Skip to content
ai-agent.review — AI Agent Reviews, home
← All news

Product updates

Ada adds API-tool latency reporting, with limits on what the clock measures

Vendor-reportedChecked: [release]

Editorial attribution: Gosia Madej

Prepared with AI assistance. Editorial attribution does not imply personal product testing or manual review of every story.

Published: · Event date:

Ada’s October 22 release adds median and slowest-ten-percent API response times to its usage report, plus the api_tool_performance MCP metric. The recorded duration measures API waiting, rather than the whole customer response.

Vendor-reportedChecked: [release]

What changes for users

Start with the slow tail, then inspect individual calls and errors. Compare the same tool mix and recorded population over time. If the customer still waits after API timings improve, investigate the unmeasured parts of the turn before attributing the delay to the integration.

Editorial assessmentChecked: [reports][conversations]

A stopwatch measures one amber segment of a longer path between two paper request cards.
AI-assisted editorial illustration of measuring one part of a response path; not an Ada report or measured latency data. · AI-assisted original vector illustration for AI Agent Reviews

The report charts response times over a selected date range and lets teams sort individual calls by duration. Conversation details compare a call with that tool’s typical response time over the last 30 days.

Vendor-reportedChecked: [release]

Ada defines the median using calls that recorded a duration. Its Slowest 10% figure is a threshold exceeded by one call in ten, rather than the longest call. The clock excludes the agent’s thinking and saving of results.

Vendor-reportedChecked: [reports][conversations]

The documented data starts on September 22, 2026; that is a historical coverage boundary, not the announcement date. Code tools and MCP tools have no response-time recording. The new reporting metric is available through MCP, but it does not make MCP-tool execution part of the measured population.

Editorial assessmentChecked: [release][reports]

Checked against sources

AI-assisted source check

How this was checked

AI-assisted checks compared the dated release with the API-usage definitions and Conversation View guide. They distinguish the report’s population, timing boundary and historical coverage; all are vendor documentation.

Editorial assessmentChecked: [release][reports][conversations]

No tool call, dashboard value or MCP response was tested. Recorded API waiting cannot establish total response latency, an improvement in satisfaction or complete timing coverage for every tool class. Source screenshot capture was blocked.

Editorial assessmentChecked: [release][reports][conversations]

Source evidence

Original source excerpt. This shows what the source says; it does not establish product performance.

Code tools and MCP tools do not record a response time.

Read original source [release]

Sources