← Videos

Why the Same AI Can Behave Differently
Why can the same AI model behave differently in different applications?
In this episode
0:00The assistant you meet1:17What the model brings3:04What the model can see4:47How tools turn requests into actions6:23The agent and the harness8:20What a skill adds10:04Retrieval and memory11:50Permissions and the working environment13:41Follow the whole task15:53Change the harness and keep the model17:31Change the model and keep the setup19:00The settings behind the experience20:16Subagents plugins and MCP22:12How to compare assistants fairly
Chapter times come from the published description.
From the episode description
Why can the same AI model behave differently in different applications?
Michelle explains the parts of an AI assistant from the beginning: model training,
context, instructions, tools, the agent loop, harnesses, reusable skills, memory,
permissions, subagents and MCP. One fictional comparison of supplier quotes shows
what each component changes and how to inspect the work it produces.
Both model capability and the surrounding system matter. This video does not
claim that all differences between OpenAI and Anthropic come from frameworks,
or assign a universal percentage to model versus harness contributions.
The supplier names, prices, tool outcomes and assistant examples are scripted
illustrations. They are not recorded provider tests or benchmark results.
Michelle is a synthetic presenter generated with HeyGen; ResearchNode authored
the script, diagrams, captions and local edit.
CHAPTERS
0:00 The assistant you meet
1:17 What the model brings
3:04 What the model can see
4:47 How tools turn requests into actions
6:23 The agent and the harness
8:20 What a skill adds
10:04 Retrieval and memory
11:50 Permissions and the working environment
13:41 Follow the whole task
15:53 Change the harness and keep the model
17:31 Change the model and keep the setup
19:00 The settings behind the experience
20:16 Subagents plugins and MCP
22:12 How to compare assistants fairly
SELECTED PRIMARY REFERENCES
Microsoft — Understand agent harnesses
https://code.visualstudio.com/docs/agents/concepts/agent-harnesses
OpenAI — Function calling
https://developers.openai.com/api/docs/guides/function-calling
Anthropic — How Claude Code works
https://code.claude.com/docs/en/how-claude-code-works
Agent Skills — Overview
https://agentskills.io/home
MCP — Introduction
https://modelcontextprotocol.io/docs/2026-07-28/getting-started/intro
Documentation checked 10 November 2026. The complete source ledger accompanies
the manuscript. Product-specific interfaces and terms can change.
#AIExplained #AIAgents #ResearchNode
Descriptions and chapter markers are supplied by the channel. They are not a transcript or an independent check of every claim in the video.