张海龙-从 Copilot 到 Coding AgentAI 驱动软件开发的未来.pdf

1、Agents Build Software演讲：张海Gru.ai 创始C 创始OsChina 联合创始The Future Software Development WorkflowProductDesignArchDesignCodingDebuggingTestingOperatingCreative WorksHandled by HumansMostly Handled by Coding AgentsRoutine WorksThe Future of Software DevelopmentFor CitizenFor EnterpriseTwo Types of Coding A

2、gentsCan Lovable Maintain Sustainable Growth?The Future of Software DevelopmentReproduce IssuePrepare PatchVerify ChangesIssuePRThe Ultimate AI DeveloperThe Challenges of Coding AgentsRank No.1 on SWE-bench verified leaderboard twice on 2024.08 and 2024.12.SWE-bench is the most recognized benchmark

3、for coding agents.https:/ fix agent is still premature,not ready for commercialization.Bug Fix Agent-Keeping on SOTAReproduce IssuePrepare PatchVerify ChangesRequirementUnderstandingCode AnalysisRead/Edit FilesTerminal UseBrowser UseSetup EnvIssuePRThe Ultimate AI DeveloperEssential Capabilities to

4、Complete the ProcessThe Challenges of Coding AgentsRequirementUnderstandingCode AnalysisRead/Edit FilesTerminal UseBrowser UseSetup EnvRequirementUnderstandingCode AnalysisRead/Edit FilesTerminal UseBrowser UseSetup EnvHuman High TouchHuman Low TouchHuman Assisted Missing Capabilities Focus on Verti

5、cal ScenariosCurrent SolutionsCurrent SolutionsCoding Agents in Your WorkflowAI currently has poor ability to understand the customers engineering environment.Low execution accuracy leads to a limited range of applicability.Enterprise Coding Agent Challenges60%unit test coverage significantly reduce

6、s bugs,but adds 30%more developer time.Unit Test Agent-Current IssuesTwo Functions of Unit Testing:1.Improve the quality of code writing2.Regression testingUnit Test Agent-Current IssuesScenario 1:Agent auto-completes unit tests on new code Code ReviewAgent Co