LinkedIn puts agents on code review as Copilot opens pull requests from Slack
LinkedIn reviews pull requests with multiple agents, while Copilot makes changes from Slack. Panasonic scans fleet faults daily, and AWS's data-pipeline blueprint runs reviewed deterministic code rather than agents in production.
Artificial Intelligence··Evening
LinkedIn splits review work among several agents
LinkedIn sends each pull request to several agents using different models and reasoning approaches instead of relying on one off-the-shelf reviewer. The agents also check one another's findings. In the company's evaluation of 5,230 comments across 1,727 pull requests, 63.9 per cent of suggestions were implemented; acceptance ranged from 100 per cent for concurrency bugs to 40.6 per cent for security fixes. Event-driven queues on Kubernetes apply organisation policies separately from repository conventions.[1]
Copilot moves from chat to pull request
In GitHub's public preview, Copilot Business and Enterprise users can start an agent session by mentioning @GitHub in Slack. The agent answers questions about code and repository activity, triages issues, investigates failures in a secure cloud sandbox, makes changes, and can open a pull request. The session is shared with the team, existing GitHub permissions still apply, and a repository administrator can require extra approval before an agent-authored pull request is merged.[2]
Fleet diagnosis and data pipelines draw different production boundaries
The multi-agent workflow from Panasonic Avionics and AWS produces a daily diagnostic report for in-flight entertainment and connectivity faults across the active fleet. AWS reports a 20 per cent to 40 per cent efficiency gain in targeted uses; it is the company's own measurement. In AWS's separate data-operations blueprint, sub-agents prepare schemas, quality checks, and transformation code at build time. Once the output is reviewed and promoted, deterministic code rather than an agent remains in the running pipeline.[3], [4]
