Quality assurance · Production

LinkedIn platform team enables thousands of engineers with enterprise-scale background and foreground AI agents using MCP and sandboxed PR workflows

The problem

LinkedIn's engineering teams were experimenting with AI in silos, each reinventing the same infrastructure for prompt orchestration, data access, safety evals, and deployment, producing inconsistent proofs of concept unable to ship to production at scale.

Workflow diagram · grounded in source

Developer writes task spec

trigger

“A spec is how we translate the developer's intent into something the agent can reliably execute. Now, it's the contract between the developer and the system and at LinkedIn we deliberately structured these in a way that we want the devel…”

Agent runs in remote sandbox

ai_action

“we then orchestrate that flow in a remote sandbox environment, where the agent is free to run and execute and do what it needs to do. But because it's under the platform, there are certain restrictions that the agent cannot cross. It can…”

Agent uses MCP and native tools

integration

“we instantiate the agent there along with the context that you provided with your spec or your prompt and then set it up with the tools that you want it to have access to, which is either through MCP or native tools that are already avai…”

Pull Request produced

output

“Once the agent is done executing, we then move it on to the next stage of like, "Okay, let's create a pull request out of this change and kind of manage that workflow".”

Human reviews PR

human_review

“I want to review the PR, like I would a teammate's PR. So I would ask for changes or requests that certain changes need to be made for certain files or certain logic that it implemented”

Agent addresses review comments

feedback_loop

“the agent then picks it back up where it left off, addresses your changes and comes back with updates to the same pull request”

Reported outcome

LinkedIn built a unified agentic platform serving thousands of developers daily, with background and foreground agents sharing MCP tools, spec-driven task execution in secure sandboxes, PR-based human review, and reduction of engineering toil.

Reported metrics

Developer adoptionthousands of developers who are using these tools every single day

Engineering toil reductionreduction of toil

historical PR data volume for evalstens or hundreds of thousands of pull requests

Reported stack

MCPGitHub CopilotGitHub

Source

https://www.infoq.com/podcasts/platform-engineering-scaling-agents/

Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

LinkedIn built a unified agentic platform serving thousands of developers daily, with background and foreground agents sharing MCP tools, spec-driven task execution in secure sandboxes, PR-based human review, and redu…

What tools did this team use?

MCP, GitHub Copilot, GitHub.

What results were reported?

Developer adoption: thousands of developers who are using these tools every single day; Engineering toil reduction: reduction of toil; historical PR data volume for evals: tens or hundreds of thousands of pull requests (source-reported, not independently verified).

How is this quality assurance AI workflow structured?

Developer writes task spec → Agent runs in remote sandbox → Agent uses MCP and native tools → Pull Request produced → Human reviews PR → Agent addresses review comments.