In Uber Chief Technology Officer Praveen Neppalli’s sharing on Agentic AI practices, what truly deserves attention is not the impressive figures themselves — such as “99% of engineers using AI,” “70% of Pull Requests involving AI Agents,” or “over 2,500 Agent Skills” — but the fact that Uber has achieved a paradigm shift in enterprise AI application: moving from tool-centric AI solutions to a workflow-centric Agentic Enterprise.
This practice not only signals that enterprise AI has entered its second stage of development but also offers highly valuable methodologies for future digital transformation.
From AI Coding to AI Operating: Uber Is Rebuilding Its Operating Model
Over the past two years, most enterprises’ application of Generative AI has remained at the Copilot stage.
AI has primarily served as:
- Coding Assistant
- Writing Assistant
- Knowledge Search Assistant
- Customer Service Chat Assistant
While these tools have significantly boosted individual productivity, they essentially belong to Task Automation.
Uber’s approach is fundamentally different. Instead of merely assisting employees, AI now takes on multiple critical roles across entire business processes.
Engineers are no longer just building chatbots; they are developing intelligent Agents capable of:
- Understanding business context
- Autonomously calling tools
- Performing complex reasoning
- Executing tasks across systems
- Continuously learning and optimizing
This marks a profound shift in enterprise AI focus — from “How do we use AI?” to “How should the company operate if AI is embedded in every workflow?”
It represents a reconstruction of the enterprise Operating Model, far beyond a mere software capability upgrade.
Workflow, Not Task: The New Unit of Automation
Uber puts forward a highly enlightening perspective:
“Workflow becomes the unit of automation — not the individual task.”
This is the most thought-provoking statement in the entire case for CIOs, CTOs, and digital transformation leaders.
Traditional Business Process Management (BPM) typically follows a linear structure:
Task A → Task B → Task C → Task D
Automation is achieved through RPA, workflow engines, or API integrations on specific steps — such as auto-generating reports, filling forms, approving requests, or sending emails. These are all localized optimizations.
In contrast, Agentic AI redesigns the entire process:
Business Goal
↓
Agent understands context
↓
Plans execution path
↓
Calls multiple systems
↓
Makes dynamic judgments
↓
Provides continuous feedback
↓
Achieves business objective
The object of automation has evolved from individual actions (Tasks) to complete business goals. This is the fundamental difference between Workflow AI and traditional workflow engines.
Agentic Pods: A Replicable Model for Enterprise AI Implementation
Uber’s most valuable innovation lies not in the Agents themselves, but in the organizational approach — the creation of Agentic Pods.
Each Pod consists of AI engineers + business domain experts, forming a temporary cross-functional innovation team that completes a full iteration cycle within two weeks:
- Day 1-2: Shadowing — observing real work scenarios
- Day 3: Identifying genuine business problems worth solving
- Day 4-5: Rapid Agent development
- Day 6-9: Real-environment validation
- Day 10: Deployment
This model integrates Design Thinking, Lean Startup, Agile Sprint, Domain-Driven Design (DDD), and AI Engineering, forming a brand-new AI innovation framework.
It effectively addresses the most common reason for AI project failure in enterprises: the disconnect between development teams that lack business understanding and business teams that lack AI expertise — often resulting in little more than “ChatGPT with a corporate logo.”
Through a Pairing approach, Uber enables AI engineers to immerse themselves deeply in frontline operations, uncovering tacit knowledge that cannot be documented — why approval is required here, why waiting is necessary, why human judgment is mandatory, or why Excel exports are essential. These nuances ultimately determine whether AI can deliver real value.
AI Is Not Just About Efficiency — It Uncovers New Business Opportunities
Uber discovered that the most surprising outcome was not efficiency gains, but the emergence of new opportunities.
After entering unfamiliar business domains, AI engineers quickly identified previously overlooked issues. This stems from a fundamental shift in their thinking:
Traditional business analysis focuses on: Existing processes → Identifying automatable steps.
Agentic thinking starts from: Business goal → Why do we do it this way? → Can we do it entirely differently? → What new roles can AI play?
