Logistics and shipping giant FedEx’s 2022 operating income reached $6.25 billion. It has yet to meet, or beat, that number since. In 2023, FedEx’s operating income fell to $4.91 billion, then grew to $5.56 billion in 2024, only to fall again in 2025 to $5.22 billion. This year, Vishal Talwar, chief digital and information officer at FedEx, has a plan to help the company return to, and exceed, those 2022 levels. That plan is to transform over half of its core operational workflows, including last-mile visibility and customer experience, network operations, and, increasingly, software development with agentic AI.
“There’s a lot of noise around AI. What matters is how it’s applied,” Talwar stated on a March 10 LinkedIn post. “At FedEx, we focus on practical, purposeful use, driving safer operations, greater reliability, and smarter decision-making at scale,” he added.
Such a big push into agentic AI is quite the structural lift for a company with 500,000 employees and $87.9 billion in annual revenue. And this latest commitment to AI is an extension of FedEx’s DRIVE efficiency program, announced in December 2022 as part of a multi-year structural cost-reduction initiative. Under DRIVE, FedEx initially targeted more than $4 billion in structural cost reductions by the end of 2025. DRIVE hit its goals. FedEx achieved $1.8 billion in permanent savings in fiscal 2024 and an additional $2.2 billion in fiscal 2025.
The AI agent workforce initiative continues FedEx’s efforts toward increasing efficiency. At the February 2026 Investor Day, FedEx set a target of $8 billion in operating income by fiscal 2029, with AI and automation explicitly named as the primary mechanism for a 53% increase over 2025.
Govern the Data, then Master the AI
For any ambitious agentic AI initiative to work, it needs to operate on a solid foundation of data. For FedEx, that’s Atlas. Atlas is an enterprise data platform built on Databricks and Unity Catalog, consolidating 600 independently managed analytics environments into a single, governed architecture that houses over 1,300 data tables and serves 2,800 enterprise users. Every scan, route decision, and customer interaction across FedEx’s network generates two petabytes of data daily. Until that data is normalized, catalogued, and structured, no agent can act on it reliably.
The timeline Talwar sequenced at FedEx’s Investor Day was deliberate: simplify processes, digitize workflows, embed AI. Atlas is the second step, and it’s still underway: FedEx projects 100% data consolidation into Atlas by the end of 2027, one year before its 2028 AI agent workflow target.
According to a recent WSJ story, FedEx’s agentic AI architecture is built as a hierarchy rather than a single autonomous system. Manager agents handle orchestration: breaking complex logistics problems into discrete tasks and assigning them to specialized worker agents. Worker agents execute tasks such as rerouting a shipment around a weather disruption, resolving a customs exception, generating a freight quote, or flagging an anomaly in a distribution center’s throughput. Audit agents run in parallel, monitoring outputs for accuracy and compliance before decisions move downstream. The entire structure operates on top of Atlas, which feeds agents with normalized, governed data from tracking scans, weather feeds, capacity data, and customer history, so that decisions are grounded in the current state rather than old inputs.
The ServiceNow partnership, announced in October 2025, enables agents to connect to workflow execution. ServiceNow’s platform handles process orchestration, such as routing agent outputs to the correct operational workflows, triggering human escalations when agents hit decision thresholds they aren’t authorized to cross, and maintaining the audit trail that regulated logistics environments require. Talwar’s AI team at FedEx has also embedded agents into software development, where agents write, test, and review code, enabling FedEx to scale its AI capability while managing related headcount costs.
In his March 10 LinkedIn post, Talwar wrote that predictive AI tools are already live in 41 U.S. facilities, and are preventing 17,000 hours of potential operational downtime.
If FedEx can establish agents that can absorb a meaningful share of the decision volume currently handled by human workers, without a proportional increase in error rates, the unit economics of moving a package improve significantly. The initiative is partly a competitive response to Amazon, which built its logistics infrastructure to run on AI systems natively and doesn’t carry the legacy architecture burden of traditional carriers.
Much of the logistics industry is moving, at varying speeds, in the same direction. DHL launched its HappyRobot AI agent platform across care centers in November 2025, claiming automated email and voice minutes processing. C.H. Robinson has the most granular public metrics of any logistics company: more than 30 deployed agents handling freight quoting, order processing, and LTL classification, with its Quoting Agent delivering quotes in 32 seconds, its Orders Agent processing 5,500 truckload orders per day, and 75% of LTL orders now fully automated — up from 50%. The company reports a 40% increase in productivity since 2022. Walmart’s Sparky AI agent is now used by roughly half of all app users, with those customers showing 35% higher order value.
The market for AI in logistics is projected to grow from $15.28 billion in 2024 to $306.76 billion by 2032, at a 42% compound annual growth rate, driven by continued growth in e-commerce, persistent labor shortages, and demand for real-time supply chain visibility. A February 2026 survey of 830 IT decision-makers by Futurum found agentic AI rising 31.5% year-over-year as an enterprise technology priority, with a direct financial impact on revenue and profitability, nearly doubling as the primary ROI measure. The “we’re saving workers four hours a week” argument, Futurum notes, is losing procurement conversations.
The AI Execution Success Record Remains Murky
These logistics AI investments do run headlong into a sobering body of research. McKinsey’s 2025 State of AI survey — 1,993 respondents across 105 nations, weighted by GDP contribution — found that only 39% of organizations report any bottom-line impact from AI at the enterprise level, and most of those attribute less than 5% of operating profit to it. Only approximately 6% qualify as high performers, defined as organizations where AI demonstrably moves the profit needle. BCG’s 2026 supply chain planning analysis lands in the same place: only 20% of organizations report real value from AI investments in supply chain planning, and fully autonomous planning remains an aspiration rather than a present operational state.
Gartner placed agentic AI at the Peak of Inflated Expectations on its most recent Hype Cycle and predicted that more than 40% of agentic AI projects would be canceled before reaching production by 2027, driven by rising costs, governance failures, and unclear business value. A Carnegie Mellon study published in June 2025 found that AI agents complete multi-step tasks correctly only 30-35% of the time in a simulated office environment — and that most deployed models demonstrate near-zero confidentiality awareness, a meaningful liability in any environment handling sensitive shipment, pricing, or customer data.
FedEx’s 2028 target is ambitious against that backdrop. A 50% workflow automation commitment requires not just functional agents but reliable data pipelines, governance frameworks that didn’t exist two years ago, a workforce that adopts rather than routes around autonomous systems, and a security posture that accounts for what happens when an agent with access to shipment routing systems receives a manipulated input. None of those are solved challenges in logistics or any industry.
The companies that will close the gap between announcement and delivery share one trait the McKinsey data makes clear: they redesigned workflows before deploying AI into them, not after. For an industry running on decades-old operational architecture, that sequencing discipline is harder to sustain than the agent deployment itself.
