Title of the talk/workshop
Your Agents.md isn't enough: Context Engineering for Coding Agents
Abstract of the talk/workshop
We all rushed to add an AGENTS.md (or CLAUDE.md, or .cursorrules) to our repos the moment coding agents got good. But dump everything into one giant file, and you'll hit the same wall researchers recently found: bloated context files can increase inference cost by over 20% and actually reduce task success — the agent drowns in irrelevant instructions instead of finding the one it needs.
This talk is a practitioner's field guide to doing context engineering right, built from hands-on work across Codex, Copilot, and Claude Code. We'll cover why giant, repeated prompts don't create memory — just cost; how splitting AGENTS.md files at the directory level lets agents find the right instruction instead of all instructions; how turning repetitive workflows into skills upgrades your prompts on demand instead of bloating them upfront; why context rots — both mid-session and across a codebase's lifetime — and what to do about it; and where plugins, connectors, and custom RAG take context engineering next.
You'll leave with a concrete framework for structuring context — not just another list of file formats to copy.
Category of the talk/workshop
AI, Developer tools
Duration (including Q&A)
30 mins
Level of Audience
Beginner - Intermediate
Speaker Bio
- Speaker Bio (Brief) - I'm a Senior Software Engineer at F5 Networks, with 9+ years of experience designing and deploying backend systems and microservice applications using Docker and Kubernetes. Over the past several months, I've brought that backend foundation into agentic AI — using coding agents like Codex and Copilot hands-on in my day-to-day engineering work, and exploring context engineering practices to make them genuinely useful on real, large codebases.
- LinkedIn Profile: https://www.linkedin.com/in/vasavi-yarramneni/
- Company/College - F5 Networks
- Email - vasavigeethay23@gmail.com
- Years of Exp - 9+
Prerequisites(if any)
Title of the talk/workshop
Your
Agents.mdisn't enough: Context Engineering for Coding AgentsAbstract of the talk/workshop
We all rushed to add an AGENTS.md (or CLAUDE.md, or .cursorrules) to our repos the moment coding agents got good. But dump everything into one giant file, and you'll hit the same wall researchers recently found: bloated context files can increase inference cost by over 20% and actually reduce task success — the agent drowns in irrelevant instructions instead of finding the one it needs.
This talk is a practitioner's field guide to doing context engineering right, built from hands-on work across Codex, Copilot, and Claude Code. We'll cover why giant, repeated prompts don't create memory — just cost; how splitting AGENTS.md files at the directory level lets agents find the right instruction instead of all instructions; how turning repetitive workflows into skills upgrades your prompts on demand instead of bloating them upfront; why context rots — both mid-session and across a codebase's lifetime — and what to do about it; and where plugins, connectors, and custom RAG take context engineering next.
You'll leave with a concrete framework for structuring context — not just another list of file formats to copy.
Category of the talk/workshop
AI, Developer tools
Duration (including Q&A)
30 mins
Level of Audience
Beginner - Intermediate
Speaker Bio
Prerequisites(if any)