100,000 Walmart Deliveries Run by AI. Human Engineers Fix What It Breaks.

AI Agents Now Run the Last Mile for Walmart and Shopify

Walmart now routes 100,000 deliveries a day through an AI orchestration layer that replaced human dispatchers. The platform, built by San Francisco startup Nash, integrates once with the retailer and connects to over 1,000 carriers across 15 countries — no spreadsheets, no 2 a.m. phone calls.

But the autonomy has a catch. Every time the platform hits a carrier API failure, a rate-card mismatch, or a webhook timeout that its agents can't resolve, a human stabilization engineer takes the page. The bottleneck has moved from routing logic to the 24/7 teams keeping the orchestration layer alive, and logistics firms are hiring engineers to fill that gap.

Nash, founded in 2021 by MIT graduates Mahmoud Ghulman and Aziz Alghunaim, raised a $20 million Series A, TechCrunch reported, led by Andreessen Horowitz in July 2022, bringing total funding to $27.8 million, TechCrunch's data shows. The company spent two years replacing the spreadsheet-and-phone-call stack that powered last-mile delivery for retailers too large for a single carrier but too small to build their own logistics network. Nash now sits between the order and the doorstep for Walmart, Grubhub, Woolworths, 7-Eleven, Kroger, Square, and Urban Outfitters.

The Walmart deployment proves the model. In 2025, the retailer scaled to over 1,000 daily delivery routes, each handling about 100 stops. Onboarding across all 5,000 store locations took five weeks, finished just before peak season. "Walmart integrates once with Nash, and we handle the rest, connecting them to thousands of national, regional, and local carriers in real time," says Charles Folsom, Nash's head of customer success. Vendors on the network reported saving over two hours per route daily.

Grubhub followed a similar arc. After a quiet February 2025 pilot, the partnership expanded nationwide by May, processing thousands of orders per week. For commerce platforms, the pitch is simpler: Shopify merchants activate local delivery, same-day, scheduled, and parcel shipping through one integration — live in minutes, no build required. Toast orders flow the same way, with intelligent dispatch strategies that weigh cost, distance, reliability scores, and custom ranking rules across hundreds of providers.

The platform operates as four products doing one job. The Platform integrates any order source and manages dispatch across internal fleets, third-party logistics providers, and Nash's own carrier network. Nash AI deploys a small fleet of agentic workers — Dispatch, Support, Analyze — tuned on logistics data. They route, reassign, and answer the 2 a.m. emails. The Network provides over 1,000 pre-vetted, pre-priced, pre-plugged-in delivery providers. Tracking & Notifications handles branded links, ETA predictions, and multi-channel customer updates.

Underneath it all runs what Nash calls a Context Fabric: live signals, memory, and provenance connecting every layer at 2,340 orders per second across 12,500 drivers, targeting 98 percent on-time-in-full performance. Gartner recognized Nash as a Representative Vendor in its December 2025 Market Guide for Last-Mile Delivery Technology Solutions.

The orchestration layer works. The agents dispatch, recover, and optimize without human intervention — until they don't.

The Autonomic Platform Still Breaks, and Someone Has to Fix It

Nash calls its platform an "autonomic intelligence layer": set the objective and AI agents work until you hit it. The marketing copy promises unified decisions across fleets, carriers, and fulfillment networks. But the job description for the company's global, 24/7 Technical Support team tells a different story: the platform generates P1 incidents, stuck integrations, and unexpected provider behaviors that no agent can resolve without human intervention.

Technical Support is Nash's intake and stabilization layer for everything beyond high-volume delivery inquiries. The role owns investigation, resolution, and clean escalation of complex platform and integration issues across the delivery orchestration system. When a configuration question needs real investigation, when an API or webhook fails in production, when a carrier behaves in a way the model didn't predict — that lands on this team.

The scope is specific. Diagnose configuration questions, known error patterns, API and webhook issues, and account-specific behavior across the Nash platform. Serve as the active shift team for P1 incidents: coordinating communication, engaging Engineering immediately, and keeping customers and internal stakeholders informed through resolution. Watch Datadog dashboards, alerts, and logs for anomalies and emerging risks. Surface issues before they become customer-impacting where possible.

Escalations from this team are expected to be actionable on arrival: clear reproduction steps, investigation performed, impact assessment, and severity. The support engineer decides when to investigate further versus escalate, and documents thoroughly either way. They also act as overflow for Delivery Support during peak windows when issues exceed that team's scope.

The physical world fights back in real time. Nash's own careers page frames it as a feature: "Logistics is one of the few domains where this happens in real time. That's a different class of engineering challenge than most SaaS products offer. Your code runs against the physical world, immediately, so the feedback loop is real." The feedback loop is also the failure mode. Couriers, fleets, gig workers, parcel carriers, in-store labor, and increasingly autonomous systems all have to be coordinated in real time, against tighter windows and rising expectations, with hard-fought customer trust on the line.

