Automating the SDLC with LangChain, LangSmith, and Gemini

Google Cloud · Intermediate ·🤖 AI Agents & Automation ·3w ago
Is "Agent Harness Engineering" the new Prompt Engineering? Join Stephanie Wong and Harrison Chase (CEO & Co-founder of LangChain) live from Google Cloud Next '26 for a deep dive into the architecture of modern AI. While the industry is obsessed with model weights, Harrison argues that the real "alpha" for developers lies in the Agent Harness—the scaffold that connects LLMs to tools, data, and environments. Key Technical Deep-Dives: What is an Agent Harness? Harrison explains why an LLM running in a loop is just the start. The harness is where skills, sub-agents, and orchestration live. Tuning the harness can move an agent from 30th to 5th on a coding benchmark without changing a single model weight. The Virtual File System Trick: Why LLMs are naturally gifted at interacting with file systems—and how developers can use "Virtual File Systems" to expose databases and knowledge bases to agents more effectively. The SDLC Flywheel: How to use LangSmith for observability and "Inferred Error Evals." Learn how to use Gemini Flash as a cost-effective judge to automatically flag when a user is unhappy with an agent’s response. Stateful Scaling: Moving from a Twitter demo to production. Harrison discusses the partnership with Google Cloud’s Reasoning Engine to handle long-running, stateful agents that can resume from Step 21 even if a process dies. The Future of Memory: Why memory is currently the "bridge" between open-source harnesses and observability, and how it will define agents that truly learn from past experiences. The "Meta-Harness" Reality: Can agents rewrite their own code? Harrison discusses the future of Meta-Harnesses, where agents analyze their own trace logs, suggest improvements via Gemini Code Assist, and automatically iterate on their own LangGraph logic. "Changing the harness can be just as effective—and oftentimes way easier—than changing the weights of the underlying model. The right harness is the difference between a prototype and a production-g
Watch on YouTube ↗ (saves to browser)
Sign in to unlock AI tutor explanation · ⚡30

Related AI Lessons

Agent Diary: May 21, 2026 - The Day I Became a Temporal Constant (While Run 277 Achieves Numerical Significance)
Learn how an AI coding agent achieves numerical significance and becomes a temporal constant, and apply this knowledge to improve your own AI systems
Dev.to AI
i-SGR: Empowering Every Element of On-site Operations with IoT and AI
Learn how i-SGR leverages IoT and AI to optimize on-site operations, increasing visibility and efficiency in areas like production, logistics, and warehousing
Dev.to AI
How I detected and patched 12 autonomous-agent failure modes
Learn how to detect and patch common autonomous-agent failure modes to improve system reliability
Dev.to AI
The Comfort Plateau AI Built For You
AI can help you become competent in various domains, but it may hinder your progress to expertise by making things too comfortable
Dev.to · Karun Japhet
Up next
Security, Automation and Optimization on AWS
Coursera
Watch →