Machine-readable, then machine-run
Every industry AI has transformed, it first made machine-readable. Text became tokens, and language models followed. Code became repositories, and coding agents followed. Physical retail — twenty trillion dollars of it — still runs on walkthroughs, spreadsheets, and gut. It is the largest domain AI has never seen from the inside.
That is the void dg2n fills, and this week we compressed the entire thesis into a thirty-second film. The exercise turned out to be more than marketing: forcing a company's argument through a thirty-second aperture is a brutal test of whether the argument is actually one idea or several stapled together.
The compression test
The temptation with a product spanning store design, space planning, planogramming, and floor execution — across stores, dark stores, and warehouses — is to enumerate. Four suites, three facility types: a feature list masquerading as a story. The resolution was to never name the suites at all. They appear only as verbs in a single breath — designing stores, planning space, placing products, executing the floor — because the product is not four tools. It is one loop.
The deeper move was choosing the category noun. "Digital twin platform" describes the means; "AI operating layer" describes the business. Twins are how AI comes to see a store. Operations are what it does once it can. The twin is infrastructure; the agents are the product; the destination is facilities that sense, decide, and act on their own.
One agent, one shelf
Abstractions do not survive thirty seconds without a witness. The narrative needed one concrete moment where the loop becomes visible — a single agent decision, told as signal, decision, action, outcome:
Somewhere right now, an agent is noticing a shelf no human would check until next week — and fixing it.
The line does double work. It dramatizes tirelessly, and — because "somewhere right now" implies thousands of parallel instances — it makes the scale claim felt rather than stated. Five thousand facilities is a statistic; one shelf being quietly fixed at 2 a.m. is a fact you can picture. Placed after the number, the story becomes a dive from the aggregate into one live point of light.
The narration
The locked cut:
Twenty trillion dollars of retail still runs on walkthroughs, spreadsheets, and gut.
dg2n is the AI operating layer for physical retail. Every store, dark store, and warehouse — a living digital twin AI can finally see.
And on that foundation, agents take over the work: designing stores, planning space, placing products, executing the floor — sensing, deciding, acting. Tirelessly. 5,000-plus facilities already run on agentic AI on dg2n — with efficiency gains of 15 to 45 percent.
Somewhere right now, an agent is noticing a shelf no human would check until next week — and fixing it.
dg2n. Agentic AI for brick-and-mortar retail.
Why this is a substrate argument
The pattern underneath is the same one that runs through everything I build: construct the missing abstraction layer rather than optimize within the existing ones. Retail software has spent two decades optimizing transactions while the physical substrate — the geometry of shelves, fixtures, and floor plates where those transactions actually happen — remained illegible to machines. Make the substrate machine-readable and an entire class of agentic operation becomes possible above it, the way form-finding became computable once shells became geometry instead of intuition.
Machine-readable, then machine-run. The order matters. Reading is the foundation; running is the point.
— Utsav