Retrieval-Augmented Generation

Evidence (20)

• 3InfoQ·10/7/2026framework_update
  InfoQ Webinar on AI in Production

  InfoQ will host a free 60-minute panel on October 14 about what breaks and what works when deploying AI in production. The discussion will cover agent autonomy versus human approval, verification of AI-generated changes, sensitive-data exposure, and production RAG, with registrants able to submit questions and receive the recording.

• 5InfoQ·10/5/2026framework_update
  Building Reusable Evaluation Frameworks for Agentic AI

  Susan Chang describes how Elastic moved from scattered, ad hoc agent evaluations to a unified framework designed for production use. The approach combines LLM-as-a-judge with deterministic rules, connects Python-based evaluation workflows to TypeScript production systems, and adds deep tracing to detect regressions in complex RAG and cybersecurity workloads.

• 5The New Stack·10/1/2026research
  When Graph RAG Is Needed for Relationship Evidence

  The article argues that vector search is useful for finding relevant documents, but it cannot reliably prove relationships such as ownership, dependency, or policy scope. It recommends adding Graph RAG when an answer depends on explicit connections between records, such as linking a vulnerable library to a service, a customer environment, and a contractual notification rule.

• 8Anthropic News·10/1/2026product_launch
  Barclays expands Claude use across operations

  Barclays is widening its collaboration with Anthropic to deploy Claude across the bank for software development, legacy modernization, and operational efficiency. The rollout includes a goal for Claude Code to reach half of Barclays’ developer population by the end of 2026 and a majority in 2027, alongside existing use cases in employee knowledge assistance and email triage.

• 5InfoQ·9/28/2026product_launch
  Forter turns employees into AI agent creators

  Ben Maraney describes how Forter enabled about 200 employees, including non-technical staff, to build AI agents in two weeks. The approach combined custom MCP servers, no-code and code-based tools, and close coordination with security and legal teams, while avoiding more complex RAG deployments.

• 3InfoQ·9/26/2026research
  Adaptive Recommenders in Production

  Mallika Rao argues that the hardest part of adaptive recommendation systems is not model architecture, but the production systems around it. She highlights real-time feedback loops, retrieval freshness, multi-stage orchestration, and latency budgeting as key elements for operating these systems under real-world constraints such as cost and observability.

• 6The New Stack·9/24/2026research
  Query decomposition can worsen context starvation

  The article argues that query decomposition can improve retrieval recall but still fail at the packing stage when relevant sub-intents are excluded from the final context. It introduces "context starvation" to describe cases where evidence is retrieved but not allocated enough space, causing multi-part queries to lose coverage despite correct search results.

• 3GitHub Blog·9/18/2026framework_update
  GitHub Podcast debates AI code review, Skills, MCP, and RAG

  GitHub’s latest podcast episode examines several common AI hot takes, including whether developers should read AI-generated code, how hiring teams view AI usage, and whether Skills replace MCP. The discussion argues that code review remains necessary, Skills and MCP serve different purposes, and RAG is still relevant for AI systems.

• 6InfoQ·9/16/2026product_launch
  Dropbox Expands Riviera to Support AI Workloads

  Dropbox has evolved Riviera from a file preview service into a universal content processing platform that supports more than 300 file formats and over 100 transformation capabilities. The platform now processes hundreds of thousands of transformations per second and provides APIs for asynchronous content extraction in Search, Replay, Sign, Dash, and AI/RAG workflows.

• 5InfoQ·9/12/2026research
  Building Production-Ready Agentic AI with Knowledge Graphs

  Cassie Shum argues that knowledge graphs provide a strong foundation for production-ready agentic AI systems, especially beyond basic retrieval-augmented generation. She outlines four architectural patterns—context bundling, decision provenance, code as truth, and agent visibility—and shows how a knowledge-graph-based engineering harness can improve feedback loops, token efficiency, and reliability.

• 5The New Stack·9/5/2026research
  Building Trust in Agentic RAG with Evidence Trails

  The article argues that agentic RAG systems need structured evidence trails because agents can rewrite queries, choose tools, and make multiple retrieval decisions before answering. It recommends logging queries, sources, filters, scores, rejected results, and reasons so users can see citations and operators can audit how an answer was produced.

• 5The New Stack·9/4/2026research
  AI Agent Evaluations Become Part of Release Gates

  The article argues that AI agents should be tested with repeatable evaluations before release, not just shown working in a demo. It recommends defining correct behavior up front, using real workflows and edge cases, and checking both outcomes and the steps the agent takes across tools, permissions, and multi-turn interactions.

• 5The New Stack·9/3/2026research
  Guide to optimizing token usage in production LLM systems

  The article explains why token costs in production LLM applications should be treated as a distributed systems problem rather than a simple billing issue. It describes how stateless APIs, repeated history replay, and large prompts create quadratic growth in cost and latency, then outlines prompt-hygiene and retrieval-based fixes to reduce token usage.

• 6The New Stack·8/30/2026research
  AI Agents Push Retrieval Engineering Into a Core Discipline

  The article argues that as AI systems evolve from chatbots into agents that investigate, reason, and act, retrieval quality becomes central to application performance. It says teams must treat retrieval as an end-to-end engineering discipline that combines hybrid search, real-time signals, ranking, machine learning inference, and experimentation to deliver the right evidence at the right time.

• 6The New Stack·8/27/2026research
  Why basic RAG fails at multi-hop reasoning

  The article argues that standard retrieval-augmented generation breaks down on multi-hop questions and global summarization because chunk-based vector search does not capture explicit entity relationships. It presents GraphRAG as a better approach, combining knowledge graphs with vector search to support structured reasoning in enterprise workflows.

• 4Hacker News·8/26/2026framework_update
  Serve Markdown to AI Agents via Accept Headers

  The item proposes using HTTP content negotiation with Accept: text/markdown so AI agents can retrieve a Markdown version of a page instead of parsing full HTML. This can reduce token usage, improve retrieval signal, and lower latency for browsing and RAG workflows.

• 5Hacker News·8/26/2026research
  RAG Can Start With Simple Full-Text Search

  The article argues that many retrieval-augmented generation systems are over-engineered and should begin with simpler approaches such as full-text search before adding embeddings or reranking. It outlines several retrieval recipes, including full-text search, agentic query rewriting, and hybrid search, with the choice driven by freshness, corpus churn, query patterns, scale, and team expertise.

• 7The New Stack·8/25/2026research
  AI agents outpace enterprise sovereignty controls

  A new IDC survey commissioned by Cohere says AI agents are already widely used in large enterprises, but understanding of sovereign AI and the controls needed to govern them remains limited. The report highlights gaps in observability, access controls, and deployment control as agents begin interacting directly with company systems and data.

• 5The New Stack·8/13/2026research
  Why AI pipelines can cost far more after launch

  The article argues that AI application costs often rise sharply in production because token usage grows with prompts, history, retrieval, tool output, and multi-step agent workflows. It says the biggest savings usually come from reducing unnecessary tokens through better architecture, prompt design, caching, routing, and memory management rather than simply switching to a cheaper model.

• 7The New Stack·7/30/2026product_launch
  OpenAI and Elastic deepen enterprise AI integration

  OpenAI and Elastic announced an expanded partnership that combines OpenAI reasoning models with Elasticsearch search, retrieval, and access-control capabilities. The integration targets enterprise AI context, observability, and security, with Elastic citing benchmark gains in retrieval efficiency, token reduction, and answer accuracy.