引言

随着现代计算机时代的到来，自然语言与机器语言之间的鸿沟逐渐缩小。硬件与软件的进步使得编程经历了多次演变，而最近人工智能（AI）和高性能计算的飞速发展更是让这一鸿沟几乎消失。通过利用大型语言模型（LLMs），生成式AI（Generative AI）正在显著提升软件开发者的生产力、软件质量和市场投放速度。本文探讨了生成式AI在软件工程领域的应用、挑战及前景。

主要发现与益处

1. 创新与软件质量提升

• 创新驱动力：61%的受访组织认为，生成式AI在软件工程中的最大益处是促进了创新工作，如开发新软件功能和服务。通过自动化重复性任务，生成式AI使开发者有更多时间专注于创新和增值任务，从而激发更大的创造力。
• 软件质量提升：49%的受访组织表示，生成式AI的使用提高了软件质量。例如，通过提供代码建议，生成式AI可以减少错误并增强测试覆盖率，从而提升整体软件质量。

2. 生产力提升

• 早期估算显示，使用生成式AI的组织在软件工程功能上的生产力提高了7%至18%。特别是在编码辅助和文档编写等特定任务中，最大潜力分别达到34%和35%的时间节省，平均分别为9%和10%。
• 大多数组织将这些生产力增益用于创新工作（50%）和员工技能提升（47%），而非减少员工数量（4%）。

3. 员工满意度与协作

• 高达69%的高级软件专业人员和55%的初级软件专业人员表示，对使用生成式AI进行软件开发感到高度满意。
• 78%的软件专业人员对生成式AI在增强业务和技术团队之间协作方面的潜力持乐观态度。

采纳现状与未来趋势

1. 采纳阶段

• 目前，生成式AI在软件工程中的采纳仍处于早期阶段，90%的组织尚未实现规模化应用。
• 27%的组织正在进行生成式AI试点，11%的组织已经开始在软件功能中利用生成式AI。
• 大型组织（年收入超过200亿美元）的采纳率显著高于小型组织（年收入在1至50亿美元之间），前者75%已采纳或试点，后者仅为23%。

2. 未来展望

• 预计未来两年内，使用生成式AI工具的软件工作者比例将从目前的46%显著增长至85%。
• 到2026年，生成式AI预计将协助完成超过25%的软件设计、开发和测试工作。

挑战与风险

1. 基础条件不足

• 仅27%的组织拥有实施生成式AI所需的平台和工具，32%的组织具备人才基础。
• 超过60%的组织缺乏针对生成式AI的软件工程治理和培训计划。

2. 非正式使用风险

• 63%的软件专业人员使用未经授权的生成式AI工具，这可能导致功能问题、安全漏洞和法律风险，如代码泄露和知识产权问题。
• 近三分之一的员工通过自学生成式AI技术，而少于40%的员工获得了组织提供的培训。

实现潜力的策略

1. 选择并优先实施高收益用例

• 组织应识别并优先实施那些能够带来最大效益的生成式AI用例，如编码辅助、测试案例生成、文档编写等。

2. 风险管理

• 制定全面的风险管理策略，以减轻安全、知识产权和代码泄露等方面的风险。

3. 组织转型

• 引入生成式AI助手以增强软件团队，并准备相应的技术前提条件，如建立平台和工具库。
• 创建一个学习文化，为员工提供培训和跨技能培训机会，以支持生成式AI的采纳和使用。

4. 监测与优化

• 采用测量协议来监测生成式AI的影响，并根据实际反馈调整和优化用例的优先级。

结论

生成式AI正在改变软件工程的面貌，通过提升创新力、软件质量和生产力，为组织带来显著优势。然而，其采纳过程中也伴随着诸多挑战和风险，需要组织制定全面的策略来应对。通过选择合适的用例、强化风险管理、推动组织转型和持续优化，组织可以最大限度地发挥生成式AI在软件工程中的潜力。对于技术、IT、产品、战略、研发/工程、一般管理和创新领域的业务领导者而言，这一报告提供了宝贵的见解和实施指南。

