# The Agentic Blog: Build, Ship \& Scale AI Agents

Generated by Yoast SEO v27.3, this is an llms.txt file, meant for consumption by LLMs.

## Pages
- [Sample Page](https://blog.appxlab.io/sample-page/)

## Posts
- [Cursor 3 Background Agents: The Config Playbook](https://blog.appxlab.io/2026/04/25/cursor-3-background-agents-setup/): Most teams discover Cursor 3 background agents by accident — and then immediately hit merge conflicts, credit overruns, and setup that lives in one engineer's head\. This guide covers the exact three\-file configuration, worktree isolation setup, and PR review checklist that turns parallel agents into a repeatable production workflow\.
- [Google ADK Tutorial: Skills, Parallel Agents \& Vertex AI](https://blog.appxlab.io/2026/04/21/google-adk-tutorial-production/): Most Google ADK tutorials stop at a single\-agent quickstart and never cover the three patterns that matter in production: SkillToolset for context\-efficient skill loading, ParallelAgent for concurrent sub\-agent execution, and zero\-code deployment to Vertex AI Agent Engine\. This walkthrough provides working Python code for all three, plus an honest comparison with LangGraph and CrewAI\.
- [OpenTelemetry LLM Observability for AI Agents](https://blog.appxlab.io/2026/04/21/opentelemetry-llm-observability/): Proprietary LLM observability SDKs lock you into a vendor's data model and leave your AI agent traces isolated from the rest of your APM\. OpenTelemetry's GenAI Semantic Conventions define a portable instrumentation standard — covering spans, metrics, and events — that ships to Datadog, Grafana, Uptrace, or any OTLP backend without changing a line of application code\. This guide covers every critical detail: which attributes to capture, why prompts belong in events not attributes, how to track cached tokens accurately, and how to build parent\-child span hierarchies for multi\-agent workflows\.
- [Fine\-Tune an LLM for Code: LoRA, Unsloth \& 2026](https://blog.appxlab.io/2026/04/21/fine-tune-llm-for-code/): Most fine\-tuning tutorials set you up to fail by training on synthetic data, skipping real evaluation, and ignoring silent failure modes like chat template mismatch\. This guide walks through a complete weekend project: extracting a training dataset from your own git history, running LoRA/QLoRA with Unsloth on a single consumer GPU, and evaluating results with an LLM\-as\-judge pipeline that actually tells you whether the fine\-tuned model is better\.
- [Reasoning Models for AI Coding: When to Use Them](https://blog.appxlab.io/2026/04/21/reasoning-models-ai-coding/): Reasoning models like o3 and Claude Opus can cost 2–5x more and take 6\+ seconds to respond — but for the right coding tasks, they're worth every token\. This framework maps your workflow tasks against real data so you stop overpaying for the wrong things\.

## Categories
- [Artificial Intelligence](https://blog.appxlab.io/category/ai/)
- [Tutorials](https://blog.appxlab.io/category/tutorials/)

## Optional
- [Sitemap index](https://blog.appxlab.io/sitemap_index.xml)