Ex-Tesla Operators Raise $12.5M to Automate Inventory Decisions
Atomic has raised a $12.5 million Series A led by Klass Capital and Madrona Venture Group, bringing total funding to just north of $15 million. TechCrunch The Boston-based company disclosed the round on Sept. 29.
The company was founded by Michael Rossiter and Neal Suidan, former Tesla supply chain leaders, with Rossiter as chief executive and Suidan as chief product officer. It was incubated at DVx Ventures, the venture studio founded by former Tesla president Jon McNeill, who sits on Atomic's board. Jeff Goodrich, a longtime Tesla planning director, later joined as chief technology officer and third co-founder.
Atomic's software decides how much inventory a company should have and where to hold it by simulating scenarios, then recommending a response or automatically choosing one. Named customers include DoorDash and HelloFresh.
Deployment depth is clearest at DoorDash. McNeill said DoorDash runs about 90% of its purchasing across hundreds of sites on Atomic. He also said Atomic's annual recurring revenue quintupled since the beginning of 2026. TechCrunch
The founders trace the approach to Tesla's 2018 Model 3 production ramp, where they built an early version of the system. Atomic previously closed a $3 million seed round. National Law Review
The broader context here is familiar to anyone who has operated planning tools. Traditional stacks separate forecasting, requirements planning and replenishment, with planners reconciling the gaps in exception queues and spreadsheets. Atomic collapses forecasting and action into a simulation loop closed by policy, either propose or execute. The trust boundary matters. Recommendation preserves control but keeps toil. Full automation removes toil but requires versioned policies, audit trails and clear limits on what the system may do without review. For expert buyers, the integration surface and data hygiene decide success more than solver math, since purchase automation is only as clean as the item, vendor and on-hand data feeding it.
Looking at what this means for technology buyers, steady-state accuracy is rarely the hard part. Most planning engines handle stable demand. Strain appears when lead times slip, promotions spike, suppliers short shipments, or perishability blocks simple rebalancing across nodes. Meal-kit and delivery networks live with those constraints daily, which makes them a demanding proving ground for closed-loop control. One point worth flagging is that autonomy also concentrates failure modes, so rollback, manual override and per-site limits become core requirements rather than extras. In this author's view, the payoff worth tracking is not fewer dashboards but fewer routine decisions. If simulation plus guardrailed execution holds up, planners shift from approving replenishment orders to supervising exceptions and tuning policies. Over the long arc, that pattern tends to stick when it reduces waste and stockouts without adding operational fragility. That is a quieter change than autopilot language suggests, and for operations teams, a more useful one.
