Rivian

Prospect Profile

Rivian ships five vehicles from two factories across three overlapping programs: R1T/R1S (premium adventure EVs, ~42K units/year from Normal, Illinois), R2 (mass-market SUV, production started April 2026, targeting 155K units/year by 2027), R3 (compact crossover, Georgia Stanton Springs facility, 2028), and the EDV delivery van for Amazon. They're past the "will this work" phase of R1 ramp and deep in "how do we run three programs at cost targets that sustain profitability."

The R2 is the test case. Designed clean-sheet for unit economics: 50% lower manufacturing cost than R1, achieved through part consolidation (2.3 miles of wiring harness eliminated), structural simplification (high-pressure die-casting cuts 2,000 lbs), and 72% lower suspension assembly cost. That kind of cost discipline decomposes into requirements — all the way down to which CAN message interval saves $45 in BMS silicon.

Systems Engineering Surface

Rivian owns the full stack. Battery cells, drive units, power electronics, the skateboard platform — no Tier 1 black boxes for critical subsystems. A single propagated change cascades through every domain they own. A new cell chemistry with 5% higher nominal voltage triggers re-verification of converter efficiency (3-4 hours FEA on AWS HPC), pack thermal dissipation (COMSOL re-run, 2 hours), BMS firmware state machine transitions (12-hour regression), vehicle electrical schematics, ISO 26262 ASIL-D traceability, and certification evidence across EPA, NHTSA, and DOT standards.

AWS case studies reveal something useful: Rivian's simulation infrastructure is already CI/CD-connected. SimOS chains thermal, structural, and electrical models on commit hooks. The bottleneck isn't compute capacity — it's requirements traceability velocity.

Pain Points

Faster iteration. R2 cost-down requires 10-15 design iterations before architecture freeze. Each V-cycle currently takes 3-4 weeks because requirement changes batch through DOORS or email-threaded ICD negotiations. Compress that to 90 minutes and you run 10 iterations in the time you used to run two.

Fewer integration surprises.Pack-to-vehicle is Rivian's historical bottleneck. The BMS firmware has a 500ms over-temperature shutdown response. The vehicle safety logic expects 50ms. The mismatch shows at V3, after mechanical assumptions are baked into the platform. With Flow, cross-domain timing conflicts surface before anyone has written code.

Greater confidence at reviews.When the VP asks "is R2 ready for Design Freeze Review?" the answer currently requires aggregating sign-off from 12+ subsystem leads, each with their own tool. Flow makes requirement status, analysis margin, test completion, and the open change-request pipeline one query.

Flow Integrations at Rivian Scale

Thermal simulation as a gated requirement check. When a systems engineer proposes an ambient temperature change for a new market (52°C for Middle East), Claude queries the linked COMSOL model, triggers a re-run via AWS API, reads the result, and proposes three restoration paths with cost and weight tradeoffs. Feedback loop: 90 minutes, not three weeks.

CI test results as verification run ingestion. The drive-unit dyno runs a 15-minute constant-torque test every commit. Results go to S3. A Flow automation watches the bucket and creates a verification test run linked to the relevant motor performance requirements automatically.

R2-to-R1 requirement reuse.Claude clones the R1 requirement tree into R2 as a baseline, flags which requirements need re-verification given the reduced cell count, and proposes new target values. Rivian's thermal lead reviews 40 flagged requirements instead of writing 247 from scratch.

Cross-domain CAN conflict detection. Pack BMS requires cell data over CAN every 10ms. Vehicle diagnostics expects pack state-of-health every 50ms. Same bus, timing conflict. Claude traverses the requirement graph across domains and opens an ICD change request before anyone has written firmware.