An Analysis of the Capgemini 2024 Generative AI in Software Engineering Report

---

Introduction

The advent of generative artificial intelligence (AI) has disrupted the landscape of software engineering, ushering in a new era of automation, productivity, and innovation. The Capgemini Research Institute’s 2024 report, “Turbocharging software with Gen AI: How organizations can realize the full potential of generative AI for software engineering,” delves into the various aspects of generative AI’s impact on the software development lifecycle (SDLC). This article will summarize the key findings, benefits, challenges, and strategies outlined in the report, offering insights into how organizations can harness the full potential of generative AI for software engineering.

Key Findings and Benefits

Innovation and Software Quality Enhancement

One of the most prominent benefits of generative AI in software engineering is its ability to augment innovative work. According to the report, 61% of surveyed organizations cite enabling more innovative work, such as developing new software features and services, as the primary advantage of generative AI. This underscores the potential of generative AI to empower developers to focus on value-added tasks rather than time-consuming, repetitive work.

Moreover, 49% of organizations report improved software quality as a significant benefit. Generative AI tools, through code suggestions, error detection, and enhanced testing capabilities, contribute to reducing bugs and enhancing overall software quality.

Productivity Boost

Early estimates indicate that organizations leveraging generative AI have seen a 7% to 18% productivity improvement in software engineering functions. This gain is particularly evident in specialized tasks like coding assistance (with a maximum potential of 34% time savings and an average of 9%) and documentation creation (35% maximum potential with 10% average savings). These productivity gains are then utilized for further innovative work (50%) and employee upskilling (47%), with only 4% of organizations aiming to reduce headcount.

Employee Satisfaction and Collaboration

Generative AI is also positively impacting software professionals’ job satisfaction. The report reveals that 69% of senior software professionals and 55% of junior professionals report high levels of satisfaction from using generative AI for software engineering. Additionally, 78% of software professionals are optimistic about generative AI’s potential to enhance collaboration between business and technology teams.

Adoption Status and Future Trends

Early Adoption Stage

Currently, generative AI adoption in software engineering is still in its early stages, with 9 in 10 organizations yet to scale their implementation. Only 27% of organizations are running generative AI pilots, and 11% have started leveraging it in their software functions. Notably, large organizations (annual revenue > 20billion)areleadingtheway,with751-5 billion).

Accelerating Adoption

The report projects a significant increase in adoption over the next two years. Currently, 46% of the software workforce uses generative AI tools for various purposes (training, experimenting, piloting, and implementing), both authorized and unauthorized. This is expected to rise to 85% by 2026. By 2026, generative AI is anticipated to assist in more than 25% of software design, development, and testing work.

Challenges and Risks

Lack of Foundational Prerequisites

A major challenge facing organizations is the lack of foundational prerequisites for generative AI implementation. Only 27% of organizations have the platforms and tools in place, and 32% have the necessary talent prerequisites. Over 60% lack governance and upskilling programs for generative AI in software engineering.

Unofficial Usage Risks

Another significant risk stems from the informal use of generative AI tools. Of those who use generative AI, 63% employ unauthorized tools, exposing organizations to functional, security, and legal risks such as hallucinated code, code leakage, and intellectual property (IP) issues. Nearly a third of the workforce is self-training on generative AI, with less than 40% receiving formal training from their organizations.

Strategies to Realize the Full Potential

Select and Prioritize High-Benefit Use Cases

Organizations should identify and prioritize use cases that offer the highest benefits, such as coding assistance, test case generation, documentation, and code modernization. By focusing on these areas, organizations can maximize their investment in generative AI and realize quicker returns.

Mitigate Risks

A comprehensive risk management approach is crucial to mitigate security, IP/copyright, and code leakage risks.