Build, Observe, Fix: A LangChain Agent Walkthrough

2026-07-19 · Source: LLM on Medium · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Novice, long

Summary

This article details the process of building and debugging a LangChain agent designed to create a "morning briefing" by fetching top India news across four categories, reading articles, and generating a structured daily digest. The author recounts hitting three common bugs: an LLM ignoring quantity constraints ("2 to 3" becoming "1"), silent step omission where the agent skipped reading articles, and Google News URLs returning blank pages. The solution involved prompt engineering, implementing agent.stream for observability, refactoring the agent into a two-phase pipeline (selector agent + Python-driven article fetching + writer LLM call), and using googlenewsdecoder to resolve URLs. The setup requires an Azure OpenAI resource with gpt-4o-mini and LangChain 1.x, taking about 90 minutes.

Key takeaway

For AI Engineers building LangChain agents, prioritize observability and deterministic design to avoid subtle failures. Implement agent.stream from day one to trace agent behavior, and move mandatory, non-judgmental steps out of the LLM's control into explicit Python code. This prevents silent omissions and ensures reliability, saving significant debugging time. Always test individual tools thoroughly before integrating them into the agent workflow.

Key insights

Debugging LangChain agents requires tracing, precise prompt engineering, and moving deterministic steps to code to prevent silent omissions.

Principles

• Tools must always return strings.
• Return errors, don't raise them.
• Agents decide, code fetches.

Method

Refactor agents into a two-phase pipeline: a selector agent uses tools to make judgment calls, followed by a Python-driven loop for deterministic tasks like article fetching, then a direct LLM call for final writing.

In practice

• Use agent.stream(stream_mode="updates") for debugging.
• Replace prompt ranges like "2-3" with "EXACTLY N."
• Test tools in isolation before agent integration.

Topics

• LangChain Agents
• LLM Debugging
• Prompt Engineering
• Azure OpenAI
• Agent Observability
• Python Development

Code references

• souravb65/morning-briefing-agent

Best for: AI Engineer, Machine Learning Engineer, AI Student

Related on AIssential

• The Loop Was the Easy Part: Evals, Observability, and Rollbacks for Your DIY Claude Code — Operationalizing AI agents requires robust observability, evaluation, and rollback mechanisms beyond the core agent loop.
• Issue #131 - Build an Agent in LangChain — LangChain's `create_agent` simplifies ReAct agent deployment by integrating reasoning, acting, and tool management.
• LangChain's CEO argues that better models alone won't get your AI agent to production — Harness engineering extends context engineering to enable autonomous, long-running AI agents by giving LLMs more control.
• How Tool Chaining Fails in Production LLM Agents and How to Fix It — Tool chaining in LLM agents often fails silently in production due to cascading errors and context loss.
• This Week in AI: Fable 5, the Clone Wave, and Uber’s AI Reality Check — Efficient AI agent development hinges on leveraging existing "ingredients" and building agent-native, observable tooling to manage costs.
• LangChain vs LangGraph vs LangSmith vs LangFlow: Choosing the Right LLM Toolkit — The LangChain ecosystem provides specialized tools for building, orchestrating, prototyping, and observing LLM applications.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by LLM on Medium.