Key Figures & Thought Leaders in Loop Engineering

Loop Engineering as a discipline was shaped by a community of engineers, researchers, and writers who recognized that the future of AI development lies not in crafting better prompts, but in designing better systems. Here are the key figures and the concrete tools they built or championed that defined the field.

Addy Osmani

Role: Popularized the term and wrote the foundational definition

Addy Osmani, engineering leader at Google and author of numerous books on web development, published the essay that crystallized the concept of loop engineering in June 2026. His definition remains the most widely cited:

"Loop engineering is replacing yourself as the person who prompts the agent. You design the system that does it instead."

Osmani's contribution was framing the shift in terms that resonated with working developers. He articulated why the move from prompt engineering to loop engineering is not just a semantic change, but a fundamental transformation in how engineers interact with AI systems. His writing emphasizes practical implementation: CLAUDE.md files, hook systems, skill definitions, and the tooling that makes loop engineering possible.

Key concepts attributed to Osmani:

The "replace yourself" framing — you design the system that does the prompting
Practical loop engineering patterns using Claude Code's hooks and skills (documented at code.claude.com/docs)
The control theory analogy — loops as feedback controllers
Emphasis on tooling and infrastructure as the enabler of loop engineering

Practical example of Osmani's principles in action:

Claude Code's hook system embodies the "replace yourself" idea. Instead of manually running claude --print "review this PR" on every pull request, a loop engineer configures a pre-commit hook that automatically invokes Claude Code with a structured prompt. The human is removed from the loop entirely:

{
  "hooks": {
    "pre-commit": {
      "command": "claude --print 'Review staged changes for correctness, suggest fixes, and approve or request changes' --allowedTools Edit,Bash"
    }
  }
}

Boris Cherny (Steipete)

Role: Coined the concept through Claude Code leadership

Boris Cherny (known as @steipete on X/Twitter), a former Apple engineer and key figure behind Claude Code at Anthropic, was among the first to articulate the concept that became loop engineering. His observation that developers should stop manually prompting coding agents and start building systems that prompt agents was foundational.

Cherny's perspective is grounded in practical experience building Claude Code (github.com/anthropics/claude-code) — one of the first production systems designed around loop engineering principles. His insights come from watching thousands of developers interact with AI coding agents and recognizing the patterns that separate effective from ineffective usage.

Key concepts attributed to Cherny:

Real-world evidence of Cherny's principles:

Peter Steinberger

Role: Industry commentator and framework for understanding the shift

Peter Steinberger, founder of PSPDFKit and a well-known figure in the Apple/iOS development community, provided a clear framework for understanding why loop engineering matters and how it differs from what came before. His analysis connects loop engineering to broader trends in AI development and provides context for engineers considering the transition.

Key concepts from Steinberger:

LangChain / LangGraph Team

Role: Technical framework and "Art of Loop Engineering"

The LangChain team, whose core framework lives at github.com/langchain-ai/langchain, published "The Art of Loop Engineering," providing a technical framework for understanding loops as the fundamental pattern of agentic AI. Their work connects loop engineering to the broader agent framework ecosystem and provides practical implementation patterns using LangGraph (github.com/langchain-ai/langgraph), a graph-based agent orchestration framework.

Key contributions:

"At its core, an agent is just a model calling tools in a loop until it's done"
LangGraph as a framework for building loop engineering systems with explicit state management and conditional routing
Connecting loop engineering to the agent framework ecosystem, which now includes CrewAI (github.com/crewAIInc/crewAI) for team-based multi-agent coordination, Microsoft's AutoGen (github.com/microsoft/autogen) for multi-agent conversations, and HuggingFace's smolagents (github.com/huggingface/smolagents) for lightweight agent deployment

Comparative insight from the community:

Anthropic Engineering Team

Role: Built Claude Code — the first mainstream loop engineering platform

The Anthropic engineering team, led by Boris Cherny and others, built Claude Code as the first widely-used platform designed around loop engineering principles. Claude Code's architecture — with hooks, skills, memory, and autonomous agent modes — embodies loop engineering in practice.

Key contributions:

Claude Code's hook system for automated iteration
CLAUDE.md for persistent context and project-level configuration
The /loop command for recurring autonomous tasks
Skills and slash commands as reusable loop components

Production engineering lessons from the ecosystem:

The Broader Tool-Building Community

Loop Engineering is not just theory. It is being built in practice by the teams behind these widely-used tools:

Research and Autonomous Platforms

Tool	Repository	Stars	Loop Engineering Contribution
OpenHands	github.com/All-Hands-AI/OpenHands	—	Autonomous coding agent platform — end-to-end loop engineering for software development
SWE-Agent	github.com/princeton-nlp/SWE-Agent	15K+	Research-oriented agent that resolves GitHub issues autonomously, demonstrating loop engineering at the PR level
MetaGPT	github.com/geekan/MetaGPT	45K+	Multi-agent framework that assigns specialized roles (PM, architect, engineer) — loop engineering through organizational design
Devin	github.com/cognition-labs/Devin	—	Cognition AI's autonomous engineer — the most ambitious closed-loop system, handling end-to-end task execution

Agent Orchestration Frameworks

Framework	Repository	Loop Engineering Contribution
LangGraph	github.com/langchain-ai/langgraph	Graph-based orchestration with state machines, conditional edges, and human-in-the-loop checkpoints
CrewAI	github.com/crewAIInc/crewAI	Team-based multi-agent framework — define roles, tasks, and delegation patterns for complex loops
AutoGen	github.com/microsoft/autogen	Microsoft's multi-agent framework for conversational agent orchestration
smolagents	github.com/huggingface/smolagents	HuggingFace's lightweight agent framework for deploying loops without heavy infrastructure
OpenAI Swarm	github.com/openai/swarm	Lightweight multi-agent orchestration — minimal abstraction for handoff-based loops

The Data Behind the Discipline

Loop engineering is supported by measurable evidence, not just opinion:

The Evolution of the Idea

The concept of loop engineering evolved through a community conversation grounded in real tools and data: