AI labs are replicating the cloud-era vendor lock-in playbook, but at the agent harness layer instead of infrastructure. As model tokens become commoditized, companies like Anthropic and OpenAI are building proprietary orchestration frameworks (Claude Agent SDK, OpenAI Agents API) to capture business logic and keep customers consuming their tokens. The argument draws a direct parallel to how AWS, Azure, and GCP locked enterprises in via CloudFormation and ARM templates — and how Terraform succeeded by providing a neutral abstraction layer above them. Model neutrality is argued to matter more than cloud neutrality because model rankings shift monthly, multi-model workflows are often optimal, and open-weight models make self-hosting viable. A neutral harness must be open source, multi-model by default, and profile-aware to exploit each model's strengths without being captive to any provider. LangChain positions its Deep Agents framework as this neutral layer.
