case-study

GE Aerospace highlights significant productivity gains from using Devin, an AI software development tool, comparing its impact to providing a top instrument to a master musician.

A case study describing how an AI coding agent dismantled a core architectural invariant across 189 files in a 717k-line codebase using a specification-first convergence methodology, with no test oracle and no human code review. The agent iteratively refined a 55-page specification and performed correction loops, fixing 201 self-identified mistakes across 31 audit loops.

This paper presents a deep reinforcement learning approach for solving vehicle routing problems, demonstrated through three industrial truck planning case studies. The proposed method achieves over 10% cost reduction compared to baseline results and discusses generalization to more VRP variants.

Virgin Atlantic adopts ChatGPT Work to streamline customer journey research and analytics, reducing weeks of competitive analysis to hours and creating secure dashboards for employees.

Austin Berke, Lead AI Product Engineer at Harmonic, shared how they used LangChain's Deep Agents framework to rebuild Scout, quadrupling user retention from week 1 to week 4. The core approach is to use the file system as shared storage to manage context, connecting agents, frontend, and deterministic processes, allowing the model to build context on demand.

LangChain's guide examines how Lyft, Fastweb/Vodafone, and LATAM Airlines deploy AI agents for customer experience, covering self-serve platforms, rep copilots, and production feedback loops.

A case study exploring prompt chaining for automated scholarly report generation, presented as an arXiv paper.

A firsthand account of building a team of AI agents at a company using the Hermes framework, detailing practical outcomes and lessons learned.

Matthew Stublefield of Fieldway built a product research agent using Tavily's web search API integrated with Claude Code via MCP, reducing a 5-hour manual cycle to 25 minutes and cutting weeks of work to 10 hours.

This article reveals, through 27 real-world cases, that people using WorkBuddy to make money primarily employ five features: scheduled automation, multi-agent parallelism, Skill packs, MCP connectors, and memory system. It demonstrates the monetization capability of AI tools in actual business scenarios.

Toyota's enterprise AI team shared the ToyotaGPT platform at the Interrupt conference. Based on LangGraph and LangSmith, it reduced AI agent delivery from 6 months to 4 days, and has over 50 agents running in production, saving millions of dollars.

Matt Chow shares how they used Fable 5 (based on Remotion) to produce a YC launch video in 1.5 days, comparing agency and Fiverr quotes and timelines, demonstrating a method to create high-quality videos quickly and at low cost.

LangChain shares a customer story where PodiumHQ's Walker Ward discusses using LangGraph and LangSmith to move AI agents from prototype to production.

Chintan Turakhia, engineering lead at Coinbase, shares how the team used AI for code review, reducing PR review time from 150 hours to 15 hours, and setting a record of merging 70 PRs in 15 minutes. He emphasizes focusing on solving problems rather than just talking about adopting AI.

This paper documents weight-level political conditioning in large language models, presenting a case study on AI bias regarding the Gaza genocide question.

Xiaomi's smart factory replaced Loki with GreptimeDB for log storage, handling billions of rows per month with tailored indexing: Bloom skipping for high-cardinality trace_id, inverted index for low-cardinality fields, and fulltext search on message bodies.

This paper presents an exploratory case study evaluating GPT-4o's ability to perform refactoring and generate gameplay features in an endless runner game, finding that refactoring tasks succeeded while feature generation tasks mostly failed.

A developer describes building a Zapier and GPT-based automation system for a real estate team that cut lead response time from 14 hours to under 3 minutes, and shares key lessons including avoiding over-personalization, building disqualification filters first, and implementing monitoring.

Martin Fowler and Bayer share the journey of building PRINCE, an agentic AI system using RAG to help researchers query decades of PDF reports, transforming preclinical data retrieval.

This paper presents a classroom case study on teaching AI-based translation and post-editing, focusing on the development of evaluative judgement in students.