About

Methodology

The Survey

The State of Agentic Coders survey was conducted in 2026 (Q1) using Tally.so. It collected 200 responses from software developers about their AI coding tool usage, workflows, preferences, and sentiment. After filtering for quality (removing troll/spam entries), 199 valid responses were analyzed.

The Agentic Coder Spectrum

The survey's central framework is the Agentic Coder Spectrum — a 1–5 scale measuring how deeply developers integrate AI into their coding workflows. This isn't about skill level; it's about adoption depth.

The Agentic Coder Spectrum

Conversationalist
Copilot
Prompt Engineer
Orchestrator
Systems Designer
Level 1 — Conversationalist: Using AI as a knowledgeable colleague to consult.
Level 2 — Copilot: AI integrated into the editor for autocomplete and boilerplate.
Level 3 — Prompt Engineer: Crafting detailed prompts, reviewing AI-generated code.
Level 4 — Orchestrator: Running multiple agents as specialized team members.
Level 5 — Systems Designer: Designing software factories — systems that write code.

Data Processing

All data is processed at build time from the raw CSV. Processing includes: filtering troll entries using keyword detection, normalizing freeform model names (e.g., "GPT 5.2", "Gpt5.2", "gpt-5.2 xhigh" are all mapped to "GPT-5"), splitting multi-select fields, and computing aggregated distributions.

Sample Characteristics

The sample skews toward experienced web developers — 48% have 10+ years of experience and 61% work in Web/Full-stack. 46% are based in Europe, 35% in North America. This reflects the distribution channels used (primarily Twitter/X developer communities). Results should be interpreted with this context in mind.

Limitations

  • Self-selected sample (not random) — responses come from people already interested in AI coding
  • Sample size of ~200 limits statistical power for fine-grained cross-tabulations
  • The survey was primarily distributed through English-language developer communities
  • Self-reported data may differ from actual tool usage patterns