Enterprise agents enter work through different control points
Examples from Google Cloud, GitHub and AWS show how agents enter code transformation, repository automation and data onboarding through settings, triggers and review points.
Artificial Intelligence··Midday
Transformation is divided into four parts
Google Cloud describes mainframe-to-cloud migration as a four-part arrangement. The Mainframe Assessment Tool visualizes dependencies, extracts business rules and generates documentation. Dual Run executes production workloads on the mainframe and Google Cloud at the same time and compares their outputs, while Mainframe Connector moves data into BigQuery, Spanner, Cloud SQL and Cloud Storage. Mainframe Modernization Agents, running on Gemini models, take the code-transformation role in this sequence. They offer two paths: reimagine the business logic or make a like-for-like conversion that preserves legacy behavior. The pipeline runs on Antigravity and uses the Model Context Protocol for context. The announcement therefore presents a division of work among components rather than one agent carrying the entire migration. Dual Run retains the comparison step, the assessment tool retains dependency and rule analysis, and the connector retains data movement. Google Cloud reports no customer count, time saving or error rate yet and directs interested organizations to contact it about pilots.[1]
A setting at task start, a trigger in comments
Two GitHub Copilot announcements expose separate control points for repository automation. When a user delegates work to the cloud agent, the reasoning level can now be selected beside the model for models that support the setting. The feature is generally available in Copilot Pro, Pro+, Business, Enterprise and Max. GitHub says a higher level may improve responses to complex problems while consuming more tokens and credits, though it gives no level names, default or numerical credit comparison. A second change lets issue or pull request comments start automations when they contain trigger text chosen during configuration. GitHub's examples include generating documentation from code changes, beginning an investigation of stack traces and error logs, and opening follow-up issues for refactoring or technical debt. All listed paid plans are covered, while Business and Enterprise accounts require an administrator to enable the cloud-agent policy. The entry states no limit on who may post a triggering comment and no cap on run frequency. One choice therefore sits at task creation; the other places a starting signal inside the repository's ordinary conversation flow.[2], [3]
The division of labor in a production example
AWS's Formula 1 customer study supplies a production example of the jobs assigned to an agent setup inside a data platform. Data Accelerator automates new-source onboarding, schema-change detection and governance classification. Running Claude Sonnet 4.6 on Amazon Bedrock AgentCore, it connects to SageMaker Unified Studio, Lambda, EventBridge, MWAA, S3 and Redshift. AWS and its customer report that adding a new data source fell from six to eight weeks to about 40 minutes of code generation followed by hours of deployment and review; agents handle 95 percent of onboarding tasks without human intervention; an 18-month integration backlog cleared in weeks; and identity-resolution processing time improved by 50 percent. Those figures come from the vendor and customer without an independent measurement method. Across the three companies, the agent's place varies: Google Cloud separates migration into components, GitHub exposes a task setting and a comment trigger, and the Formula 1 example preserves deployment and review after code generation. The common feature is a defined entry, trigger or approval point around the automated work.[4], [1], [2], [3]
