Claude Code, Cursor, Codex, and Gemini CLI — running through Databricks AI Gateway with centralized governance, observability, and cost controls. No API keys. No vendor lock-in.
Direct adoption of AI coding agents creates fragmented governance, disconnected procurement, and limited visibility.
MCP tools can grant agents excessive privileges to sensitive data — engineering tickets, design docs, customer issues. No centralized policy enforcement across tools.
Multiple teams adopting different AI tools create disparate billing streams. No unified view of spend, no budget controls, no capacity planning.
When teams use different assistants simultaneously, leadership loses insight into adoption rates, usage patterns, ROI, and developer productivity impact.
Direct API keys to a single provider create dependency. Over-provisioning or under-provisioning capacity as the market evolves rapidly.
Separate logins per service, inconsistent authentication policies, no unified audit trail across coding tools.
Token usage, model costs, and team attribution scattered across vendor dashboards. No way to tie AI spend to business outcomes.
Databricks AI Gateway centralizes governance, observability, and cost management for all AI coding agents.
Single identity system via OAuth — no separate API keys per tool. Agent data access governed centrally with audit logs in Unity Catalog. MCP servers managed within Databricks infrastructure. MLflow tracing for compliance.
Consolidated billing across all coding tools via Foundation Model API. Per-team quotas, rate limiting, and cost guardrails applied organization-wide. One invoice — developers choose their preferred tool.
OpenTelemetry metrics flow into Delta tables in Unity Catalog. Track adoption by team, department, or region. Measure developer velocity. Self-service AI/BI dashboards on gateway telemetry.
Developers use their preferred coding agent. The gateway handles authentication, routing, governance, and telemetry.
The ucode CLI auto-detects AI Gateway, authenticates via OAuth, and configures each agent automatically.
First launch prompts for workspace URL, authenticates via OAuth, and writes managed config files. No API keys, no manual setup, no environment variables.
Add Databricks MCP servers to any compatible agent — SQL execution, Vector Search, UC Functions. Centrally managed, token-authenticated.
Switch between Claude, GPT-4o, Gemini, and open-source models without changing tools. The gateway handles routing — developers choose their preferred agent.
A phased approach to rolling out governed AI coding at enterprise scale.
Databricks workspace with Foundation Model API + AI Gateway. Tier 4 capacity for production scale.
10–50 developer cohort using ucode CLI. Validate adoption, measure velocity, collect feedback.
Per-team quotas, model allow-lists, PII guardrails in AI Gateway. Cost attribution by team.
Provisioned throughput for guaranteed capacity. System tables for FinOps. AI/BI dashboards on telemetry.