Thoughts on coding agents

2026-02-22 · Source: Denny's Blog · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Advanced, long

Summary

The author, with over 20 years of software engineering experience, describes coding agents as the most significant workflow shift observed, despite acknowledging current overhyped claims of 10-100x productivity gains or engineer replacement. While not replacing experienced developers, agents offer a consistent ~1.5x productivity increase. The author's workflow has shifted from IDE-centric to agent-first, primarily using CLI agents like Codex CLI and Claude Code, with IDEs like Zed used mainly for navigation and code review. The nature of software development has evolved from direct code writing to detailed specification and context provision for agents. The author emphasizes that software development remains a process of model-building and learning, which agents can inadvertently diminish if not managed deliberately. Productivity gains are highest (10x+) for mechanical refactors or greenfield scaffolding with strong guardrails, while typical brownfield work yields 1-2x gains.

Key takeaway

For NLP Engineers evaluating agent-assisted development, recognize that while agents offer substantial productivity gains (1.5x average, 10x+ for specific tasks), your role shifts to context provision and managing the agent's output. Focus on structuring code and documentation for agent readability and implement robust guardrails (e.g., strong typing, integration tests) to maximize agent effectiveness and prevent costly rework. Be mindful that agents may reduce implicit learning, necessitating deliberate model-building.

Key insights

Coding agents fundamentally shift software engineering workflows, demanding explicit context provision and deliberate model-building from developers.

Principles

• Consistent ~1.5x productivity gains are significant.
• Software development is model-building, not just text generation.
• Guardrails are crucial for agent productivity and correctness.

Method

Adopt an agent-first workflow using CLI agents, leveraging IDEs for code understanding and diffs. Structure code and documentation for optimal agent context, including comprehensive docstrings and tests as API examples.

In practice

• Use CLI agents (e.g., Codex CLI, Claude Code) over resource-heavy IDE wrappers.
• Prioritize "planning mode" for agents on complex tasks.
• Implement strong guardrails like typed languages and integration tests.

Topics

• Coding Agents
• Software Development Workflow
• Large Language Models
• Developer Productivity
• Behavior-Driven Development

Best for: NLP Engineer, Software Engineer, Machine Learning Engineer, AI Engineer

Related on AIssential

• Faster Code, Slower Understanding: A Survival Note from a Year of Agent-First Engineering — AI shifts the engineering bottleneck from code production to comprehension and maintenance, demanding new optimization strategies.
• Open Source Coding Agents: The Future of AI-Driven Software Development — Open source coding agents enable autonomous, scalable software development by orchestrating multi-agent workflows and offering model flexibility.
• Pete Was Right...(Again) — Coding agents are more effective when designed to self-prompt and manage review cycles, rather than relying on direct user prompts.
• Jun 16, 2026Economic ResearchAgentic coding and persistent returns to expertise — Domain expertise, not coding skill, is the primary driver of success and agent utility in agentic coding.
• Judgment and creativity are all you need. — Coding agents move software development bottlenecks from implementation to judgment and creativity.
• 10 Lessons for Agentic Coding — Agentic coding demands new development practices that prioritize learning, testing, and strategic focus, acknowledging code's low cost but high maintenance.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Denny's Blog.