管理人工智能生成代码中的风险，以更好地保护软件

Please refer to important disclosures on pages 42-44. Analyst certification is on page 42.William Blair or an affiliate does and seeks to do business with companies covered in its research reports. As aresult, investors should be aware that the firm may have a conflict of interest that could affect the objectivity of thisreport. This report is not intended to provide personal investment advice. The opinions and recommendations here- Introduction.......................................................................................................................3Key Findings......................................................................................................................4How Do AI Coding Assistants Work and Who Are the Main Players?...............................8AI Coding Massively Expands the Dev Tools TAM...........................................................23Risks of AI Coding............................................................................................................24AI Is Fundamentally Reshaping Software Development................................................27Are Developer Jobs at Risk? Tackling the Million-Dollar Question................................32AI’s Impact on Incumbent DevSecOps Vendors..............................................................36AI’s Impact on Infrastructure Software Ecosystem........................................................38AI Forcing All Software Companies to Reexamine Competitive Moats—but Not in theWay Many Fear..........................................................................................................38Conclusion.......................................................................................................................40 Apart from general-purpose chatbots, customer support AI agents, and third-party AI applicationstargeted at vertical markets like legal and content creation, adoption of AI in the enterprise hasbeen slower than many expected. The main gating factors have been scarce AI skillsets, model At the same time, we have seen widespread adoption of AI coding tools, which we believe is by farthe most prevalent use-case today for enterprise AI. Since the release of GitHub Copilot in 2022, AIcoding has evolved from the partial integration of AI within the coding process (e.g., code sugges-tions, autocomplete) to full compilations/blocks of code being generated from natural language While software development is unlikely to be fully automated by AI anytime soon—and the actualproductivity gains thus far for the average developer have been hotly debated—the implicationsof this paradigm shift in software development are vast and still underrecognized, with uncertainimpact on incumbent tool vendors, business models, and engineering headcount. What is clear isthat AI coding is the tip of the spear, with the massive influx of capital and start-ups into AI coding With this report, we aim to provide investors with an overview of the rapid evolution of AI codingtools, the competitive landscape for these tools, the true benefits to developer productivity fromAI coding, and the downstream impacts of AI coding on the traditional software development life- Our overarching conclusions are: 1.AI coding is fundamentally disrupting the SDLC– AI coding is materially accelerating soft-ware development cycles, automating workflows, and reducing time, cost, and resource con-straints historically associated with the SDLC. LLM vendors and innovative start-ups are keydisruptors in this new era. 2.AI coding massively expands the dev tools TAM– AI coding has introduced an entirely newmonetizable layer in the developer toolchain. We expect the average developer will ultimatelypay roughly $2,000 per year for a set of these tools, which would amount to an annual TAM of 3.Systems of record will be more critical than ever– with the rise of AI coding and softwareagents leading to orders of magnitude more code creation and pipeline complexity, systems ofrecord across the toolchain will be even more necessary to ensure code quality, auditability, 4.AI coding unlikely to reduce developer headcount– while developer roles (and the defini-tion of who is a developer) will change, human skills and specialization will still be needed toarchitect, supervise, troubleshoot, secure, and provide context to an ever-expanding numberof software projects (and agents). 5.AI coding creates secular tailwinds for infrastructure software– the expected flood of newAI applications and agents will result in vastly greater data volumes that will ripple across theinfrastructure software ecosystem, driving demand for databases, data security/governance,data observability, data protection, and storage. 6.AI coding will grow the software pie– instead of “AI eating software,” Jevon’s Paradox suggeststhat AI will grow the software pie as it becomes easier and cheaper to create new apps, enablingcode to eat into more of the labor economy.