Abstract
Agentic AI is the next stage in enterprise artificial intelligence, moving beyond solely using generative AI’s content creation toward orchestrated reasoning and autonomous actions, leading to measurable business outcomes. The collaboration between Amazon Web Services (AWS) and Deloitte provides a model for scaling agentic AI by combining cloud-native capabilities, industry experience, and structured methodologies. Toyota Motor North America (TMNA) demonstrates how this approach modernizes complex supply-chain operations, enhances forecasting accuracy, reduces manual workload, and elevates employees’ roles. This report examines the challenges of deploying agentic AI at scale, how AWS and Deloitte jointly helped TMNA overcome those barriers, the methodology applied to TMNA’s transformation, and the outcomes that illustrate leading practices for enterprise adoption.
Introduction
As enterprises seek to move AI from experimentation to operational impact, interest in agentic AI has grown significantly. Unlike traditional machine learning models or generative AI applications, agentic AI integrates reasoning, planning, and action-taking across business systems. It enables organizations to automate multi-step workflows, improve decision-making, and create more responsive and resilient operations. During AWS re:Invent 2025, leaders from AWS, Deloitte, and TMNA described how agentic AI is reshaping supply chain operations by reducing manual effort, improving forecasting precision, and enabling more dynamic, self-healing processes. This discussion highlighted why agentic AI is central to the next evolution of enterprise transformation and how organizations can deploy it responsibly and at scale.
The State and Challenges of Agentic AI
Enterprises exploring agentic AI face a combination of technical, operational, and organizational challenges. Many organizations struggle to define what agentic AI truly means and how it differs from prior AI approaches. According to AWS, the shift from rules-based automation and generative AI to agentic AI centers on the system’s ability to understand business problems, navigate multiple data sources, reason across constraints, and execute tasks autonomously. Yet this increased intelligence also increases complexity.
TMNA recognized that operations such as demand and supply planning relied heavily on manual processes, including more than 70 spreadsheets stitched together each month by dozens of planners. This reflects a broader enterprise challenge: legacy workflows, data silos, and fragmented systems make it difficult for agentic AI to operate effectively without foundational modernization.
Organizational readiness adds another layer of difficulty. Agentic AI influences customer interactions, employee workflows, and operational decisions. AWS emphasized that technology alone accounts for only half the transformation; the other half is people. Enterprises need to invest in skills, trust, and transparent governance so users understand when the AI should act, when humans stay in control, and how decisions are validated. Without this, AI adoption can stall due to uncertainty, resistance, or perceived risk.
Additionally, companies face a common “pilot-to-production gap”. Agentic AI prototypes often show promise but fail to scale due to a lack of architecture standards, compliance controls, and enterprise-wide adoption strategies. Deloitte discussed how businesses need a structured approach to identify high-value use cases, manage risk, and operationalize AI securely and repeatably.
How AWS and Deloitte Helped to Overcome These Challenges
AWS and Deloitte have created a solution that helps enterprises address obstacles that prevent them from scaling agentic AI. AWS brings the secure AI infrastructure, foundational models, and agentic capabilities needed for multi-step reasoning and workflow automation. Deloitte complements this with its industry experience, implementation frameworks, and ability to translate AI potential into operational results.
Deloitte outlined a three-part methodology. The first step focuses on identifying high-value use cases that deliver measurable ROI within 6 to 12 weeks. This “business-first” prioritization enables the technology to be applied where it matters most, avoiding scattered experimentation. The second step introduces a multi-agent system accelerator, jointly developed with AWS, which integrates with Amazon Bedrock and AgentCore. This environment includes compliance guardrails; Deloitte helped build one hundred controls with AWS Audit Manager to safely test agentic AI, evaluate performance, and transition use cases into production. The third step centers on measuring value through a defined framework that evaluates operational efficiency, forecasting accuracy, and other enterprise KPIs. Only use cases that meet the threshold are scaled.
This model provides the architecture, governance, and repeatability needed to bring agentic AI into mainstream business processes, reducing risk while accelerating deployment. AWS emphasized that its collaboration with Deloitte also includes joint investment and specialized AI resources aligned to customer outcomes, enabling transformations that occur in weeks rather than months or years.
