Uploaded November 2026 | Updated November 2026, 3 days ago
Same model, very different results. The difference is the agent harness: the code around the model that runs the loop, gives it tools, manages its context, keeps its memory and decides what is allowed to run. In this video we open a harness up, follow one real coding task through it step by step, and then compare the harnesses you can use today: Claude Code, Codex, DeepSeek Harness (dsh), OpenClaw, Hermes Agent and the agent SDKs. I have run these harnesses myself, with cloud models and with local models on my own GPU, and I share what changed when only the harness changed. At the end you get my verdict on which harness fits which kind of work.
If this helped, like the video, and on the phone app tap Hype under the video. It helps the channel reach more people.
Subscribe for more AI explainers: youtube.com/@KGPTalkie?sub_confirmation=1
Chapters
0:00 Same model, two different results
0:38 Model vs harness: engine and car
1:12 Proof: same model, better harness
2:00 What I saw in my own tests
2:24 The five parts of a harness
3:24 One task, step by step
4:52 Context: the hard part
5:49 Control: who decides what runs
6:26 Claude Code, Codex, DeepSeek Harness
7:27 Local models need a tighter harness
7:47 OpenClaw, Hermes Agent and SDKs
8:32 Do you even need an agent?
9:00 My verdict: which harness to pick
9:33 The whole idea in one line
Sources
LangChain, Improving Deep Agents with harness engineering (Terminal Bench 2.0, 52.8% to 66.5%): langchain.com/blog/improving-deep-agents-with-harness-engineering
Anthropic, Effective harnesses for long-running agents: anthropic.com/engineering/effective-harnesses-for-long-running-agents
Claude Code docs: code.claude.com/docs
Codex docs: developers.openai.com/codex
DeepSeek Harness: deepseek.com/en/harness
DeepSeek Harness on GitHub: github.com/deepseek-ai/deepseek-harness
OpenClaw docs: docs.openclaw.ai
Hermes Agent by Nous Research: github.com/NousResearch/hermes-agent
YouTube Hype: support.google.com/youtube/answer/15509925
Which harness do you use every day, and what would you change about it? Tell me in the comments.
#AIAgents #ClaudeCode #AgentHarness
Same model, very different results. The difference is the agent harness: the code around the model that runs the loop, gives it tools, manages its context, keeps its memory and decides what is allowed to run. In this video we open a harness up, follow one real coding task through it step by step, and then compare the harnesses you can use today: Claude Code, Codex, DeepSeek Harness (dsh), OpenClaw, Hermes Agent and the agent SDKs. I have run these harnesses myself, with cloud models and with local models on my own GPU, and I share what changed when only the harness changed. At the end you get my verdict on which harness fits which kind of work.
If this helped, like the video, and on the phone app tap Hype under the video. It helps the channel reach more people.
Subscribe for more AI explainers: youtube.com/@KGPTalkie?sub_confirmation=1
Chapters
0:00 Same model, two different results
0:38 Model vs harness: engine and car
1:12 Proof: same model, better harness
2:00 What I saw in my own tests
2:24 The five parts of a harness
3:24 One task, step by step
4:52 Context: the hard part
5:49 Control: who decides what runs
6:26 Claude Code, Codex, DeepSeek Harness
7:27 Local models need a tighter harness
7:47 OpenClaw, Hermes Agent and SDKs
8:32 Do you even need an agent?
9:00 My verdict: which harness to pick
9:33 The whole idea in one line
Sources
LangChain, Improving Deep Agents with harness engineering (Terminal Bench 2.0, 52.8% to 66.5%): langchain.com/blog/improving-deep-agents-with-harness-engineering
Anthropic, Effective harnesses for long-running agents: anthropic.com/engineering/effective-harnesses-for-long-running-agents
Claude Code docs: code.claude.com/docs
Codex docs: developers.openai.com/codex
DeepSeek Harness: deepseek.com/en/harness
DeepSeek Harness on GitHub: github.com/deepseek-ai/deepseek-harness
OpenClaw docs: docs.openclaw.ai
Hermes Agent by Nous Research: github.com/NousResearch/hermes-agent
YouTube Hype: support.google.com/youtube/answer/15509925
Which harness do you use every day, and what would you change about it? Tell me in the comments.
#AIAgents #ClaudeCode #AgentHarness










