Harnesses
Also written: Runtime harness, Multi-model harnesses
A framework or structure that helps AI agents interact with tools and perform tasks.
A harness is like a set of tools and instructions that helps an AI agent do specific jobs. Think of it as a specialized toolkit and a user manual for an AI.
For example, an AI agent might need to book a flight. A harness would provide the code and rules for the AI to access flight booking websites, understand the available options, and complete the reservation. Without a harness, the AI wouldn't know how to interact with those external systems.
In AI development, a harness refers to the software framework or set of tools that enables an AI agent to interact with its environment, utilize external software or APIs, and execute complex tasks. It decouples the core AI model (the 'brain') from the specific actions it can take (the 'hands').
These harnesses manage aspects like tool selection, input formatting for external services, parsing of tool outputs, and maintaining state across multiple interactions. They are crucial for building agents capable of long-horizon tasks, multi-step reasoning, and robust tool use, adapting to underlying model improvements without requiring a complete redesign.
Frontier AI labs are developing more complex AI agents that can perform long-running tasks. These agents require specialized software frameworks, or harnesses, to manage their memory, interactions, and tool use effectively. This focus on robust agent management is crucial for building more capable and reliable AI systems.
- Antigravity AgentGoogle · Gemini Models ·
- Self-Compact Pi Agent: ZERO HYPE Agentic Coding DevlogAnthropic · IndyDevDan ·
- How to Navigate the Next Wave of AI CompetitionOpenAI · The AI Daily Brief ·
- Orchard: An open framework for scalable agentic AIMicrosoft · Microsoft Research Blog ·
- We Need An Ecosystem in AI, And Every Company Can Win A Place In ItMicrosoft · No Priors ·
- Scaling Managed Agents: Decoupling the brain from the handsAnthropic · Anthropic Engineering ·
- Sign of the future: GPT-5.5OpenAI · One Useful Thing (Ethan Mollick) ·
- Effective harnesses for long-running agentsAnthropic · Anthropic Engineering ·
- The future of coding with AI - Episode 6OpenAI · The OpenAI Podcast ·
- Anthropic Claude
- Anthropic is exploring effective harnesses for long-running agents, aiming to improve how their AI agents manage memory and perform tasks over extended periods.
- Microsoft Copilot
- Microsoft's Orchard is presented as an open framework for scalable agentic AI, suggesting a harness for developing and deploying AI agents.
- OpenAI ChatGPT
- OpenAI's discussions on AI competition and the future of coding with AI imply the use of underlying frameworks that enable advanced agent capabilities and development.
- Testing Harness
- A testing harness is specifically designed to run automated tes
- Software Framework
- While a harness is a type of software framework, the term 'h
- Agent Core
- The agent core is the reasoning engine of the AI, while the harness
The term 'harness' in software engineering has a long history, referring to a framework that facilitates testing or execution. In the context of AI agents, it has evolved to specifically denote the systems that connect AI models to external tools and environments.
Try using a tool within your favorite AI assistant. For example, ask ChatGPT or Copilot to 'set a timer for 5 minutes' or 'tell me the weather in London'. The underlying system that allows the AI to access and use these functions is a form of harness.
The Playbook: reusable how-tos, prompts, configs and tips pulled out of the news.
- HeuristicVirtualize agent components like operating systems virtualize hardwareDesign agent interfaces to be stable abstractions that outlast specific implementations, similar to how OS abstractions outlast hardware. · 20 min
- PromptRole-play a student struggling with math for AI tutor simulationSimulate a student's math struggles and uncertainty for AI tutor evaluation. · 5 min
- TemplateReusable workflow templates for healthcare teamsTeams can create and share templates for common workflows like documentation or prior authorizations. · 15 min
- TemplateTemplate for building an AI tool tailored to your workA structured approach to defining and building a custom AI tool for specific business needs. · 15 min
- TemplateEvals CSV prompt list for agent skillsA CSV file template for listing prompts used in evaluating agent skills. · 20 min
Questions people ask
- What is the main purpose of an AI harness?
- The main purpose is to allow an AI agent to interact with external tools, software, or data sources to perform tasks beyond its core knowledge.
- How does a harness help an AI agent?
- It provides the necessary code, interfaces, and logic for the AI to understand how to use specific tools, process their outputs, and complete complex actions.
First seen on Subvolts Nov 26, 2025. Updated Oct 5, 2026. Written by an AI model from the items above and reviewed against them weekly; report an error.