AI Harness
An AI harness is an AI-enabled development environment or agent framework that integrates AI assistance directly into a developer's workflow — turning the execution environment itself into an active collaborator.
Approved Harnesses
SOK operates a whitelist policy: only approved harnesses may be used on SOK codebases and within SOK infrastructure. This protects data, ensures auditability, and keeps AI use within agreed governance boundaries.
Currently approved
- GitHub Copilot
- Amazon Kiro
If you need a harness that is not on this list, raise it with the DevEx team or your AI Champion.
Opensource Harnesses
Opensource harnesses (like OpenCode) are allowed when properly configured to only use SOK-approved LLMs and to not leak information. See the SOK Engineering Handbook for the latest guidance.
What a Harness Is Not
A harness is not a standalone chat tool (like Claude.ai or ChatGPT) used in isolation. Those are useful, but a harness turns your terminal, editor, or CI environment into an AI-capable agent — it can read files, run commands, write code, and take multi-step actions on your behalf.
That distinction matters for governance: harnesses have broader execution access than chat interfaces, which is why the approved list exists.
LLMs
Approved harnesses must be configured to only use SOK-approved LLM providers, listed by cloud platform.
| Model | AWS Bedrock | MS Foundry | Snowflake Cortex |
|---|---|---|---|
| Anthropic Claude | ✓ | 𐄂 | ✓ |
| Mistral AI | ✓ | ✓ | ✓ |
| Meta Llama | 𐄂 | 𐄂 | 𐄂 |
| Amazon Nova, Titan | ✓ | 𐄂 | 𐄂 |
| OpenAI GPTs | 𐄂 | ✓ | 𐄂 |
✓ supported · 𐄂 not supported