Lessons on retrieval for autonomous coding agents with Cline head of ai.
RAG promised to give coding agents long-term memory and reduce hallucinations. But what if it's actually making them worse at their job? In this talk, Nik Pash (Head of AI, Cline) joins us to challenge the conventional wisdom around RAG for autonomous coding agents, based on his viral essay "Why I No Longer Recommend RAG for Autonomous Coding Agents." We discuss: • Why embedding search and vector databases distract coding agents from logical code exploration • The "plan and act" paradigm and "narrative integrity" - letting agents explore code like senior engineers do • The "bitter lesson" - how the application layer keeps shrinking as models get better • When RAG still makes sense: cost optimization and massive unstructured data Nik shares insights from enterprise adoption at Cline, context management for long-running tasks, and why companies like Cursor use RAG primarily for cost reasons rather than performance. The discussion sparked significant debate in the AI community about whether we're over-engineering solutions to problems that modern LLMs have already solved. About Cline: https://cline.bot/ Connect with Nik: X/Twitter: https://x.com/pashmerepat Substack: https://pashpashpash.substack.com LinkedIn: / nikpash TIME STAMPS 00:00 Introduction and Viral Essay 01:30 The Death of RAG: Initial Reactions 03:24 The Original Promise of RAG 05:48 Building and Scaling Vault 08:00 Challenges and Criticisms of RAG 14:10 Pine Cone's Response and Agentic RAG 20:26 The Bitter Lesson and Application Layer Shrinking 24:43 When RAG Still Makes Sense 31:14 Enterprise Adoption of Klein 33:34 Knowledge Graphs and Coding Agents 37:20 Maintaining Context in Long-Running Tasks 41:56 Multi-Agent Systems vs. Solo Geniuses 44:26 CLI Tools and MCP Servers 52:10 Memory Management in Coding Agents 55:42 Conclusion and Contact Information If you want to learn more about improving rag applications check out: https://improvingrag.com/ Stay updated: X/Twitter: https://x.com/jxnlco LinkedIn: / jxnlco Site: https://jxnl.co/ Newsletter: https://subscribe.jxnl.co/

Is RAG Still Needed? Choosing the Best Approach for LLMs

How AI agents & Claude skills work (Clearly Explained)

Andrej Karpathy: From Vibe Coding to Agentic Engineering w/ Stephanie Zhan

RAG in the age of agents.

I Manage a Team of AI Agents. Here's My Framework.

Tokens can make you rich, just do this – Mario Zechner

RAG Crash Course for Beginners

Model Context Protocol (MCP) Explained for Beginners: AI Flight Booking Demo!

The most rational take on AI you’ll hear this year

OWASP's Top 10 Ways to Attack LLMs: AI Vulnerabilities Exposed

Python, Go, Rust, TypeScript and AI with Armin Ronacher

Skill Issue: Andrej Karpathy on Code Agents, AutoResearch, and the Loopy Era of AI

Yann LeCun on What Comes After LLMs

Don't learn AI Agents without Learning these Fundamentals

Harnesses in AI: A Deep Dive — Tejas Kumar, IBM

Why This Dev Ships 100x Faster Than 99% of Engineers

Agentic Engineering: Working With AI, Not Just Using It — Brendan O'Leary

Something is jamming GPS over Europe. Here's what we found

Exposing The Solid State Donut Battery. It's Over.

