Harness Engineering logopeterindia.net
Knowledge Hub

Harness Engineering

A curated guide to harness engineering — the memory, context, tools, skills, guardrails and feedback loops around an AI agent — with the Thoughtworks Radar theme, primary write-ups, open standards and research.

34 topics•6 categories•Updated November 15, 2026

What Is Harness Engineering?

Harness engineering is the practice of building the environment around an AI agent: memory and context, tools, skills, instructions, guardrails and the feedback loops that let the agent and its team improve. A common shorthand is Agent = Model + Harness. The Thoughtworks Technology Radar Vol 35 makes harness engineering at scale a theme because a plugin ecosystem is forming, with standards such as Agent Plugins, marketplaces and package managers for sharing skills. It also stresses that a harness is a living asset that needs ownership, evaluation and a feedback loop, and suggests the enterprise harness may become a product. This page summarises the theme in my own words and collects the primary write-ups, standards, tools and research.

The Radar Theme

3 entries

The Thoughtworks Technology Radar Vol 35 (November 2026) names harness engineering at scale as a theme, defining the harness as everything built around an AI agent, from memory and context to feedback loops and guardrails.

Core Concepts & Definitions

4 entries

Primary write-ups that define harness engineering and show it in practice.

Radar Entries on Harness Engineering

7 entries

The seven November 2026 Radar entries behind the theme, with the ring Thoughtworks gave each. The summaries are mine; read each entry for Thoughtworks' own assessment.

Open Standards for Harness Assets

6 entries

Shared formats that let skills, instructions and tool servers move between agents.

Distribution & Supply Chain

5 entries

Tools and standards for sharing harness assets and tracking where they came from.

Research on Harness Design

4 entries

Recent papers on harness optimisation and architecture.

Related Pages

5 entries

Related directories on this site.

Frequently Asked Questions

What is harness engineering?

It is the discipline of building everything around an AI agent except the model: context and memory, tools, skills, instructions, guardrails and feedback loops. A common shorthand is Agent = Model + Harness.

How is harness engineering different from context engineering?

Context engineering is mostly about what the model sees. Harness engineering also covers the tools, guardrails, checks and feedback loops that steer the agent and test its output, so context is one part of the harness.

What belongs in a harness?

Typical parts are instruction files such as AGENTS.md, skills, MCP servers, hooks and deterministic guardrails. Böckeler groups them into feedforward guides that steer the agent before it acts and feedback sensors, such as tests and linters, that check its work afterwards.

How do you test a harness?

Thoughtworks suggests a test suite for deterministic guardrails, so each one is shown to work, and evaluations for non-deterministic parts such as skills and plugins, run on a schedule such as nightly or alongside CI.

What is an enterprise harness?

It is the organisation-wide layer that coding agents consume: shared skills, tools, connectors to internal context, approved technologies, architecture decision records and security policies. Thoughtworks advises keeping it thin and federated and sharing assets through open standards and a plugin marketplace.