# AI Model vs Agentic Harness: What Actually Drives AI

## Executive summary

The video distinguishes between the core AI model (the 'brain') and the surrounding infrastructure, or 'agentic harness' (the 'jar'), which is responsible for advanced system capabilities. While models like ChatGPT provide intelligence, it is the harness—comprising tools, memory management, and agentic loops—that allows an AI to perform complex, multi-step software tasks by interacting with external systems.

## Key takeaways

- AI Model vs. Agentic Harness: The core AI model (e.g., ChatGPT, Claude) is the neural network itself but cannot independently interact with the outside world; it requires an agentic harness to function as a powerful system.
- Components of the Agentic Harness: An overall AI agent consists of the AI model plus the agentic harness, which manages tools, memory, and execution loops. These components are critical for real-world performance.
- Capability Gains Source: Most recent capability gains in generative AI come from improvements in the harness (better tools, memory handling, smarter loops) rather than solely from improvements to the underlying model.

## Technical details

- Tools and External Access: The harness provides tools allowing the model to read/write files, run code (potentially in a sandbox), browse the web, or perform computer use (cursor movement). For external services, it supports the Model Context Protocol (MCP) for plug-and-play integration.
- Memory Management: The harness overcomes the model's fixed context window limitation by persisting instructions (e.g., `agents.md`) and implementing context compaction, summarizing past events while pruning redundant tool outputs. It also enables targeted search using grep-like text search or semantic/code indexes.
- Agentic Loop and Verification: The agentic loop is the cycle where the model plans a step, the harness executes an action, the model observes results, and the process repeats. Modern harnesses include continuous verification, running tests or even spinning up separate models as reviewers.

## Practical implications

- Understanding the separation between model intelligence and harness capability is crucial for evaluating AI performance in complex tasks like code generation or customer support.
- System architects must consider both the choice of AI model and the robustness/features of the agentic harness when designing automated workflows.

## Topics

generativeai, aimodel, aitools, agentic_harness, IBM Technology, https://ibm.biz/~fAH1kt7jT

Source: https://www.youtube.com/watch?v=ZELPNFXJ4_o