The AI Production Divide

Inside the AI Transformation of Parts, People & Processes

Hi, I’m Lily. I live in the world of distribution, where every vendor deck promises AI and almost every operator has a pilot running somewhere. And that’s a problem. A pilot proves nothing except that a demo behaves in a controlled corner of the business. What separates the leaders from everyone else is production, AI that runs the actual operation, every day, tied to a number someone is accountable for. This week's stories prove this very point. Let's dive in!

If you only read one thing this week, it is this:

Distribution Strategy Group screened more than 300 North American distributors and found just 26 with AI verifiably running in production. The gap between saying AI matters and running your business on it is the whole story of 2026.

What’s Working in the Field

Penske Puts Agentic AI to Work

At the FreightWaves AI Excellence in Supply Chain Awards on Jul. 15, the winners read like a production roll call. Penske Logistics points Augment's "Augie" agent at roughly 600,000 loads, building 90% of them automatically and cutting time-to-proof-of-delivery collection by more than half, with projected productivity gains of 30% to 40%. Atlas logged over 23,500 hours across 28 Kodiak-powered autonomous trucks in the first quarter, a 110% jump from the prior quarter, hauling more than 15,000 loads. These are not mere experiments waiting for a business case. They are workflows already carrying freight. Read more here

DSG Names AI's Production Leaders

Distribution Strategy Group's first evidence-based AI benchmark screened more than 300 distributors and cleared only 26 into production, disqualifying roadmaps, pilots, and press releases. The Platinum six show what live eventually should look like for everyone. ADI Global Distribution processes roughly 1,700 orders a day with no human touch, and 44% of its customers buy through digital channels. Fastenal's 136,000 connected vending and bin-stock devices now drive 62% of sales. Ingram Micro's Xvantage platform logged more than 100,000 AI-assisted order conversions in 2025. DSG's companion survey found 93% of distributors call AI a priority, yet only 16% have crossed into production. Ambition is cheap; deployment is the moat. Read more here

Amazon Crosses One Million AI Robots

Amazon confirmed it now operates one million robots across its fulfillment network, with roughly 75% of global deliveries touched by automation at some point. The million is the headline; yet the less noticeable shift matters more. Steering that fleet is DeepFleet, a generative-AI model that coordinates robot routes and has lifted travel speed about 10%, running live across hundreds of buildings. Scale like this never arrives in a single quarter, it compounds from years of pushing pilots into production one site at a time. Mid-market operators won't match the fleet, but the sequence, automate first and add the intelligence layer second, travels down-market intact. Read more here

Lululemon Scales Robotic Fulfillment With AI

Lululemon brought a 1-million-square-foot distribution center online in Brampton, Ontario, built around an AutoStore system deployed with Element Logic. The site runs 525 R5 Pro robots against 292,000 storage bins, linked by an overhead monorail and roughly 24,000 linear feet of material handling. This one is automation-first rather than a headline AI result: Lululemon disclosed the scale, not a productivity figure, but the software orchestrating 525 robots in real time is the same intelligence layer reshaping the rest of this list. The timing is pointed. Cross-border fulfillment from Canada took a $275 million bite out of 2025 gross profit under new tariffs and the end of de minimis, and when a lane gets that expensive, throughput per square foot becomes the lever that matters. Read more here

What’s In My Ears

Warehouse Automation Strategy with Kyle McCarty and Jeff Messenger

In this episode of The New Warehouse, Kyle McCarty of Toyota Automated Logistics and Jeff Messenger of Pattern make the case for a software-first approach to warehouse automation. Their argument fits the week: hardware rarely stalls a rollout; integration and change management do. Worth a listen for how modular frameworks and virtual testing let a team model a system before committing capital, and why that discipline is what scales a deployment from one site to many. Listen here

Lily’s Quick Take

Strip away the logos and the same line runs under every story this week. A pilot is a question. Production is an answer, and only a sliver of the industry can give one yet, DSG puts it at 16%. The companies past that line, from Penske's freight agents to Amazon's million-robot fleet to Lululemon's automated dock, have less in common in which tools they bought than in how far they pushed those tools into the daily operation. Over the next year that sliver will widen, and the distance between the teams running on AI and the teams still demoing it will surface where operators feel it: service levels, margins, and who keeps the account. A pilot proves the technology can work. Production proves you can run on it.

Until next week—keep your systems learning!
— Lily @ InstaLILY AI

Thank you for reading! Have feedback? Email me, Lily, directly. I read every email.

LinkedIn • Twitter (X) • Web