Skip to main content
Optimize analyzes agent activity across users, teams, departments, and the organization. It combines session evidence with identity, tool, connector, policy, skill, and artifact data to measure how AI work is performed and where operating changes can improve quality, efficiency, and reuse. Optimize includes three related views:

​
Data used by Optimize

Optimize uses evidence from observed agent sessions and related Forge records. Depending on the integration and available source data, a session can include:
  • Model requests and responses
  • Tool calls, results, and errors
  • Files, repositories, URLs, commands, and working directories
  • MCP servers, connectors, and gateway routes
  • Policy decisions, approvals, reviews, and exceptions
  • Verification events and downstream artifacts
  • User, device, team, department, and subteam associations
Forge groups related activity into session timelines and derives normalized session facts from the available evidence. Signals are then associated with a scope, time window, taxonomy label, and supporting sessions. Analytics and recommendations use those persisted facts and signals as their source of truth. Fields that are not observed or retained by a source are not inferred.

​
Analytics

Analytics provides time-series and scoped rollups for the following areas.

​
Adoption and activity

Adoption metrics describe the amount and distribution of AI activity in the selected period. Results can be grouped by department, team, or subteam to show differences that are hidden by organization-wide totals. Adoption is evaluated alongside outcome metrics, including verification, rework, durable artifacts, and reuse. High activity does not necessarily indicate high organizational value.

​
Rework and verification

Optimize identifies signals associated with repeated generation, redo, refactoring, manual correction, and changes in direction. These signals can be related to under-scoped tasks, missing planning or acceptance criteria, and manual edits that are not converted into reusable instructions. Verification metrics cover available tests, automated checks, browser checks, query validation, review events, and human interventions. They indicate whether a workflow closes its verification loop before work is accepted.

​
Environment and tooling

Optimize reports constraints caused by the available AI environment, including: These distinctions help separate an operating issue from a platform, connector, procurement, or registry gap.

​
Reuse and durable value

Optimize measures whether AI-assisted work produces reusable organizational assets. Examples include skills, templates, tests, saved flows, documentation, and shared context. The analysis can identify repeated workflows that are candidates for a shared skill, successful context that remains local to one person, recurring manual corrections that have not been codified, and durable artifacts without an owner, audience, or distribution path.

​
Operating-model signals

Optimize can identify patterns associated with inconsistent agent use, such as tool hopping, abandoned sessions, repeated prompting without a changed plan, context fragmentation, and sessions that do not reach a clear outcome. These signals can be compared with adoption and improvement density over time to show whether early workflow friction is declining as usage grows.

​
Improvements

Improvements are actionable operating changes associated with repeated signals. They are linked to supporting sessions, signal evidence, a scope, and one measurement target. Examples include:
  • Create and publish a shared skill for a repeated workflow.
  • Convert recurring manual corrections into a template, acceptance criterion, or reusable instruction.
  • Add a connector or approved access path for repeated manual context transfer.
  • Add automated verification for work that repeatedly reaches human review without a reliable check.
  • Replace stale, unused, or local-only skills with an owned and governed version.
  • Create an owned context pack for a workflow with fragmented or low-quality shared context.
An improvement describes the operating change to install. The related finding and session evidence remain separate so that recommendations are auditable and can be evaluated against their target metric.

​
Goals

Goals define a target metric, scope, owner, and time horizon. Improvements can be linked to a goal to show which operating changes are expected to affect the target. Typical goal metrics include:
  • Manual rework rate
  • Verified-output or closed-loop rate
  • Connector-backed or approved-path usage
  • Durable artifact or shared-skill creation
  • Stale, unused, or local-only skill rate
  • Tool-sprawl or capability-gap pressure
Goal progress uses the same session-derived analytics as the Analytics view. Scope and organizational dimensions are retained so that aggregate progress can be compared with the teams and subteams contributing to it.

​
Scope and filtering

Analytics, Improvements, and Goals support a consistent scope model based on:
  • Time range
  • Department
  • Team
  • Subteam
Changing scope recalculates the associated metrics, signals, improvements, and goal relationships for the selected population and period.