hebel. today · 14 November 202632 / 80
Agentic Engineering · Coding Agents
LinkedIn automates 40,000 weekly pull requests
LinkedIn processes 40,000 pull requests weekly using an AI-native review system that performs an initial pass before human engineers engage. This system addresses the bottleneck created by coding agents, which reduce writing time from…
From 4 newsletters · 1 min read · one thing to try
LinkedIn processes 40,000 pull requests weekly using an AI-native review system that performs an initial pass before human engineers engage (ByteByteGo Newsletter). This system addresses the bottleneck created by coding agents, which reduce writing time from hours to minutes but leave review as a manual, fixed-capacity task (ByteByteGo Newsletter). The system functions by reading diffs and commenting on specific lines, allowing human reviewers to focus their limited time on judging the agent's fixes rather than performing initial discovery.
Engineers at LinkedIn built the system to be grounded, high-signal, codebase-aware, and timely, ensuring it flags real bugs rather than minor style issues. This approach mirrors the broader shift toward software factories, where teams design systems for agent-to-agent handoffs rather than manual code production (Arjun Iyer from The New Stack). Adam Tornhill similarly advocates for this eval-driven workflow, where engineers define precise test suites as specifications before agents implement the corresponding code, ensuring the output remains verifiable by domain experts (Adam Tornhill from Code for Humans and Machines).
Reliability remains a challenge, as agents can produce code that passes review but fails due to hidden system connections that a pull request diff cannot reveal (luminousmen Helping robots conquer the…). While automated review catches bugs, human oversight is still required to maintain the mental map of the codebase and its architecture, particularly when production failures occur outside the scope of the changed lines.
Try today · 30 min
Define a test-driven specification for your next feature before prompting an agent to implement the code.
Write a failing test case that captures the expected outcome, then commit it to your repository to serve as the agent's precise requirement.
You need
- Existing test framework
- Feature requirement description
Steps
- Identify the core behavior of the new feature.
- Write a test case that describes the expected input and output.
- Run the test to confirm it fails.
- Commit the failing test to your codebase.
- Prompt your coding agent to implement the solution that satisfies this test.