--- title: Customer Experience (CX) Agents in Production: Lessons... description: Three companies — Lyft, Fastweb+Vodafone, and LATAM Airlines — share how they built and operate AI customer experience agents in production. Lyft created a... canonical: https://daily.dev/posts/customer-experience-cx-agents-in-production-lessons-from-lyft-vodafone-and-latam-airlines-hiwilqp5n twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Customer Experience (CX) Agents in Production: Lessons from Lyft, Vodafone, and LATAM Airlines | daily.dev og:description: Three companies — Lyft, Fastweb+Vodafone, and LATAM Airlines — share how they built and operate AI customer experience agents in production. Lyft created a... og:url: https://daily.dev/posts/customer-experience-cx-agents-in-production-lessons-from-lyft-vodafone-and-latam-airlines-hiwilqp5n og:image: https://api.daily.dev/og/posts/HIWilQp5n.png og:image:alt: Customer Experience (CX) Agents in Production: Lessons from Lyft, Vodafone, and LATAM Airlines og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Customer Experience (CX) Agents in Production: Lessons from Lyft, Vodafone, and LATAM Airlines **[LangChain](https://daily.dev/sources/langchain)** · 24 min read · 0 upvotes · 0 comments ## Summary Three companies — Lyft, Fastweb+Vodafone, and LATAM Airlines — share how they built and operate AI customer experience agents in production. Lyft created a self-serve platform using LangGraph that reduced agent development time from six months to two weeks, with a 65% deflection rate. Vodafone's Super TOBi serves 9.5 million customers with a 90% correctness rate, while Super Agent helps call-center consultants achieve 86%+ one-call resolution. LATAM's Concierge agent serves 4,000 daily users and its Compass pipeline converts unstructured conversations into structured business intelligence stored in BigQuery Graph. Key lessons: prompt quality becomes the bottleneck when non-engineers build agents, observability must be in place before launch, architectures evolve through production use, and conversations are a valuable source of business intelligence. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### How did Lyft reduce the time to build a new customer support agent from six months to two weeks? Lyft split agents into two categories: specialized agents built by ML engineers for complex workflows like damage claims, and configurable agents initialized at runtime from a JSON configuration and a prompt stored in LangSmith's Prompt Hub, which domain experts can write without engineering help. This self-serve layer, combined with automated checks catching contradictory instructions, cut build time from roughly six months to about two weeks. _daily.dev surfaces engineering deep dives like this for teams designing self-serve AI agent platforms._ ### Why did LATAM Airlines redesign its Concierge travel agent from a triage architecture to a supervisor architecture? Production traces in LangSmith showed the original triage-based design, where each specialist agent independently formatted its own structured output, was creating roughly 15% overhead in latency and token consumption even when no downstream component needed that structure. LATAM redesigned the system so a supervisor stays in control and structures the final output only once, preserving quality while cutting cost by about 15%. _engineers debugging agent architecture inefficiencies can find production case studies like this on daily.dev._ ### How does Lyft make its LLM-based evaluation judges reliable for grading customer support agent conversations? Lyft calibrates LLM judges against human reviewers by collecting human labels and iterating on each judge until it reaches a sufficiently high agreement rate with those labels. The team also moved from generic scalar quality scores to narrow, behavior-specific pass-or-fail rubrics, and fine-tuned its simulated test users on real customer verbatims because early LLM-generated customers were too articulate and cooperative, producing offline pass rates above 90% that didn't match production. _daily.dev helps teams building LLM evaluation pipelines keep up with real-world calibration techniques._ ## Similar posts on daily.dev - [Fastweb \+ Vodafone: Transforming Customer Experience with AI Agents using LangGraph and LangSmith](https://daily.dev/posts/fastweb-vodafone-transforming-customer-experience-with-ai-agents-using-langgraph-and-langsmith-ruvglrcz6) · LangChain · 4 upvotes · 0 comments - [Agents Work. Now the Hard Part: Notes from LangChain…](https://daily.dev/posts/agents-work-now-the-hard-part-notes-from-langchain--cmz5x6chx) · 8th Light · 1 upvotes · 1 comments - [Building AI agents for 127 million customers](https://daily.dev/posts/building-ai-agents-for-127-million-customers-yk2dcgusz) · Nubank · 1 upvotes · 0 comments - [Scaling Agents in Europe & The Middle East: Lessons from Schneider Electric, Vodafone, and monday.com](https://daily.dev/posts/scaling-agents-in-europe-the-middle-east-lessons-from-schneider-electric-vodafone-and-monday-co-owwihwiar) · LangChain · 0 upvotes · 0 comments --- Tags: [#llm](https://daily.dev/tags/llm), [#ai-agents](https://daily.dev/tags/ai-agents), [#langgraph](https://daily.dev/tags/langgraph), [#langsmith](https://daily.dev/tags/langsmith) [View this post on daily.dev](https://daily.dev/posts/customer-experience-cx-agents-in-production-lessons-from-lyft-vodafone-and-latam-airlines-hiwilqp5n) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Customer Experience (CX) Agents in Production: Lessons from Lyft, Vodafone, and LATAM Airlines","url":"https://daily.dev/posts/customer-experience-cx-agents-in-production-lessons-from-lyft-vodafone-and-latam-airlines-hiwilqp5n","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/customer-experience-cx-agents-in-production-lessons-from-lyft-vodafone-and-latam-airlines-hiwilqp5n"},"datePublished":"2026-08-04T16:04:12.252Z","dateModified":"2026-09-13T21:57:22.455Z","description":"Three companies — Lyft, Fastweb+Vodafone, and LATAM Airlines — share how they built and operate AI customer experience agents in production. Lyft created a...","image":"https://media.daily.dev/image/upload/s--P4t4XyoV--/f_auto/v1722860399/public/Placeholder%2001","thumbnailUrl":"https://media.daily.dev/image/upload/s--P4t4XyoV--/f_auto/v1722860399/public/Placeholder%2001","isAccessibleForFree":true,"articleSection":"LangChain","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"LangChain","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/0f4f2629e0724a849823f8cd0d913e13","url":"https://daily.dev/sources/langchain"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/customer-experience-cx-agents-in-production-lessons-from-lyft-vodafone-and-latam-airlines-hiwilqp5n","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"llm,ai-agents,langgraph,langsmith","timeRequired":"PT24M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"LangChain","item":"https://daily.dev/sources/langchain"},{"@type":"ListItem","position":3,"name":"Customer Experience (CX) Agents in Production: Lessons from Lyft, Vodafone, and LATAM Airlines"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/customer-experience-cx-agents-in-production-lessons-from-lyft-vodafone-and-latam-airlines-hiwilqp5n#faq","mainEntity":[{"@type":"Question","name":"How did Lyft reduce the time to build a new customer support agent from six months to two weeks?","acceptedAnswer":{"@type":"Answer","text":"Lyft split agents into two categories: specialized agents built by ML engineers for complex workflows like damage claims, and configurable agents initialized at runtime from a JSON configuration and a prompt stored in LangSmith's Prompt Hub, which domain experts can write without engineering help. This self-serve layer, combined with automated checks catching contradictory instructions, cut build time from roughly six months to about two weeks. daily.dev surfaces engineering deep dives like this for teams designing self-serve AI agent platforms."}},{"@type":"Question","name":"Why did LATAM Airlines redesign its Concierge travel agent from a triage architecture to a supervisor architecture?","acceptedAnswer":{"@type":"Answer","text":"Production traces in LangSmith showed the original triage-based design, where each specialist agent independently formatted its own structured output, was creating roughly 15% overhead in latency and token consumption even when no downstream component needed that structure. LATAM redesigned the system so a supervisor stays in control and structures the final output only once, preserving quality while cutting cost by about 15%. engineers debugging agent architecture inefficiencies can find production case studies like this on daily.dev."}},{"@type":"Question","name":"How does Lyft make its LLM-based evaluation judges reliable for grading customer support agent conversations?","acceptedAnswer":{"@type":"Answer","text":"Lyft calibrates LLM judges against human reviewers by collecting human labels and iterating on each judge until it reaches a sufficiently high agreement rate with those labels. The team also moved from generic scalar quality scores to narrow, behavior-specific pass-or-fail rubrics, and fine-tuned its simulated test users on real customer verbatims because early LLM-generated customers were too articulate and cooperative, producing offline pass rates above 90% that didn't match production. daily.dev helps teams building LLM evaluation pipelines keep up with real-world calibration techniques."}}]} ```