Six things have to be true before AI changes a workflow. Most companies have built two.

Source: Nate’s Newsletter Date: 2026-05-14 URL: https://natesnewsletter.substack.com/p/enterprise-ai-deployment-layer

Summary

Nate’s Newsletter argues that model access alone is not what separates companies seeing real workflow transformation from those stuck at impressive demos. The case: six conditions must hold simultaneously — a specific role in a specific workflow, the right data, correct permissions, a review process, and a measurable success metric — and most organizations have only assembled two. The piece is timed to the emerging investment pattern from major financial players into deployment-layer services rather than model licensing, reading that shift as the market acknowledging that implementation infrastructure is now the scarce variable.

Implications

This feeds the enterprise deployment layer thread directly — the argument is that the gap between “has access to frontier models” and “runs real workflows on AI” is a systems-integration and change-management problem, not a capability problem. That’s the same dynamic visible in the Databricks/GPT-5.5 partnership and the Codex-for-work usage pieces: the model is table stakes; the integration scaffolding is what compounds.

• Mid-market as the contested ground. Large enterprises can hire the integration capacity; self-serve PLG tools reach individuals. The mid-market (enough operational complexity to benefit, not enough engineering to self-serve) is where deployment-layer services will compete hardest over the next 18 months.
• The compounding gap. Once an organization has built the six conditions for one workflow, the second one is cheaper — this is the moat being assembled right now, not model access.
• Watch: Whether any vendor bundles the deployment-layer checklist as a managed onboarding product, effectively selling “all six conditions” rather than just the model.