Atomic raises $12.5M; DoorDash buys 90% autonomously

DoorDash lets software place about 90% of the purchase orders for its convenience business. That is the number underneath Atomic's $12.5 million Series A, announced September 29. Autonomous inventory purchasing stopped being a pitch deck this year, and the reason it works has less to do with the model than with something most stores have never done.

What actually happened

Atomic raised $12.5 million led by Klass Capital and Madrona, bringing total funding past $15 million. Adrian Schauer of Klass and Matt McIlwain of Madrona join the board. The founding team came out of Tesla supply chain: Michael Rossiter (CEO), Neal Suidan (chief product officer) and Jeff Goodrich (CTO), a former Tesla planning director. The origin story is a prototype they built during the 2018 Model 3 production ramp, when planning changes outran spreadsheets.

The product simulates supply chain scenarios, recommends inventory levels and placement across distribution sites, and executes purchasing decisions autonomously. Customers named in the announcement: DoorDash's DashMart, Good Chop (HelloFresh), Starface World, OOFOS, LMNT and Vincero. Atomic says Good Chop cut inventory in half while more than doubling revenue. ARR has quintupled since the start of 2026, per TechCrunch.

The detail worth the whole article: Atomic's system works out a company's decision rules even when those rules were never documented. Rossiter's framing of the optimization problem — "AI can play the role of finding all of the best paths through that forest."

Why autonomous purchasing matters for your business

Purchasing is the highest-leverage automation target in a physical-goods business and almost nobody attacks it, because getting it wrong costs cash. Over-order and you have paid for a pallet you will discount in March. Under-order and you stock out on your best SKU during your best week. So it stays with the one person who "knows," and it stays a spreadsheet.

Here is what Atomic's customer results actually demonstrate: the decision was always rule-based. The person doing it by feel was running a model — reorder points, lead time buffers, seasonal multipliers, the vendor who is always two weeks late — they just never wrote it down. Software that infers those rules and executes them is not smarter than your buyer. It is faster, and it never forgets the vendor who is always late.

Which makes the precondition clear. You do not need an AI vendor to start:

Write your reorder rules down this week. For your top twenty SKUs: reorder point, order quantity, supplier lead time, and the reason each number is what it is. If the reason is "that's what we've always done," that is the SKU to look at first.

Find out whether your own data can support the decision. Pull 24 months of unit sales by SKU by week. If you cannot get that out of your systems in under an hour, your data plumbing is the project, not the agent.

Set the autonomy threshold in dollars, not in confidence. DoorDash runs 90% autonomously, not 100%. The 10% that stays human is almost certainly the large, unusual and irreversible orders. Pick your line — say, anything under $2,000 with a known supplier goes automatically — and hold it.

Keep the rules portable. Reorder logic encoded in a vendor's platform is logic you rent. Written down, it survives your next system migration and it is the asset that makes any tool work faster.

The pattern generalizes past inventory. Any recurring decision a person makes by feel, on data you already have, with a bounded cost when wrong, is automatable now. Purchasing just happens to be the one with a number attached.

Sources: PR Newswire, TechCrunch.