Fundamental knowledge SWE's in 2026 must have before I will even consider hiring them

2026-01-16 · Source: Geoffrey Huntley · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Intermediate, long

Summary

Jeff Huntley, a former tech lead at Canva and engineer at Sourcegraph, defines an "agent" as approximately 300 lines of code running in a loop with LLM tokens. He asserts that understanding and building agents is a new fundamental skill for software engineers by 2026, akin to knowing what a primary key is. The presentation outlines core concepts for agent development, including distinguishing between agentic and non-agentic LLMs (e.g., Grok for low safety, Anthropic/OpenAI for high safety), and critical context window management to avoid "pollution" by limiting it to one activity and carefully selecting MCPs (Multi-Capability Prompts). Huntley demonstrates building a baseline coding agent by incrementally adding tools like chat, list files, read file, bash, and an edit tool, showcasing how these primitives stack to create functional automation. He stresses that this capability transforms engineers from AI consumers to producers, enabling significant automation and personal development.

Key takeaway

For software engineers aiming to remain competitive by 2026, you must prioritize learning to build LLM agents. This fundamental skill, comparable to understanding primary keys, transforms you from an AI consumer to a producer, enabling significant automation within your organization. Invest in developing this expertise now to avoid being outpaced, focusing on practical agent construction and nuanced context management for optimal results.

Key insights

Building an LLM agent involves a simple loop integrating specialized tools and careful context management.

Principles

• Not all LLMs are agentic; select models based on task behavior (e.g., safety, tool-calling).
• Dedicate the LLM's context window to a single activity to prevent "pollution".
• Avoid over-allocating Multi-Capability Prompts (MCPs) to maintain LLM effectiveness.

Method

An agent operates via a loop that takes user input, performs LLM inference, and conditionally invokes registered tools (functions with "billboards" instructing the LLM). Each message or "turn" reallocates previous conversation history.

In practice

• Implement a baseline agent with chat, list files, read file, bash, and edit tools.
• Combine different LLMs (e.g., Anthropic for core, O3 for checking) for robust agentic workflows.
• Use "new chat" frequently in tools to reset context and avoid pollution.

Topics

• LLM Agents
• Agentic Development
• Context Window Management
• Tool Calling
• Software Engineering Skills
• AI Automation

Best for: Software Engineer, AI Engineer, AI Student

Related on AIssential

• Our New Agentic AI Engineering Course! — Building production-ready AI agents requires extensive engineering beyond basic LLM calls.
• Anatomy of Agent SKILLS — Agent "skills" prevent context bloat by progressively disclosing information, improving efficiency and modularity.
• 10 LLM Engineering Concepts Explained in 10 Minutes — Modern LLM applications are complex systems built on engineering principles beyond basic prompt design.
• 🤖AI Agents Weekly: LLMs in 2025, YOLO in the Sandbox, Plan Caching for Agents, DeepTutor — Coding agents and reasoning models significantly advanced LLM capabilities and market impact in 2025.
• humanlayer / 12-factor-agents — Reliable LLM agents require robust software engineering principles, not just advanced AI frameworks.
• 4 Things Everyone Must Know About LLMs and Agentic AI — Effective LLM and agentic AI development prioritizes model selection, task simplification, context management, and structural prompting over complex prompts.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Geoffrey Huntley.