Deloitte’s Methodology for Working with TMNA on Agentic AI
TMNA’s objective was to reimagine its operations horizontally, transforming supply-chain planning, responsiveness, and customer experience. Deloitte’s methodology aligned directly to these goals by embedding agentic AI into end-to-end processes rather than layering it on top of legacy workflows.
The engagement began with a clear understanding of TMNA’s business problems: complex planning cycles, manual data collection, demand volatility, and the need for more resilient supply chain networks. Deloitte and AWS worked with TMNA to build an agentic ecosystem that included a standardized platform layer, an intelligence layer for AI models, an agentic foundation for building and managing agents, and an experience layer for planners and operational teams. This architecture facilitated TMNA’s business units in avoiding reinvention of solutions and could rely on a consistent, secure, and scalable framework.
The shift to agentic workflows enabled planners to work with an AI “companion” that provided recommendations, simulated scenarios, and iterated monthly to improve outcomes. TMNA emphasized that this approach did not replace planners but elevated their role. Instead of stitching spreadsheets together, planners now oversee broader operational scope, manage more complex decision-making, and focus on higher-value work. Deloitte and AWS emphasized that humans remained in the loop, especially for cases requiring nuanced judgment or contextual awareness.
Workforce enablement played a central role. AWS highlighted its large-scale AI training initiatives and its alliance with Deloitte helped TMNA’s teams develop the skills needed to confidently and responsibly adopt agentic workflows.
Outcomes for TMNA and Lessons for Other Enterprises
TMNA’s transition to agentic AI has produced measurable improvements across operations. For supply-chain planning, the shift from seventy-plus spreadsheets and forty to fifty planners to an agent-assisted model represents a significant simplification. TMNA anticipates that as agents mature, the planning function can be handled by roughly ten planners, not through labor reduction but by enabling planners to take on broader, higher-impact responsibilities.
Forecasting accuracy increased by approximately 20%, and planner productivity improved by 18%. The agentic system’s proactive simulations also introduced self-healing capabilities. Instead of reacting to supply-chain disruptions, TMNA can anticipate issues and receive automated recommendations to maintain operations at full capacity.
In customer-facing workflows such as vehicle delivery ETA, agentic AI helped evolve TMNA’s systems from legacy green-screen lookups and swivel chair between systems, to a modern interface where agents now detect delays, resolve routine issues autonomously, and escalate only when human judgment is required. This enhances both team-member experience and customer satisfaction.
Overall, the collaboration model, TMNA’s operational expertise, Deloitte’s transformation skillset, and AWS’s AI platform have created shared ownership, shared risk, and shared outcomes. TMNA confirmed it would have moved more slowly and scaled more slowly without the combined strengths of AWS and Deloitte.
Our Angle
The collaboration among TMNA, AWS, and Deloitte highlights essential practices for enterprises pursuing agentic AI transformation. Organizations should begin with clearly defined business problems rather than technology-first thinking. Agentic AI is most successful when targeted at operational bottlenecks, repetitive tasks, and data-intensive workflows. Transformation must include the workforce from the beginning, with AI being more of a ‘buddy’ than an agent, and employees getting the needed training, clarity, and confidence to adopt AI-augmented roles.
Enterprises should also establish a standardized architectural foundation with built-in security, compliance, and governance to provide trust and transparency as base requirements. This prevents fragmented AI efforts and ensures scalability. Starting with narrow use cases allows teams to validate value, build trust, and iterate safely before scaling to multi-agent systems. Finally, success requires rigorous measurement, ongoing refinement, and cross-functional collaboration among operations, technology, and leadership teams.
When these principles are followed, agentic AI becomes not just an innovation initiative but a sustainable engine for operational excellence, resilience, and enterprise modernization.
Feel free to reach out and stay connected through robs@siliconangle.com, read @realstrech on x.com, and comment on our LinkedIn posts.
Disclosure: TheCUBE is a paid media partner for Amazon Web Services and Deloitte, the sponsor of theCUBE’s event coverage, and neither Amazon Web Services nor Deloitte nor other sponsors have editorial control over content on theCUBE Research, theCUBE, or SiliconANGLE.