Each merchant runs its own configuration, its own carrier mix, its own integration points. The platform orchestrates across Planning, Orchestrate, Fleet, and Experience products — each a surface for integration failure. A stuck integration at a 5,000-store retailer isn't a ticket; it's a revenue event.

The hiring bar reflects the complexity. Five to six years in Technical Support or Support Engineering in a SaaS environment. Hands-on experience investigating complex platform issues — not routing tickets, but actually diagnosing what is happening. Comfort with APIs, webhooks, and integration troubleshooting in production. Experience with monitoring and observability tools: Datadog, New Relic, Grafana. Clear, calm written communication during incidents. Judgment about escalation thresholds. Prior incident response or post-mortem participation. Experience supporting logistics, marketplace, or high-volume transactional platforms.

This is not a tier-one help desk role. It is a stabilization engineering function disguised as support. The autonomic platform decides; the human stabilizes when the decision hits the physical world. That shift has moved from routing algorithms to the team that keeps the orchestration layer from collapsing under its own complexity.

The Stabilization Engineer: A New Hybrid Role

The job posting does not call it "logistics support engineer." It calls it Technical Support Specialist. But the description (owning this escalation process) describes a hybrid that did not exist five years ago. Nash is hiring for that team. The role sits at the intersection of logistics operations, distributed systems debugging, and carrier API integration.

The company's workforce data shows the geographic shape of this constraint: a 24/7 stabilization team covering enterprise merchants across multiple continents needs follow-the-sun coverage. But it also needs data-residency compliance. The United States has no single federal law: requirements come from HIPAA, GLBA, state privacy statutes, and public-sector procurement terms. When a Woolworths order in Australia or a Loblaws order in Canada hits the platform, the support engineer triaging the incident may need to access logs that legally cannot leave that jurisdiction. That forces hiring in specific regions, not just time zones.

The salary band runs $53,000 to $184,000 with a median of $95,000 across five salaried roles. That spread reflects the hybrid skill set: part logistics operator who knows what a "proof of delivery" exception looks like, part engineer who can read a distributed trace and spot a stuck Kafka partition. The posting emphasizes "clean escalation", meaning the engineer resolves what they can and hands off the rest with enough context that the platform team does not waste hours reproducing the issue. That discipline is the difference between a support queue and a stabilization function.

Every new carrier integration adds a new failure mode: a rate-card mismatch, a webhook timeout, a status-code ambiguity. The AI agent retries, reassigns, and reroutes. When it cannot, a human takes the page. That human needs to understand the carrier's SLA, the merchant's promise, and the platform's internal state — all at once.

The same pressure appears across the orchestration market. Onfleet, Bringg, and Locus all run enterprise last-mile platforms with similar integration surfaces. They all need engineers who can debug a stuck delivery at 3 a.m. without waking a backend engineer. The role is not "support" in the traditional tier-one sense. It is the operational immune system for an autonomic platform. The hiring signal is clear: the focus has shifted from routing logic to stabilization capacity. The next hire you make for this team will not optimize a route. They will keep the optimization running when the physical world breaks the assumptions.

The Skill Stack Is Hybrid by Necessity

A logistics support engineer for AI fleets needs fluency in four layers that used to live in separate orgs. First, the integration layer: REST, EDI 204/214, SOAP, SFTP file drops (sometimes all in one carrier relationship). Research shows enterprise teams managing multi-carrier operations accumulate 15 to 30 point-to-point integrations, and the variation across carriers, not the complexity of any single API, drives the cost. Second, the orchestration layer: understanding how the platform's AI agents consume normalized carrier events, apply fallback routing rules, and surface exceptions to human queues. Third, the data layer: mapping status codes across carriers that use different vocabularies for the same conceptual state while preserving raw codes for audit and dispute resolution. Fourth, the operational layer: incident command for P1s during peak, shift handoffs across time zones, and the discipline to write postmortems that feed back into platform hardening.

Employers expect proficiencies in Python, TensorFlow, and PyTorch, along with ethical reasoning and clear communication with non-technical stakeholders. But the day-to-day tooling looks more like curl, jq, Postman, and a tail on Kafka topics than model training. The role is closer to a site reliability engineer who speaks logistics than a data scientist who ships code.

Compensation Reflects the Scarcity

Tech hubs including NYC offer significantly higher compensation than national averages. The spread between a platform vendor's support specialist and a robotics onsite engineer reflects the difference between triaging API failures remotely and physical robot fleet recovery, but both now require the same integration literacy.

Geography Is a Hard Constraint, Not a Preference

There is no single federal data residency law in the United States; requirements come from sectoral regulations such as HIPAA and GLBA, state privacy laws, public sector procurement terms, and contractual obligations. A 24/7 support rotation covering Walmart's 5,000+ stores and Shopify's merchant base needs follow-the-sun coverage where data can legally reside. Engineers handling P1 incidents for North American merchants must operate inside compliant enclaves: VPNs, data-local VPCs, audit-logged access. The time-zone spread is a feature; the data-residency boundary is a wall.

The architecture choice you make at three systems is the one you live with at fifteen.

