On July 22, a seven-month-old startup called Arrakis came out of stealth with $38 million in funding. Blossom Capital led the round. Accel participated. The Datadog CEO put in personal money. So did OpenAI’s head of business products.
None of that is unusual in 2026. What is unusual is where Arrakis is pointing its AI agents. Not at knowledge workers. Not at marketing teams or customer support desks. At aerospace, energy, logistics, and manufacturing. The 70 percent of the economy that most AI companies have ignored entirely.
CEO Rafael Quintanilla told Fortune the thesis plainly: “Most AI investment to date has targeted the 30% of workers behind a desk. The real ROI lies in the 70% running industrial operations.”
He is not wrong. And the reason he is not wrong tells you more about where AI value actually lives than any model benchmark ever will.
The Spreadsheet That Already Existed
Here is how Arrakis landed one of its first customers, a publicly traded shipping company in New York.
The goal was cash-flow visibility. The company had monthly snapshots. Arrakis was hired to get that to daily. The engineering approach most AI companies would take: build a new dashboard, connect a dozen APIs, deploy a custom model, train the team on new software.
Arrakis did something different. They rebuilt the spreadsheet the operators already used. Same layout. Same columns. Same workflow. The only change was that AI populated the data instead of a person manually pulling it from three systems. As operators corrected entries, the system learned their logic and started codifying their knowledge.
No new interface. No retraining. No change management deck. The operators opened the same file they opened yesterday, and it was better.
That is not a technology breakthrough. It is a design choice. And it is the design choice that separates companies getting real value from AI and companies stuck in pilot programs wondering why adoption stalled.
Why Complexity Keeps Losing
Arrakis reported a 90 percent reduction in procurement cycle times for its customers. Ninety percent. Not from a more powerful model. Not from a proprietary dataset. From removing friction between the AI and the person doing the work.
This pattern keeps showing up, and most organizations keep missing it.
The instinct when deploying AI is to build something new. New platform. New workflow. New training program. New role to manage it all. Every layer adds cost, adds timeline, and adds one more reason for the people who actually do the work to not use it.
The companies getting results are going the other direction. They are finding the existing workflow, the spreadsheet or checklist or routing process that someone already does forty times a week, and making AI do the manual part of it. No migration. No transformation. Just subtraction of the part nobody wanted to do.
Quintanilla’s playbook starts at headquarters, proves the value with one team, then expands to field operations only when there is pull from the people on the ground. Not push from management. Pull. The team asks for it because they saw what it did for the first group.
That is adoption driven by evidence, not mandates. It works because the barrier to entry is nearly zero.
The Model-Agnostic Bet
Arrakis is also model-agnostic. They start with OpenAI or Anthropic, then shift customers to open-source alternatives and cut token costs by roughly 70 percent. One Swiss executive told Quintanilla: “Everyone told us we had to be on Copilot. Then we went to OpenAI. Now it’s Anthropic. My head is going like this.”
That executive’s frustration is the frustration of every business leader who has been told the model is the decision that matters. It is not. The model is a commodity input. It changes every quarter. What does not change is whether the deployment fits into how your team actually works.
The organizations spending months evaluating which model to standardize on are solving the wrong problem. The organizations deploying in weeks by plugging AI into existing workflows are getting the results.
What This Means for Your Team
Arrakis has five customers. It is six months old. It is not yet proof that this scales to every industry. But the pattern it represents is already proven across dozens of companies and sectors.
The highest-value AI deployments are not the most sophisticated. They are the most accessible. The ones a new hire could understand in an hour. The ones that do not require a specialist to maintain. The ones that look, from the outside, almost boring.
If your AI initiative requires a change management program to get people to use it, the problem is not change management. The problem is that you built something nobody asked for.
The companies pulling ahead right now are not pulling ahead because they picked the right model or hired the right AI team. They are pulling ahead because they did the simplest possible thing first, proved it worked, and let the results do the selling.
That is the unlock most organizations are still looking for. It was never complicated. That was the point.