For New Disruptors: Welcome. Every week we cut through the AI noise and hand you something you can actually use — no hype, no fear, no jargon. Just the shift that matters and what to do about it.

 This week I've been reflecting on a quiet contradiction. For two years, the story we were told was simple: the company with the best model wins. Pick the right AI, and the results follow. Then, in the span of a single week this month, four of the largest technology companies on earth spent billions telling us the opposite. Nobody said it out loud. But if you read what they did instead of what they said, the message is impossible to miss.

On July 2, Microsoft announced it is investing $2.5 billion and adding 6,000 employees to a new group called the Frontier Company. Its job is not to build a better model. Its job is to send Microsoft engineers into customers' companies to build and run the AI systems those customers have already paid for.

 Two days earlier, Amazon committed $1 billion to the same idea. Back in May, Anthropic and OpenAI launched their own versions. The industry even has a name for it now: forward-deployed engineering. Put your own people inside the customer's building and make the thing actually work.

Stop and think about what that means. The companies that sell the models are now paying to place humans next to the customer because selling the model alone was not enough. If the product worked the way the sales deck promised, you would not need to embed an engineer in someone's operations for six months to see a result.

 That's the admission. The bottleneck was never the model. It was everything around it.

Here's the number driving all of this. Research from MIT's Project NANDA found that 95% of enterprise AI pilots deliver zero measurable impact on profit and loss. Not a small impact. Zero.

That should stop every leader cold. Ninety-five out of a hundred AI projects produce a demo, a round of applause, and nothing on the bottom line. The tools work in the demo. They fail in the business. And the gap between those two places has almost nothing to do with the quality of the AI.

We see the same pattern in the culture data. A recent global study of 2,400 knowledge workers and executives found that 54% of C-suite executives say adopting AI is tearing their companies apart. More than half report power struggles between departments. Nearly eight in ten say it created tension between IT and the rest of the business. One in two called their own rollout a "chaotic free-for-all."

Read those two findings together. The pilots fail on the P&L, and the people fail on trust. That is not a technology story. That is a human story wearing a technology costume.

When a pilot dies, it usually dies in one of four places. None of them are the model. This is the work forward-deployed engineers are actually being paid to do — and it's the same work your own team has to own.

One: naming the job. Most pilots start with "let's try AI on this" instead of "here is the specific decision or task we want to make faster, cheaper, or better." A tool pointed at a vague goal produces a vague result. The engineer's first job inside a company is to force that clarity. You can do it before you spend a dollar.

Two: fitting the workflow. A model that lives in a separate tab is a model nobody uses. The win comes when it sits inside the process people already run — the intake form, the ticket queue, the Monday report. Someone has to redesign that workflow. That's a people problem, not a licensing problem.

Three: owning the handoff. Every AI task has a moment where the machine hands work back to a human, or the human hands work to the machine. When nobody owns that handoff, work falls through the crack and trust collapses. Naming an owner costs nothing.

Four: proving the number. The 95% failure is really a measurement failure. Most teams never defined what a win looked like, so they can't tell if they got one. Pick the metric before you start. Baseline it. Then you'll know.

Notice something. All four are things a person decides, not things a model computes. You can hire Microsoft to send an engineer to do them. Or you can build the muscle yourself and keep the ownership — and the savings — inside your own walls.

Here's what I keep coming back to. The forward-deployed engineering wave is being sold as innovation. I read it as an honest correction. For two years, the industry told business leaders that intelligence was the scarce thing — buy enough of it and you'd win. That was never true. Intelligence is becoming abundant and cheap. What remains scarce is the human work of pointing it at the right problem and rebuilding how we work around it.

That's actually good news, and it's the whole reason I'm optimistic. If the winning move were simply "buy the biggest model," then whoever had the most money would win, and everyone else would lose. But that's not the game. The game is clarity, ownership, and the discipline to redesign your own workflows. Those are available to a ten-person shop in Cincinnati just as much as to a Fortune 500 company. The billion-dollar deployment armies are proof that the money can't skip the human part. Neither can you — and you shouldn't want to. That’s where the ownership lives.

 For enterprise teams of 20+ trying to turn stalled AI pilots into real results — let's talk. Disruption Now trains your people to name the job, fit the workflow, and own the outcome, so the value stays inside your walls instead of walking out with a vendor's consultant.

→ Book a Consultation

 MidwestCon is where policy meets innovation, creators ignite change, and tech fuels social impact. This year's theme—"The Era of Abundant Intelligence"—explores how AI is reshaping what's possible when intelligence becomes accessible to everyone.

 → Get a Ticket Now

Disruption Now® interviews leaders focused on the intersection of emerging tech, humanity, and policy.

 → Subscribe to the Podcast

Keep Disrupting, My Friends.

Rob Richardson – Founder, Disruption Now® & Chief Curator of MidwestCon