The confidence gap just got a scoreboard

Thirty per cent of the apps deployed on Vercel are now built by AI agents, not people. TechCrunch reported that the company's annual recurring revenue has grown 240%, from $100M to $340M, and its CEO is openly signalling IPO readiness. If you're inside the AI industry, this is the prevailing mood: revenue up, agents building the apps, IPO talk on every earnings call.

If you're outside it, Stanford has your number. The 2026 AI Index found that only 10% of Americans feel more excited than concerned about AI. On jobs, the gap is starker: 73% of AI researchers are positive about employment effects versus 23% of the public. Americans trust their government to regulate AI responsibly at just 31%, the lowest of any country surveyed.

The standard reading of this divide is that it's a communications problem: if only the public understood what the insiders understand. I think the data says something less comfortable. Both sides are reading the situation correctly, from where they sit.

PwC's 2026 AI Performance study makes the economics concrete. Seventy-four per cent of AI's economic gains are captured by 20% of companies. The top performers deliver 7.2x more AI-driven revenue and efficiency gains than the average firm. This follows a pattern economists have documented in every major technology wave since electrification. In the 1920s, factories that reorganised production around electric motors saw enormous productivity gains. Those that swapped steam engines for motors and changed nothing else saw almost none. Same technology, wildly different returns, determined entirely by whether you rebuilt the system around it.

The AI version is playing out now. Vercel's growth isn't because "AI is good." It's because the company redesigned its platform around agent workflows, turning AI-generated code into a deployment advantage. PwC's leaders differentiated by pursuing new revenue streams, not bolting chatbots onto existing processes. The 20% didn't adopt faster. They adopted differently.

That makes the public scepticism rational, not ignorant. If you work at a company in the other 80%, your lived experience of AI is probably a chatbot tucked into an internal tool and a vague all-hands about "transformation." Nature confirmed that human researchers still outperform the best AI agents on complex, multi-step scientific tasks. AI handles the routine brilliantly. It struggles where novelty and judgement matter. For the majority of workers, AI's promise remains a rumour from someone else's office.

The builder's problem

The confidence gap isn't a perception failure waiting to be fixed by better messaging. It's an economic distribution with receipts. The gains are real, but they accrue to organisations that restructure around the technology rather than layer it on top.

For product teams, the PwC finding that leaders started with business strategy rather than technology selection ought to redirect a lot of roadmap conversations. The question worth asking isn't "are we using AI?" Nearly everyone is. It's whether you're using it in a way that puts you in the 20% or the 80%. The scoreboard is up. So far, both the optimists and the sceptics are winning.

---