AIToday
Large Language ModelsAI Coding AssistantsITmedia AI+Published: Sep 1, 2026, 06:01 JST2 min read

AI Coding Shifts from Prompts to Context to Harness

AI Coding Shifts from Prompts to Context to Harness

Key takeaway

  • AI coding has evolved from prompt to context to harness engineering.

  • The harness is a framework that controls AI agents.

  • This shift is changing developer roles and software creation.

3 Key Points

  1. What happened

    AI coding has moved through three stages—prompt engineering, context engineering, and now harness engineering—where the harness is a structured framework that controls AI agents. This shift was a key topic at the AI Ops Conference held July 30, 2026.

  2. Why it matters

    As AI agents become more autonomous, the human role changes from writing prompts to designing the context and, now, the harness that governs agent behavior. Kyle Daigle, GitHub COO, notes that within two years, coding has moved from assistive to autonomous, a major change for developers.

  3. What to watch

    Harrison Chase, Anysphere co-founder, introduced a five-level framework for AI coding maturity, modeled on NHTSA's autonomous driving levels. Level 5 is the 'Dark Factory,' where software is built without human input, shifting development from processes to a black box.

Ask the AI about this article →

Context & Analysis

The article traces the evolution of AI coding from prompt engineering in 2022 to context engineering in 2024-2025, and now to harness engineering in 2026. This progression reflects a deeper change in what developers control: from writing instructions to curating information to designing the entire system that governs AI agents. The harness, likened to a car's steering system, sets the overall path and rules for the AI, reducing human involvement in granular details.

This shift is driven by the growing autonomy of AI agents. As they operate with more independence, the human role becomes more about supervision and design than direct input. Harrison Chase's five-level framework, paralleling NHTSA's levels for self-driving cars, provides a way to assess where an organization stands in this transition. The framework suggests a future where software development becomes largely automated, a prospect that Chase views as both inevitable and disruptive.

FAQ

What are the five levels of AI coding maturity?
The levels range from Level 0, where AI is a code engine and humans write code, to Level 5, the Dark Factory, where robots build software without human light. Intermediate levels include AI as a copilot, autopilot, safety driver, and PM.
How does the harness differ from prompts?
The harness is a structured framework that defines the pipeline and guardrails for AI agents, whereas prompts are instructions. The shift means humans design the harness rather than just writing prompts or context.

Get the latest Large Language Models news every morning

For example, today's edition would include:

  • Nvidia revives Rubin CPX chip with major redesignYahoo Finance AI · 30m ago
  • AI advice followed by 79%, but well-being unchangedITmedia AI+ · 3h ago
  • Enterprises face agent governance gapSiliconANGLE AI · 6h ago

AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.

Free · takes 30 seconds · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. Q&As are published on this page for other readers too.

Related Articles

Next articleDow Falls on Trump AI Warning; Tesla Rises