[{"data":1,"prerenderedAt":150},["ShallowReactive",2],{"blog-post-blog_en-coding-agenten-im-pull-request-workflow":3},{"id":4,"title":5,"body":6,"cover":134,"date":135,"description":136,"draft":137,"extension":138,"meta":139,"navigation":140,"path":141,"seo":142,"stem":143,"tags":144,"__hash__":149},"blog_en\u002Fen\u002Fblog\u002Fcoding-agenten-im-pull-request-workflow.md","Coding Agents in the Pull Request Workflow: Governance for Growing Teams",{"type":7,"value":8,"toc":127},"minimark",[9,13,18,21,24,59,66,70,73,76,108,111,115,118],[10,11,12],"p",{},"Coding agents in the pull request workflow are not another autocomplete step. Tools such as OpenAI Codex, GitHub Copilot coding agent, and Google Jules can now take whole tickets, work in cloud environments, and deliver branches or pull requests. For growing teams, the leadership question is changing: not \"may developers use AI?\", but \"which work may an agent do under which rules?\"",[14,15,17],"h2",{"id":16},"what-coding-agents-change-in-the-pull-request-workflow","What Coding Agents Change in the Pull Request Workflow",[10,19,20],{},"A coding agent is productive when the task is small, verifiable, and protected by tests. It becomes risky when it has to guess unclear product decisions, hidden architecture rules, or missing security boundaries.",[10,22,23],{},"The key differences from classic AI assistants:",[25,26,27,35,41,47,53],"ul",{},[28,29,30,34],"li",{},[31,32,33],"strong",{},"Asynchronous work:"," The agent works in the background. That increases throughput, but also the number of parallel changes that reviews must handle.",[28,36,37,40],{},[31,38,39],{},"Repository context instead of chat context:"," The agent reads code, issues, tests, and documentation. Weak internal documentation is not compensated for. It becomes more visible.",[28,42,43,46],{},[31,44,45],{},"Automated branches:"," Changes appear as commits or pull requests. Branch protection, CI, review rules, and ownership become part of the agent strategy.",[28,48,49,52],{},[31,50,51],{},"More execution power:"," Agents can run tests, change dependencies, or touch configuration files. That requires clear rights and boundaries.",[28,54,55,58],{},[31,56,57],{},"New quality risks:"," A plausible pull request can still be wrong in domain logic, too large in scope, or inconsistent with the architecture.",[10,60,61,62,65],{},"Coding agents should therefore not be counted as additional developers. It is more useful to treat them as ",[31,63,64],{},"delegated automation"," with human accountability.",[14,67,69],{"id":68},"where-teams-need-rules-before-the-first-agent-pr","Where Teams Need Rules Before the First Agent PR",[10,71,72],{},"The typical mistake is a pilot without hard working boundaries. An agent receives a ticket, delivers code, and only during review does the team discuss whether it should have been allowed to change that area at all.",[10,74,75],{},"Before production use, teams should define at least:",[25,77,78,84,90,96,102],{},[28,79,80,83],{},[31,81,82],{},"Suitable tasks:"," Adding tests, small bug fixes, refactorings with clear acceptance criteria, and documentation updates. Product logic, authentication, and data migrations should stay tightly controlled at first.",[28,85,86,89],{},[31,87,88],{},"Repository access:"," Which repositories, files, and secrets are off limits? Agents do not need blanket trust across the whole organisation.",[28,91,92,95],{},[31,93,94],{},"Mandatory review:"," No agent PR is merged without human review. Review focus should be domain logic, side effects, security, and architecture boundaries.",[28,97,98,101],{},[31,99,100],{},"CI as the minimum standard:"," Type checks, tests, linting, and security checks must run before review starts. Missing tests are a signal against delegation.",[28,103,104,107],{},[31,105,106],{},"Traceability:"," Prompt, plan, relevant logs, and test output must be visible in the pull request.",[10,109,110],{},"A good starting point is a small backlog of recurring, well-tested tasks. If quality and review effort stay stable there, adoption can gradually expand into other areas.",[14,112,114],{"id":113},"why-this-matters","Why This Matters",[10,116,117],{},"Coding agents can increase delivery speed, but they also amplify existing weaknesses. Unclear architecture, missing tests, poor tickets, and vague ownership are not solved by agents. They are simply turned into pull requests faster.",[10,119,120,121,126],{},"For leaders, the economic point is simple: an agent PR only saves time when review, rework, and operational risk do not rise more than implementation cost falls. Teams that introduce coding agents well are not only investing in tools. They are investing in better specifications, stronger tests, and clearer technical leadership. An ",[122,123,125],"a",{"href":124},"\u002Fen\u002F#packages","Architecture & AI Review"," can help define suitable use cases and guardrails early.",{"title":128,"searchDepth":129,"depth":129,"links":130},"",2,[131,132,133],{"id":16,"depth":129,"text":17},{"id":68,"depth":129,"text":69},{"id":113,"depth":129,"text":114},"\u002Fimg\u002Fblog\u002Fki-coding-assistenten-cover.jpg","2026-06-05","Coding agents are entering the pull request workflow. What growing software teams need to clarify for security, quality, and delivery speed.",false,"md",{},true,"\u002Fen\u002Fblog\u002Fcoding-agenten-im-pull-request-workflow",{"title":5,"description":136},"en\u002Fblog\u002Fcoding-agenten-im-pull-request-workflow",[145,146,147,148],"AI","Developer Tools","Engineering Leadership","Software Quality","v7QFRu5W6xlQD2hiWHNxefjybU5e_bp65vUb4daxV70",1781596426445]