AIToday
Open-Source AIDatabricks AI BlogPublished: Nov 19, 2026, 04:00 JST

Databricks launches Unity Gateway Insights to surface agent token waste

Send on LINE
Databricks launches Unity Gateway Insights to surface agent token waste

3 Key Points

  1. What happened

    Databricks introduced Unity Gateway Insights, which uses Genie to analyze up to 30 days of agent traces and rank token waste by cost — covering recurring MCP errors, simple tasks run on expensive frontier models, and repeated workflows worth turning into shared skills.

  2. Why it matters

    Insights points teams to the sessions where tokens are being spent without a matching return, such as a workstream with 10% of sessions but 35% of tokens, so cost-cutting starts from evidence rather than budget guesses.

WHO IT HITSEngineering leaders and platform teams running coding agents such as Claude Code, Codex, or GitHub Copilot through Unity Gateway can see which sessions waste tokens and where to fix MCP errors or adopt Smart Routing.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, every morning.

Context & Analysis

Databricks's own rollout of coding agents to thousands of engineers produced concrete lessons that shaped Insights. Offline replays showed no performance gain from upgrading from Opus 4.8 to Opus 5, so the company stayed on the cheaper 4.8. Trace analysis also showed that simple prompts were going to expensive models, prompting adoption of Smart Routing. The largest single finding was $1.2M in costs recovered a year from just seven small bugs in MCP servers — $499K in wasted tokens plus 12,000 hours of engineers waiting on agents — and fixing all seven took about an hour.

Insights packages that investigation into a product. It samples sessions and groups them into workstreams by kind of work, then checks each workstream for failing MCP tools, bloated context, frontier models on simple tasks, and repeated workflows. It depends on capabilities a standalone gateway does not have: automatic OpenTelemetry tracing for agents launched with the ug CLI, a governed Delta trace table in Unity Catalog, Genie with the Genie Ontology for business context, and Databricks SQL for scale across millions of traces.

Governance is built in by default. Because traces hold prompts, code, and tool output, only the table's owner can query it until admins grant access with Unity Catalog grants and row filters. Insights reads the table in place, so sensitive data is not exported for analysis.

FAQ
What time ranges can I analyze?
Insights can generate a report for the last day, 7 days, or 30 days of traffic flowing through Unity Gateway.
Can I ask my own questions in a report?
Yes, you can add up to 10 questions of your own to a report, such as whether the newest model is worth it for your team's work.
What counts as a common workflow to codify?
A workflow counts when it recurs in at least three sessions or across two engineers.
Databricks AI BlogRead Original Article
Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.

Questions and answers are published on this page.

Related Articles

Next articleAIO Helper blocks AI calls before they break its $50 budget