By Damian Mathews and The Last Mile Team
Last week PwC published its 2026 AI Performance Study.
1,217 senior executives. 25 sectors. There was one finding that I think should make most leaders uncomfortable: 74% of all AI-driven economic value is being captured by just 20% of organizations.
Everyone else is stuck in what PwC calls “pilot mode.” Sound familiar? Activity without returns. Experiments without outcomes.
The 80% have AI tools. They have pilots. They have innovation teams and budget. What they don’t have is a measurable financial return from any of it.
So what separates the 20%?
According to PwC, the single strongest predictor of AI-driven financial performance is whether the company uses AI to pursue growth and reinvent how it creates value. The top 20% are 2.6 times more likely to say AI improved their ability to reinvent their business model. They’re generating 7.2 times more AI-driven gains than the average competitor.
In other words, the 80% are using AI to do the same work slightly faster. The 20% are using it to change what work they do.
For CX, that distinction is everything.
Most contact center AI deployments start with a cost question. How do we reduce headcount? Deflect more volume? Shorten handle time? Those are optimization goals. They make the existing operation cheaper without changing what it does or how it creates value for customers.
The 20% approach starts somewhere different. Customer evidence: what are people truly calling about, where does the process break, what would resolution look like if you designed it from scratch? That leads to different use cases, different outcomes, and a customer experience that changes in ways people can feel.
PwC also found that the leaders are redesigning workflows around AI rather than layering AI on top of broken processes. They’re building governance frameworks that create trust at scale. Their employees are twice as likely to trust AI outputs, which creates a compounding cycle: trust enables automation, automation enables scaling, scaling enables more value.
The 80% are launching pilots without killing the ones that don’t perform. They have no repeatable system for taking something from “worked in a demo” to “live, governed, and improving every week.”
We lived this.
We spent 2025 transforming our own operation before taking any of it to clients and documented the whole thing in A1B: Customer Zero to AI-First. Five plays that took us from scattered AI curiosity to thirty deployed tools in production. The pattern maps almost exactly to what PwC found in the top 20%: start with evidence, prototype before you plan, prove before you scale, build governance first, and make sure the people doing the work own the method.
On our recent LinkedIn Live, Fish and Kerry dug into why this divide keeps widening. Fish made the point that AI absorbs the operational layer of work faster than most people realize. Kerry added that the durable human contribution is taste, judgment, and knowing what “good” looks like for the customer. The 20% have figured out how to pair those two things. The 80% are still debating which vendor to pick.
PwC’s own warning: “Without a shift in approach, the performance gap between AI leaders and laggards is likely to widen further.”
The gap isn’t closing. It’s compounding. Every production deployment teaches the 20% something that makes the next one faster. The 80% are restarting the same pilot cycle every quarter.
We’ve been building something specifically designed to close that gap. More very soon.
Are you in the 20% or the 80%?
— Damian
Here’s what went down this week.
Bleeding Edge
Early signals you should keep on your radar.
Amazon bet $25 billion more on Anthropic, on top of the $8 billion it already committed. An initial $5 billion lands now, with the rest tied to milestones and up to five gigawatts of Trainium capacity for Claude. Hyperscaler commitments at this scale could well shape compute pricing and supply dynamics for the next two years.
Nvidia’s chip challengers are raising record money, with AI silicon startups pulling in $8.3 billion so far this year. Dutch startup Euclyd, backed by ASML’s former CEO, claims 100x inference efficiency over Nvidia’s Vera Rubin, while UK peers Fractile and Optalysys chase nine-figure rounds. Training is locked up for now, but inference workloads could hand specialists a genuine opening against GPU incumbents.
Leading Edge
Proven moves you can copy today.
OpenAI’s Codex now drives your Mac, clicking and typing with its own cursor across any app on the desktop. The update adds in-app browser support, persistent memory, and more than 90 new plugins spanning dev tools and app integrations. For teams already shipping with AI pair programming, this moves Codex from code suggestor to actual desktop operator.
Cursor is raising $2 billion at a $50 billion valuation, two years after the coding assistant first crossed $1 billion in annualized revenue. SpaceX reportedly also holds an option to acquire the startup outright for $60 billion. Developer tooling is shaping up as the cleanest commercial wedge in AI, with enterprise IT budgets following fast.
Off the Ledge
Hype and headaches we’re steering clear of.
Meta will record employees’ keystrokes and mouse movements to feed its AI training pipeline. There is no opt-out, and staff are already pushing back on what critics call workplace surveillance dressed up as AGI research. Enterprise buyers watching this story should ask any vendor, Meta included, exactly whose keystrokes trained the model they’re about to deploy.
ChatGPT went dark for hours on Monday, dragging Codex and the API platform down with it. Downdetector logged over 13,000 reports at peak, with the UK hit hardest before OpenAI shipped a fix around the 90-minute mark. Workflows increasingly run through a single vendor, so a midmorning wobble in San Francisco quietly froze teams from London to Lagos.