That line from the integration literature applies to hiring too. The first stabilization hire sets the runbook template, the escalation paths, the on-call culture. Companies that treat this as a tier-1 support role get tier-1 outcomes: burned-out engineers, missed SLAs, merchants churning to competitors with better incident transparency. Companies that treat it as a systems engineering discipline (with rotation, blameless postmortems, and a feedback loop into the orchestration platform) build a moat.

The Constituencies Are Reorganizing Around This Role

Platform vendors (Nash, Onfleet, Bringg, Locus) are hiring stabilization engineers to absorb carrier volatility so their merchants don't have to. Enterprise logistics teams at retailers and 3PLs are hiring them to own the platform relationship, to be the internal counterpart who can pressure-test a vendor's fallback routing, demand status normalization, and audit carrier settlement matching. Carrier networks (UPS, FedEx, USPS, regional last-mile providers) are hiring them to maintain the API contracts that orchestration layers depend on, because unannounced version updates cost enterprises massive resources per hour in operational disruption. The merchants don't hire these engineers directly. They buy the outcome: a delivery promise that holds. But their procurement teams now ask for the stabilization org chart in RFPs. They want to see the on-call rotation, the mean-time-to-acknowledge, the postmortem cadence.

The Orchestration War: Four Vendors, Same Bottleneck

The last-mile delivery technology market remains extremely fragmented, Gartner's December 2025 Market Guide concluded. Most vendors still lack the full capability set required to support every fulfillment process and delivery service type. That fragmentation is the battlefield. Nash, Onfleet, Bringg, and Locus each stake a different claim, but all four have pivoted toward the same AI-agent architecture that replaces human dispatch with autonomous decisioning, and all four now face the same stabilization burden.

Locus sits at the enterprise pole. The Bangalore-headquartered platform, operated by Mara Labs and acquired by Ingka Group (IKEA's parent) in 2025, bills itself as the world's first agentic transportation management system. Its DiSCO framework runs eight specialized AI agents coordinated through a Sense-Decide-Execute-Learn cycle. Six governance mechanisms, including explainability, traceability, and human-in-the-loop checkpoints, underpin enterprise compliance. The platform evaluates more than 250 constraints per dispatch decision across 350-plus enterprise deployments in 30-plus countries, orchestrating over 1,000 carriers. Locus reports 1.5 billion deliveries powered, $320 million in transit cost savings, and 800 million miles reduced across its customer base. Its ShipFlex multi-carrier product earned Representative Vendor status in the 2026 Gartner Multi-Carrier Parcel Management Solutions Market Guide. Deployments run three to six months and integrate above existing SAP, Oracle, and custom order management systems through an API-first architecture.

Bringg occupies a different vector. Built for brands and logistics providers that must coordinate owned fleets, third-party carriers, and gig networks under a single branded customer experience, Bringg treats routing as one component of a larger orchestration system. Its native Salesforce integration and retailer-grade communication tools distinguish it from pure routing platforms. The platform evaluates which delivery provider should fulfill each order, not just how to route it.

Onfleet anchors the mid-market. Founded in 2015 and headquartered in San Francisco, the platform has powered 150 million-plus deliveries across 90-plus countries. Its strength lies in a clean REST API and webhook ecosystem that lets engineering teams embed dispatch and tracking in days rather than months. The driver app consistently ranks as the most intuitive on the market — a practical advantage because driver adoption determines data quality. Onfleet's automated SMS and email notifications with real-time ETAs cut "where is my order" calls by up to 70 percent. Enterprise plans add dedicated support reps, full data access, brand customization, and onboarding teams that compress launch from months to days. United Supermarkets cut fuel costs 45 percent; Core & Main saved over $1 million in two years. The trade-off: Onfleet typically requires a separate TMS for upstream planning, creating a visibility gap at the network level.

The segmentation is sharp. Operations pushing 500-plus deliveries per day across multiple carriers and depots gravitate toward Locus or Bringg. Mixed carrier orchestration points the same way. Salesforce shops buy Bringg; SAP or Oracle shops buy Locus or Bringg; API-first teams with custom order management systems buy Onfleet. Enterprise platforms demand eight to 16 weeks and usually a systems integrator. A 10 percent mileage reduction across a 50-vehicle fleet at $0.60 per mile pays for a $50,000 annual license.

Nash enters this fight with Walmart, Grubhub, Woolworths, and 7-Eleven on its roster. Enterprise reviewers cite strong multi-carrier orchestration, dynamic dispatch, and real-time visibility. Customers praise integration speed with OMS and front-end teams. Buyers value auto-reassignment and configurable triggers that return control over the delivery experience.

Every vendor in this tier now sells autonomy. Each of them runs a 24/7 stabilization team. The orchestration war has moved from routing algorithms to incident response, and the hiring follows.

When the next 100,000 Walmart deliveries route through Nash tonight, the AI agents will make the calls. But somewhere, an engineer will be watching the dashboards — ready to catch what the agents miss.

Working in frontier tech? Zero G Talent tracks the openings: see every open Locus Robotics role, browse frontier tech jobs, openings at Nash, and the people building the field.