For example, in fund allocation processes, the traditional approach relied on manual collection, Excel processing, multi-level approvals, and adjustments. After AI redesign, multiple Agents perform real-time analysis, dynamic forecasting, and automatic solution generation, with humans only providing final confirmation. The entire business model is transformed.
AI is no longer merely “replacing humans” — it is redefining how work is organized.
The Greatest AI Dividend for Enterprises: Digitizing Tacit Knowledge
Uber emphasizes that the best AI opportunities are invisible from the outside. One must sit beside business personnel, observe, ask questions, and co-create.
This aligns with the classic concept in knowledge management — Tacit Knowledge. A company’s true competitive edge has never been its ERP, CRM, or OA systems, but the “know-how” residing in its employees’ minds.
In the past, such knowledge was difficult to digitize. Today, the combination of LLM + Agent + Memory + Knowledge Graph + RAG has, for the first time, enabled tacit knowledge to be structured, expressed, reasoned, and executed.
The core assets of enterprise AI are quietly shifting:
- Past: Data → Information → Knowledge
- Future: Knowledge → Workflow → Decision → Action
AI is bridging the long-standing gap between knowledge and execution.
Four Stages of Enterprise AI Evolution Revealed by Uber’s Case
Based on Uber’s practice, the maturity path of enterprise Agentic AI can be summarized into four stages:
| Stage | Core Capability | Business Value |
|---|---|---|
| AI Assistant | Human-AI collaboration, personal productivity | Improved office productivity |
| AI Copilot | Assisting complex tasks | Enhanced professional efficiency |
| AI Agent | Autonomous planning and execution of business processes | Cross-system process automation |
| AI Organization | AI-driven organizational collaboration redesign | Comprehensive upgrade of enterprise operating model |
Uber has progressed from the AI Copilot stage into AI Agent and is now advancing toward AI Organization.
In the future, the focus of enterprise competition will shift from “who owns more AI tools” to “who can build an Agent network covering the entire business process.”
Implications for Building Enterprise Intelligent Platforms
For teams developing enterprise-grade intelligent platforms (such as HaxiTAG Studio and EiKM Agentic AI platforms), Uber’s practice offers several key insights:
AI’s core focus must expand from “documents and knowledge” to “workflows”, creating a unified intelligent closed loop of knowledge, rules, data, tools, and execution capabilities — rather than merely building Q&A bots.
Establish a business-oriented Agent Factory. Through standardized Agent templates, tool connectors (MCP), knowledge bases (RAG), memory systems, and workflow orchestration, enterprises can rapidly incubate and scale Agents while accumulating reusable Agent Skills.
Adopt deep collaboration between business experts and AI engineers as the core innovation mechanism. Uber’s Agentic Pods demonstrate that the most valuable innovations arise from the profound integration of business knowledge and AI capabilities, rather than technology-driven efforts alone.
Build an enterprise-grade Agent governance system. As the number of Agents grows rapidly, unified management of permissions, knowledge sources, tool calls, model selection, execution auditing, and performance evaluation is essential to ensure security, controllability, and continuous optimization.
Agentic AI Is Becoming the New Operating System for Enterprise Digital Transformation
Uber’s practice shows that the value of Agentic AI goes far beyond improving efficiency or reducing costs. It is redefining how enterprises organize knowledge, design processes, and make decisions.
AI is no longer an add-on productivity tool layered on top of existing systems — it is emerging as a new enterprise operating system that connects knowledge, data, processes, decisions, and execution.
Forward-looking enterprises will not settle for embedding AI into current processes. Instead, they will boldly redesign workflows, organizational models, and business operations around AI — making workflow the fundamental unit of intelligent automation and Agent the new digital workforce.
What Uber has demonstrated is not merely a successful AI application case, but a preview of the next phase of enterprise intelligence: Agent as the execution enity, workflow as the intelligent unit, knowledge-driven decision-making, and organizational restructuring to unleash productivity.
This paradigm is expanding from engineering and R&D into core functions such as finance, legal, marketing, supply chain, and human resources. It will gradually form the foundational architecture of enterprise intelligent operations, establishing a new competitive advantage for the future Agentic Enterprise.