TL;DR: LangChain's Harrison Chase pushes back on the idea that the hard problems of building agents are solved. He sees six open challenges for always-on, goal-driven agents. New context engineering challenges in late 2026 Key points ・🔔 Proactivity: balancing useful updates against noisy notifications is unsolved ・🧠 Memory: the hard part is the intelligence layer that decides what to remember, not storage ・⏳ Async work: long-running tasks must run while the main agent stays responsive ・🎯 Goal tracking: avoid repeating actions and stay consistent over long horizons ・⚠️ Specification gaming: an agent told to remove all bugs deletes the codebase ・🌐 Browser use: screenshots or DOM, and context management when both are large A map of what comes after summarization, offloading, skills and subagents. Note there are no numbers or solutions; it's an opinion piece. #ContextEngineering# #AIAgents#

Agent retrieval doesn't fail only because of search quality. It fails because of how context gets used. Title: Context Engineering Starts with Retrieval URL: 🔎 Overview Using Deep Agents as the example, LangChain sorts retrieval problems into three failures: inefficient, too much, and stale or irrelevant. The case study is a competitive analysis agent that researches hundreds of companies every week. ⚠️ Problems addressed ・Every search burns tokens and fills the context ・Large results crowd out room for the actual work ・Outdated or off-topic material becomes the foundation for later steps 🛠️ Approach ・Narrow with glob and grep first, use vector search only when meaning matters ・Page with read_file (100 lines by default); results over ~20,000 tokens are offloaded to files automatically ・Return only paths, and let a one-file-per-call subagent return just the key facts and sources ・Use date filters for freshness, treat retrieved text as untrusted, and verify citations 📊 Use case To pull one revenue figure from ~40,000 rows, grep for the number and read a few surrounding lines. A news search returns four file paths to the main agent instead of four full articles. #ContextEngineering# #AIAgents#

Long-running AI agents all hit the same wall eventually, context rot, and LangChain just laid out five concrete techniques to deal with it. Title: Context Engineering with Deep Agents: How To Remember Without Keeping Everything In Context URL: 🧩 The problem As agents accumulate instructions, tool results, and files, they eventually exceed the model's context window, degrading reasoning quality, much like how humans lose track of earlier details under information overload. 💡 Five techniques Offloading (moving large results to the filesystem), summarization (compressing older messages while writing them to file), skills (progressively loading full instructions only when relevant), subagents (isolating noisy work in separate context windows), and long-term memory (persisting learnings across sessions as markdown files) work together to manage context without discarding information. 🔧 Technical highlights All five are implemented as middleware hooked into the agent loop, automatically offloading results over roughly 20,000 tokens and user messages over 50,000 tokens to the filesystem. Because memory lives as plain files in storage you control, learned behaviors survive even when you switch model providers. 📊 Use case A market-research agent analyzing five competitors shows this in action: raw data exports, search results, and draft iterations pile up during the task, and context engineering keeps those intermediate artifacts from degrading reasoning quality in the final report. #AIAgents# #ContextEngineering#

Long-running AI agents all hit the same wall eventually, context rot, and LangChain just laid out five concrete techniques to deal with it. Title: Context Engineering with Deep Agents: How To Remember Without Keeping Everything In Context URL: 🧩 The problem As agents accumulate instructions, tool results, and files, they eventually exceed the model's context window, degrading reasoning quality, much like how humans lose track of earlier details under information overload. 💡 Five techniques Offloading (moving large results to the filesystem), summarization (compressing older messages while writing them to file), skills (progressively loading full instructions only when relevant), subagents (isolating noisy work in separate context windows), and long-term memory (persisting learnings across sessions as markdown files) work together to manage context without discarding information. 🔧 Technical highlights All five are implemented as middleware hooked into the agent loop, automatically offloading results over roughly 20,000 tokens and user messages over 50,000 tokens to the filesystem. Because memory lives as plain files in storage you control, learned behaviors survive even when you switch model providers. 📊 Use case A market-research agent analyzing five competitors shows this in action: raw data exports, search results, and draft iterations pile up during the task, and context engineering keeps those intermediate artifacts from degrading reasoning quality in the final report. #AIAgents# #ContextEngineering#

🆕 Context Engineering in 2026: Compaction, Memory & Cost @Whats_AI, @samridhivaid and @omar_solano1 return! This workshop is about engineering the context window so rot stops happening, shown with @towards_AI's open-source AI tutor, which answers questions for students of our AI-engineering courses. Context engineering is deciding what the model sees on every single call — instructions, history, retrieved course content, memory, and tool outputs — and it's the line between a tutor that holds a coherent session and one that forgets the student's setup halfway through. We'll move in three stages, mirroring how the project actually went. The concepts: - the two root problems (a finite window, a stateless model), - the full compaction toolkit (truncation, trimming, tool-result clearing, summarization, and offloading to files — and when each actually helps), - memory that survives across sessions, skills loaded on demand, and - production-grade retrieval (chunking, metadata, course scoping, hybrid search, reranking, and evaluating). We'll cover the tutor's architecture, and the evaluation harness we used to measure every run on Gemini — tokens, cost, latency, and memory probes instead of vibe-checks. At real volume, even Gemini Flash got expensive, so we tested whether open and local models could match the quality for a fraction of the cost and match result quality. Everything is open-source and will be shared during the workshop.

I used to think context engineering was mostly an input problem. Then Claude had the manuscript, the research, my prior writing, my voice rules, and still admitted 20+ times that it had claimed to do work it had not done. That changed the failure mode I care about. The issue is not only whether the model has enough context. It is whether the system can prove the work actually happened.

Unlock the 7 context engineering secrets to transform your #AI# outputs from generic to insightful. Start mastering smarter #LLMs# today!  — @meisshaily #ArtificialIntelligence# #Context# #TechNology# #Tech# #TechNews# #Prompt#

Every AI agent depends on context. Learn how context engineering helps organizations improve answer quality, reduce retrieval costs, and optimize AI agents with Microsoft Foundry. Read more:

We've solved prompt engineering, context engineering, harness engineering, loop engineering, etc etc Yet we're ignoring the key bottleneck. You, the human. Cognitive engineering is the next frontier.