AI是协作者，而非替代者--探讨AI高效协作开发

如果无法正常显示，请先停止浏览器的去广告插件。

1. AI-First Software Development Scott Shaw, APAC CTO, Thoughtworks

9. Coding assistance Our developers really like copilot Experience level： 5 4 3 2 How likely are you to recommend Copilot? 5 5 4 4 3 3 2 2 1 1 1 ● 1-5 x-axis is a Likert scale of “Strongly Disagree (1)” to “Strongly Agree (5)” ● 1-10 x-axis is the NPS Likert scale for “How likely are you to recommend” ● The stack colors show self-assessed experience level from 1-5 Copilot was frequently useful to me © 2024 Thoughtworks | Confidential 5 4 3 2 1 5 4 3 2 1 I would keep I would like like to to keep the the license license 5 5 4 4 3 3 2 2 1 1 5 4 3 2 1 9

14. Code assistants amplify everything The good and the bad Poisoned Context Automation Bias Anchoring Effect Security flaws Code is read far more than it is written. There are objective standards for “good” code but training sets only take correctness into account, not what makes the overall code base better. © 2024 Thoughtworks | Confidential 14

15. The trends in code quality are not encouraging Github codebases are getting bigger without getting better Churned Added https://gitclear-public.s3.us-west- 2.amazonaws.com/Coding-on- Copilot-2024-Developer- Research.pdf © 2024 Thoughtworks | Confidential Moved 15

16. Empirical evidence is not encouraging Github codebases are getting bigger without getting better Click to add body copy text https://gitclear-public.s3.us-west- 2.amazonaws.com/Coding-on-Copilot-2024- Developer-Research.pdf © 2024 Thoughtworks | Confidential 16

17. Practices still matter Engineering practices still matter, if not more. If you can code faster, can you fill the backlog faster? © 2024 Thoughtworks | Confidential Higher coding throughput If you can code faster, can you review faster? Can you ship faster? If you can produce more code, can you also keep your technical debt in check? 17

19. Software lifecycle Beyond coding: Potential for AI across the lifecycle ● ● ● ● ● ● ● ● ● ● Market Research Competitor Analysis User Needs Identification Feasibility Study Technology Exploration Regulatory Requirements Identification Security Considerations Existing System Analysis (for modernization) Mobile Platform Analysis (for mobile projects) Data Analysis (for data projects) Research ● ● ● ● ● ● ● ● ● ● ● ● Software Architecture Analysis Data protection impact assessment Requirement Gathering Compliance Requirement Identification Requirement Analysis Business Process Analysis User Interface Analysis Security Analysis Existing System Analysis Data Analysis Vulnerability Assessment Data Source Analysis Planning ● ● ● ● ● ● ● ● ● ● ● ● Project Scope Definition Communication Planning Project Governance Documentation Plan Mobile Platform Selection Resource Planning Budget Planning Risk Management Quality Assurance Planning Data Migration Plan Security Planning Project Scheduling © 2024 Thoughtworks | Confidential Analysis Design ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● Refactoring Coding Database Management UI/UX Implementation Security Implementation Reverse engineering Data Modeling Data Integration Data Security Implementation Integration API Development Testing (Unit and Component) Data Processing Implementation Continuous Integration Database Design Data Migration Data Validation Database Implementation Development System Design Software Architecture Design UI/UX Design Compliance-Driven Design Security Design API Design (for application development) Microservice Design (for modern applications) Mobile Design (for mobile applicalions） Data Pipeline Design (for data platforms) Risk management policy creation Incident response plan creation Infrastructure Design Database Design Privacy policy creation Terms of service creation Security policy creation Data retention policy creation ● ● ● ● ● ● ● ● ● ● ● ● ● Testing ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● Change Management Cutover Management Performance Tuning Mobile App Store Submission Environment Setup Release Preparation Data Migration Software Installation System Monitoring Load Balancing Compliance monitoring Backup and Recovery Setup SEO Setup Deployment Alpha Testing Compliance Testing Accessibility Compliance Defect resolution System Testing Non-functional Testing User Acceptance Testing (UAT) Beta Testing Mobile App Testing (for mobile projects) Data Testing (for data projects) Accessibility Testing Load and Stress Testing Compatibility Testing Defect tracking Unit Testing Integration Testing Functional Testing Regression Testing Automation Testing Test data management Test reporting Maintenance ● Release and patch management ● System audits ● Incident Management ● Security Updates ● Performance Tuning ● Backup and Recovery ● System Documentation ● Compliance Checks ● Decommissioning ● Mobile App Updates ● Data Validation ● Security monitoring ● Performance monitoring ● Security monitoring ● Capacity planning ● Software Upgrades ● User Support ● Performance monitoring ● Application monitoring 19

20. How we can help Big picture: The AI-first software delivery stack Buy OSS & self-host Chatbot Team assistance portal Coding assistant 3rd party model Buy Build Extensions to the existing delivery toolchain, using 3rd party models Issue tracker Wiki Developer Portal Prompt sharing platform Data stores for retrieval © 2024 Thoughtworks | Confidential Text chat Embeddings Coding specific chat Cloud Console OSS & self-host Foundation model access OSS Buy … … Design tools CI/CD Tool Buy Buy OSS & self-host 20

21. Consider the factory pre- electification © 2024 Thoughtworks | Confidential https://upload.wikimedia.org/wikipedia/commons/thumb/6/61/Line_shaft_and_powe r_looms_at_Boott_Mills%2C_Lowell%2C_Massachusetts.JPG/1600px- Line_shaft_and_power_looms_at_Boott_Mills%2C_Lowell%2C_Massachusetts.JP G?20140715052744 21

22. How did the electric motor change manufacturing? Electrify the line shaft (1890) Unit drives didn’t become common until the 1920s © 2024 Thoughtworks | Confidential Devine, W. D. (1983). From Shafts to Wires: Historical Perspective on Electrification. The Journal of Economic History, 43(2), 347–372. http://www.jstor.org/stable/2120827 22

27. How we can help Our Team Assistant Accelerator Knowledge Amplification Engineering Organization Team Thoughtworks recommended practices “just in time” Generative AI and prompt upskilling Team Assistant (web application) Prompts Knowledge Organisational experience and advice Low cost of experimentation 3rd party model services of choice ●Domain and architecture descriptions ●Organisational resources ●External resources Documentation / Guidance © 2024 Thoughtworks | Confidential 27

28. Haiven™ Team assistant How we can help We’re continually adding new features. Haiven TM now includes: ● General Chat, with saved ‘useful’ prompts for the team ● Epic breakdown ● Story brainstorming ● Technical Tasking ● Test DB exploration ● Knowledge Base (PDF / vector-DB query) ● Diagram visual understanding and discussion © 2024 Thoughtworks | Confidential 28

29. 1. Generative AI is poised to transform software delivery 1. Obsessing on coding assistants is a distraction 1. The real revolution will come when teams use AI to augment their collective knowledge of the business © 2024 Thoughtworks | Confidential 29