Build agents, not pipelines

2026-05-30 · Source: All posts - seangoedecke.com RSS feed · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Robotics & Autonomous Systems · Depth: Intermediate, long

Summary

The article distinguishes between two primary methods for integrating Large Language Models (LLMs) into computer programs: pipelines and agents. Pipelines involve explicit code-defined control flow, while agents empower LLMs with tools to manage their own control flow. While pipelines offer greater predictability in cost and latency, agents demonstrate superior flexibility and intelligence, capable of handling more complex tasks by looping and gathering information dynamically. This makes agents, like Claude Code, Codex, Cursor, and Copilot, particularly effective for challenging problems such as coding. Pipelines excel in scenarios requiring strict context size limits, predictable GPU costs, or local model deployment, often staying below 32k or 6k tokens. Agents are generally easier for context-gathering, as they can fetch data as needed, contrasting with pipelines' reliance on pre-assembled context, a challenge Retrieval-Augmented Generation (RAG) often failed to solve. The author suggests agents are more future-proof due to ongoing model improvements and their ability to delegate more to the LLM.

Key takeaway

For AI Engineers or Architects designing LLM-powered applications, if you face complex, open-ended problems where context assembly is challenging, prioritize building agentic systems. Your initial investment in agents will yield greater flexibility and intelligence, especially as models improve. Conversely, if your project demands strict cost predictability, low latency, or local model deployment, opt for pipelines. You should evaluate the trade-offs carefully, but when in doubt, start with agents for easier development and future adaptability.

Key insights

Agents offer superior flexibility and intelligence for complex LLM tasks, while pipelines provide predictability and cost control.

Principles

• Agents are smarter and more flexible.
• Pipelines offer predictability and cost control.
• Context-gathering is simpler with agents.

In practice

• Use pipelines for strict context/cost needs.
• Employ agents for complex, unpredictable tasks.
• Prioritize agents for future-proof LLM systems.

Topics

• LLM Agents
• LLM Pipelines
• Context Management
• Retrieval-Augmented Generation
• AI System Design
• Model Predictability

Code references

• sgoedecke/gatsby-blog

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

Related on AIssential

• When to use Small LM for AI Agents: New Insights — Small, open-weight LLMs can handle many AI agent tasks, significantly reducing operational costs.
• How to build AI product sense — Context engineering optimizes LLM agent performance by managing context growth through offloading, reduction, isolation, and layered action spaces.
• Why I Stopped Using One Agent and Built a Multi-Agent Pipeline Instead — Single LLM agents struggle with complex, multi-faceted tasks by attempting too many competing mental models in one context.
• Top 10 GitHub Repositories Trending in July 2026 (AI, ML & GenAI Edition) — The AI ecosystem's innovation focus has shifted from LLMs to agent frameworks and infrastructure for practical applications.
• Top 100 Claude Certified Architect — Professional Questions and Answers — Choosing the optimal LLM architecture depends on task predictability and the need for dynamic adaptation.
• You Probably Don’t Need an Agent Framework — Explicit LLM workflows offer more reliability and control than autonomous agents for many 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 All posts - seangoedecke.com RSS feed.