harness
The surrounding software system in which an AI agent runs, managing how it executes, accesses tools, and operates.
As defined in Lex Fridman Podcast #491 - OpenClaw
Other definitions
The scaffolding of prompts, tools, and control logic built around a model to make it perform tasks reliably.
Sarah Guo - Funding the Frontier - [Invest Like the Best, EP.489]
Where it comes up · 12
- Rethinking AI Agents: The Rise of Harness Engineering35 claims
- Building Investing Superintelligence4 claims
- Christina Cacioppo - Building an Enduring Software Company in the Age of AI - [Invest Like the Best, EP.496]4 claims
- Lex Fridman Podcast #501 - DHH3 claims
- Alex Sacerdote - How to Invest Through Technology Cycles - [Invest Like the Best, EP.477]2 claims
- Gavin Baker - Watts and Wafers - [Invest Like the Best, EP.473]2 claims
- How Bridgewater Built an AI Analyst That Does Hours of Expert Research in Minutes2 claims
- Conversations with Tyler - Luis Garicano1 claim
and 4 more
Most important claims · 10 of 57Search all claims for “harness” →
- Anthropic identified five canonical patterns in agent harness design: prompt chaining, routing, parallelization, orchestrator workers, and evaluator optimizer loops, each representing a different strategy for when and how the model gets called.Rethinking AI Agents: The Rise of Harness Engineering
- Architectural choices in harness design—not the model underneath—drive the six-fold performance gaps observed across systems with identical models and benchmarks.Rethinking AI Agents: The Rise of Harness Engineering
- Stanford researchers found that orchestration code wrapping a language model drives more performance variation than the model itself, with the same model and benchmark showing six times performance difference depending on harness implementation.Rethinking AI Agents: The Rise of Harness Engineering
- By making the agent fully aware of its own source code, harness, model, and documentation, Steinberger created software that can modify itself via the agentic loop, effectively realizing self-modifying software that he didn't even explicitly plan.Lex Fridman Podcast #491 - OpenClaw
- Roughly 90% of all compute in agent harnesses flows through delegated child agents, not the parent, establishing that the harness is fundamentally an orchestration pattern (decomposes, delegates, verifies) rather than a reasoning pattern.Rethinking AI Agents: The Rise of Harness Engineering
- When OS Symphony, a native code harness for desktop automation, was migrated from native code into NLAH representation (same strategy, different representation), performance jumped from 30.4% to 47.2%, runtime dropped from 361 minutes to 141 minutes, and LLM calls collapsed from 1,200 to just 34.Rethinking AI Agents: The Rise of Harness Engineering
- LangChain confirmed the orchestration finding by modifying only harness infrastructure on their coding agent, jumping from outside the top 30 to rank five on Terminal Bench 2, without changing the underlying model.Rethinking AI Agents: The Rise of Harness Engineering
- Human agency comes not just from raw recall and IQ but from many interacting microservices in the brain for storing, retrieving, and triggering knowledge, with emotions acting as triggers, and AI systems need analogous components built out.Building Investing Superintelligence
- Frontier AI models are needed for research, innovation, and some military applications, but good law, medicine, accounting, and professional services can run on models three or four months behind the frontier, provided the harness and data are in-house.Conversations with Tyler - Luis Garicano
- Naive harnesses suffer from two failure modes: one-shotting (where the agent tries everything at once and exhausts its context) and premature completion (where a later session sees partial progress and declares victory).Rethinking AI Agents: The Rise of Harness Engineering