CX Agents Go Live: What Lyft, Vodafone, and LATAM Airlines Learned

AI Summary

LangChain has published a detailed account of how Lyft, Vodafone, and LATAM Airlines deployed Customer Experience (CX) agents in live production environments. These are not chatbots that deflect queries to FAQ pages. They are autonomous agents that reason through customer problems, take actions across connected systems, and resolve issues end to end. The fact that three large consumer-facing companies across ride-sharing, telecommunications, and aviation are willing to share operational lessons signals that agentic AI for customer service is crossing from experimentation into real deployment. For Malaysian businesses running contact centres, e-commerce support, or any high-volume customer service operation, the practical lessons are directly applicable.

Key Takeaways

• CX agents are moving from controlled demos to live production at major companies — Lyft, Vodafone, and LATAM Airlines are running real customer interactions through autonomous agents
• The focus has shifted from "can AI answer questions?" to "can AI resolve complete customer issues across multiple steps and systems?"
• Telecom, ride-sharing, and aviation represent three of the highest-volume, most complex CX environments — if agents work here, they work in most industries
• Production deployment forces companies to solve for accuracy, system integration, and human handoff — not just model quality
• Malaysian contact centres, e-commerce platforms, and service businesses face identical CX pressures and can adopt similar agent architectures using available tools

What Happened

LangChain — the company behind one of the most widely used frameworks for building LLM applications — published a blog post documenting how three major global companies have put CX agents into production. The three companies operate in very different industries: Lyft handles ride-sharing customer support across North America, Vodafone runs telecommunications customer service across multiple European markets, and LATAM Airlines manages aviation customer operations across Latin America.

Each of these companies runs a massive customer service operation at a scale most Malaysian businesses will recognise in principle, if not in size. Lyft processes constant streams of fare disputes, lost item reports, driver-rider conflicts, and safety incidents. Vodafone serves hundreds of millions of mobile and broadband subscribers, each with different plans, languages, and regulatory frameworks depending on the country. LATAM Airlines deals with the notoriously complex world of airline customer service — flight changes, cancellations, baggage issues, seat upgrades, loyalty programme redemptions, and refund processing, where a single customer query might require the system to check booking databases, real-time flight status, loyalty programme records, and payment systems.

The LangChain post focuses on what these companies learned from putting AI agents — systems that can plan, reason, take actions across connected tools, and resolve customer problems — into live, customer-facing environments. This represents a step change from the chatbot model that has dominated customer service for the past decade. Traditional chatbots follow scripted decision trees and frequently hit dead ends that force human handoff. The agents described in the LangChain account operate with greater autonomy, working through multi-step problem resolution without a human guiding each step.

The specific technical architecture, deployment metrics, and operational details are available in the original LangChain post. What matters here is the signal: three companies known for high-volume, high-stakes customer interactions have moved beyond pilot programmes and testing sandboxes. They are running autonomous agents in front of real customers, processing real issues, and generating real operational data about what works and what breaks.

Why It Matters

The customer service industry has been talking about AI-driven support for years. Most of that talk has centred on chatbots — scripted or LLM-powered systems that answer frequently asked questions, route queries to human agents, or deflect simple requests to self-service portals. These systems have value, but they hit a ceiling fast. A chatbot can tell you your flight is delayed. It cannot rebook you on the next available flight, apply your loyalty discount, issue a meal voucher, and confirm the change — all within a single interaction.

CX agents can. That is the core distinction. An agent does not just retrieve information. It takes action. It reads the customer's request, figures out what needs to happen, checks which systems it needs to access, executes the steps in the right order, verifies the outcome, and reports back to the customer. If something goes wrong midway, a well-designed agent can detect the failure, adjust its approach, or hand off to a human with full context about what it already tried.

This matters because customer service is one of the largest operational cost centres for consumer-facing businesses. Contact centres in Malaysia employ tens of thousands of people across outsourcing firms like Teleperformance Malaysia, Scope International, and local BPO operations based in Cyberjaya, Penang, and the Klang Valley. The ability to resolve a meaningful percentage of customer issues through autonomous agents — without degrading service quality — changes the economics of customer operations.

Lyft, Vodafone, and LATAM Airlines publishing their lessons also matters because it normalises the practice. When only tech companies like OpenAI or Anthropic talk about agents, it feels distant. When a telco, an airline, and a ride-sharing platform share production experiences, it becomes a reference point that other companies can evaluate against their own operations. It moves the conversation from "should we try this?" to "how should we implement this?"

What This Means for Malaysia

Malaysia has a direct stake in this trend on two levels. First, as a major destination for contact centre outsourcing. Malaysia competes with the Philippines and India for shared services and BPO contracts. Global companies set up operations here because of the multilingual workforce — Bahasa Malaysia, English, Mandarin, Tamil — and the lower cost base relative to Singapore. If CX agents begin handling 20 to 40 percent of customer interactions that currently require human agents, the economics of the BPO sector shift. Some roles will disappear. Others will evolve into agent supervision, quality assurance, and exception handling. Malaysian BPO firms that adopt agent technology early can offer clients lower costs and faster resolution times, protecting their competitive position.

Second, Malaysian consumer-facing businesses — banks, telcos like Maxis and CelcomDigi, e-commerce platforms like Shopee Malaysia, airlines like AirAsia and Malaysia Airlines, and large retailers — face the same customer service pressures as Lyft, Vodafone, and LATAM Airlines. Customers want instant resolution at any hour. Human-only contact centres cannot scale to meet peak demand during promotional periods like Mega Sales, festive seasons, or service outages. CX agents can absorb that volume.

The regulatory environment in Malaysia also matters here. The Personal Data Protection Act (PDPA) governs how customer data is collected, processed, and stored. Any CX agent that accesses customer records, payment information, or communication history must comply with PDPA requirements. The Malaysia Digital Economy Corporation (MDEC) has been actively promoting AI adoption under the MyDIGITAL framework, and Budget 2024 and 2025 included allocations for digital transformation. Companies deploying CX agents should ensure their data handling practices — especially when using cloud-based LLMs — align with PDPA and any sector-specific guidelines from Bank Negara or the Malaysian Communications and Multimedia Commission (MCMC).

How Your Business Can Use This

Start by auditing your current customer service queries. Pull the last three months of contact centre tickets, live chat logs, and email support records. Categorise them by type: billing enquiries, order status, refund requests, appointment scheduling, technical troubleshooting, complaint escalation. Identify the top five categories that account for the largest volume of repetitive, rules-based interactions. These are your agent candidates.

Do not attempt to build a fully autonomous agent from scratch on day one. Begin with a narrowly scoped agent that handles one specific workflow — for example, processing refund requests for e-commerce orders under a certain value. The agent needs access to your order management system, your payment gateway, and your customer database. It should follow clear business rules: verify the order, check return eligibility, process the refund, and notify the customer. Build in a human handoff trigger for any edge case the agent cannot confidently resolve.

For Malaysian SMEs without in-house AI engineering teams, platforms like LangChain's framework provide the building blocks for constructing agents, and managed services from cloud providers like AWS Bedrock, Google Cloud Vertex AI, or Microsoft Azure AI Foundry offer infrastructure that handles the heavy lifting. Work with a local AI consultancy or system integrator — several have emerged in the Klang Valley and Penang tech corridors over the past two years — to scope a pilot project with clear success metrics: resolution time, customer satisfaction score, cost per interaction, and percentage of queries resolved without human intervention.

The Agentic AI Angle

The distinction between a chatbot and a CX agent is the difference between answering a question and solving a problem. A chatbot tells a customer their flight is cancelled. A CX agent rebooks the customer on the next available flight,