-stage CRM.

But the cost of all that caution doesn't vanish. It shows up in the people. Three years of "not yet" builds employees whose AI muscle never got used. And they carry that with them when they walk into your office.

The Real Advantage Belongs to the 22-Year-Old

Here's the contrarian part: the most AI-ready hire on the market probably isn't the one with ten years of experience.

It's the undergrad or grad student who's been messing with every tool in sight.

  • HEPI's 2026 Student AI Survey (UK): 95% of undergrads use AI. 94% use it on assessed work.
  • Gallup/Lumina 2026: 21% of U.S. college students use AI daily. Compare that to 15% of the workforce.
  • And more than half of those students say their school discourages or prohibits it.

Read that again. Students are out-adopting the workforce while their institutions tell them not to.

That's the trait you're hiring for. Not experience with AI. The reflex to try it anyway.

The caveat: not every student is building. Plenty are just outsourcing their homework. The edge goes to the ones who tinker, the kids who build the workflow instead of copying the answer. Find those people and they'll be running circles around your senior hires in a quarter.

What This Means

For founders:

  • Stop trusting "AI-forward" on a résumé. Test it. Hand candidates a real task and watch how they attack it. Do they reach for a workflow, or a text box?
  • Price the ramp into the hire. A corporate veteran who needs four months to go AI-native isn't a senior hire yet. They're an expensive junior with great instincts.
  • Rethink your junior bench. The tinkering 22-year-old with no title is often worth more than you'd think.

For corporate leaders:

  • Your people are already ahead of you. 90% of them are using AI anyway. Quit policing it and start channeling it.
  • Reward redesign, not just output. When only 13% of people get rewarded for reinventing work, don't be surprised when nobody does it.

The Bottom Line

The open question I keep coming back to: how do you upskill people fast enough?

Because the real skill isn't prompting. It's critical thinking, applied at a new speed. Knowing what to automate, what to question, and what's still worth a human's judgment.

You can't train that in a lunch-and-learn.

Startups have a massive lead right now. Not because they're smarter, but because they've been reps-deep for three years while everyone else was waiting for approval.

Corporates will catch up eventually. They always do.

The question is whether you want to be the one paying their tuition.