Context engineering with Dex Horthy

2026-07-08 · Source: The Pragmatic Engineer · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Intermediate, extended

Summary

Dex Horthy, founder of Fumilir and coiner of "context engineering," discusses building effective AI-assisted software. His experience includes shutting down a "lights-off" software factory after four months due to unmanageable code quality degradation. Horthy emphasizes context engineering as crucial for maximizing AI output quality by carefully crafting model inputs, moving beyond basic prompt engineering. He also introduces "harness engineering" for optimizing coding agent environments and "loop engineering," including "slow loops" that generate small, reviewable PRs daily. The discussion covers the "physics of context windows," noting quality degradation beyond the "smart zone" (first 100K-200K tokens), and the evolution of software factories from a 1968 NATO conference to modern agentic systems.

Key takeaway

For AI Engineers building production-grade software, prioritize "token smarter" approaches over "token harder" to avoid unmanageable codebases. Your team should adopt incremental "slow loops" for continuous code quality improvements, generating small, human-reviewable pull requests daily. Additionally, integrate human-agent planning and design checkpoints early in the SDLC to guide agents, ensuring maintainable architecture and program design, thereby achieving 2-3x faster development without sacrificing quality.

Key insights

Context engineering, the precise crafting of LLM inputs, is key to achieving high-quality, reliable AI-generated software.

Principles

• Own your context window to impact output quality.
• Automate feedback loops for models to check their own work.
• Prioritize human effort for efficiency over raw model intelligence.

Method

Implement "intentional compaction" by compressing context into verified artifacts (e.g., research docs, design plans) and starting fresh sessions to stay within the "smart zone" of the context window.

In practice

• Use "slow loops" for incremental code quality improvements via daily PRs.
• Integrate human-agent planning checkpoints before code generation.
• Focus on program design and architecture for long-term maintainability.

Topics

• Context Engineering
• AI Agents
• Software Factories
• Loop Engineering
• Harness Engineering
• LLM Context Windows
• Code Quality

Best for: CTO, VP of Engineering/Data, Director of AI/ML, AI Engineer, Machine Learning Engineer, MLOps Engineer

Related on AIssential

• Context Is the New Code — Patrick Debois, Tessl — Effective AI agent performance hinges on meticulously engineered context, necessitating a dedicated Context Development Life Cycle.
• DAY 1 Livestream - 5-Day AI Agents: Intensive Vibe Coding Course With Google — Software development is fundamentally shifting from syntax-based coding to intent-driven expression via AI agents.
• Boost Productivity with Vibe Coding and IBM Bob — AI coding tools boost productivity for greenfield, low-complexity tasks, but require careful integration and human oversight.
• From prompt engineering, to context engineering, and now… loop engineering? — AI-driven software development evolved from simple prompts to complex agentic "loop engineering" for sustained task execution.
• AIEWF Daily Dispatch: The great loops debate and the state of AI engineering — AI engineering faces a tension between rapid agent adoption and current limitations in control, cost, and human oversight.
• 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.

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 The Pragmatic Engineer.