Open harnesses with customer-owned databases are the antidote to model-provider lock-in

The article’s prescription is explicit: memory must be owned by whoever is developing the agentic experience, and memory — and therefore harnesses — should be separate from model providers. Chase positions LangChain’s Deep Agents as the concrete pattern: open source, model-agnostic, using open standards like agents.md and skills, with plugins for Mongo, Postgres, and Redis, deployable either via LangSmith or any standard web hosting framework. The through-line is that the storage backend is pluggable but the harness itself is open and client-controlled, so the proprietary dataset accumulated through user interactions is never held hostage to a model provider’s platform decisions.

This gives builders a third option beyond “build a harness from scratch” and “adopt a closed provider harness” — adopt an open harness but own the memory substrate. It aligns with Boring tech wins for AI-native startups — simpler stack means faster AI-assisted shipping: Postgres and Mongo as memory backends are deliberately unglamorous choices. It pairs with Harness improvements that encode domain patterns transfer across models — invest once and keep the gain through model swaps — if your harness is open and your memory is in your own database, swapping the model becomes a prompt-tuning exercise rather than a full rebuild. For domain-intelligence products where memory is the moat, this is the default stance implied by Closed harnesses behind APIs create memory lock-in by design. It is also exactly what passes The sovereignty test — can you swap out a generalist model without losing your 'company veteran' expertise? — Nadella’s bar that a firm should be able to swap its generalist model without losing the company-veteran expertise built into its learning system.