Goldman Sachs Expands Deployment of Devin to Advance the Implementation of Autonomous Programming AI
The Cryptonomist
08-06 16:13
Ai Focus
Goldman Sachs deploys Devin in its technology department and plans to introduce Claude into trading and customer onboarding processes, indicating that its independent AI is accelerating its entry into the core business of financial institutions.
Helpful
No.Help

高盛正把一项面向软件开发的 AI 投入推进到实际生产环境。该行已在技术部门部署由 Cognition 开发的自主编程智能体 Devin,用于配合内部工程团队处理真实开发任务,而不再只是生成代码片段的辅助工具。

2025年已在技术部门铺开

报道显示,高盛在 2025 年将 Devin 引入技术部门,让数百个 AI 智能体与约 1.2 万名工程师和开发人员协同工作。与此前银行业试用的代码助手不同,Devin 可以在既有 IT 基础设施内独立完成部分流程。

它不仅能按指令写代码,还能自行梳理项目需求,完成编码、测试、提交人工审核,并在需要时继续修复问题。这意味着,AI 在软件开发中的角色,已从“辅助生成”转向“承担完整任务链的一部分”。

效率数据高于此前工具

高盛首席信息官 Marco Argenti 表示,Devin 的表现相当于一名新员工,预计效率可达到该行此前 AI 工具的 3 至 4 倍。高盛没有披露本行内部的完整统计,但 Cognition 在 2025 年底对客户部署情况做出的评估,给出了部分参考数据。

  • 某大型机构用 Devin 处理代码安全问题后,开发时间节省约 5% 至 10%
  • 另一个客户的单个漏洞修复时间,从 30 分钟降至 1.5 分钟
  • 经人工审核后无需大幅重写的代码提交,占比从约三分之一升至约三分之二

这些数据表明,这类自主编程系统的价值不只在于首次上线时的能力,还在于嵌入工作流后,可靠性会继续提升。对大型银行而言,日常修复和维护环节节省的时间,可能进一步改变技术预算和项目周期安排。

下一步扩展至交易与客户流程

在软件开发之外,高盛还计划把智能体 AI 用到更接近核心金融业务的环节。Argenti 表示,该行准备在 2026 年引入 Anthropic 的 Claude,用于交易、结算以及客户审查和开户流程。

他对外表示,现在更关注的已不是有多少员工在使用 AI 工具,而是一个想法转化为原型、再进入生产环境的速度。这也显示出,高盛正在把 AI 视为提升交付效率的基础设施,而不只是单点工具。

岗位结构变化引发讨论

随着自主 AI 覆盖范围扩大,岗位影响也成为外界关注重点。Argenti 认为,自动化会把员工从重复性任务中释放出来,转向更高价值的工作。但市场研究机构的判断更为谨慎。

彭博行业研究分析师估计,随着智能体 AI 承接更多原本由初级员工完成的任务,美国银行业最多可能有 20 万个岗位受到影响,其中包括大量初级开发岗位。文章提到,这也带来一个更现实的问题:如果初级岗位减少,未来资深工程师的人才培养链条可能受到冲击。

此外,Cognition 对客户部署的复盘还显示,Devin 在理解代码方面接近资深开发者水平,但在执行目标不清晰的任务时,表现更像初级员工。这意味着,企业在部署此类系统时,仍需要清晰定义任务范围,难以直接把模糊工作整体交给 AI。

Tip
$0
Like
0
Save
0
Views 161
CoinMeta reminds readers to view blockchain rationally, stay aware of risks, and beware of virtual token issuance and speculation. All content on this site represents market information or related viewpoints only and does not constitute any form of investment advice. If you find sensitive content, please click“Report”,and we will handle it promptly。
Submit
Comment 0
Hot
Latest
No comments yet. Be the first!
Related
Stacks Launches Bitcoin Staking with PoX-5: Self-managed time locks do not equate to a lack of new trust
Stacks has been activated at Bitcoin block height 960230, triggering a PoX-5 hard fork, and preparations are underway to implement the first batch of Bitcoin Staking arrangements. SIP-045 is designed to allow participants to pair their self-managed BTC time locks with STX locks in order to earn BTC rewards. At the same time, the staking process will be adjusted, and miners will receive 1000 STX of rewards per Bitcoin block during the launch phase. This attempt aims to address a long-standing challenge: how to generate on-chain profits for BTC while minimizing the need to entrust assets to centralized custodians.
币界网
·2026-08-23 12:28:56
128
OpenAI Funds 14 Studies on the "Intelligent Era": Why Can't AI Policy Experiments Be Answered Only by Model Companies?
On August 17, OpenAI announced funding for 14 projects led by independent organizations, with themes focusing on economic opportunities and social resilience, continuing the discussions on "industrial policies in the intelligent era" initiated in April of this year. The selected projects cover areas such as labor force, education, local governance, social security, and technology diffusion. Compared to a single product launch, this arrangement is more worth observing from a methodological perspective: when AI affects not only software functionality but also employment structures, skill investments, and public institutions, it becomes difficult to determine who will bear the costs, how the benefits will be distributed, and which policies are effective in different regions, relying solely on internal research within model companies.
CoinMeta
·2026-08-23 12:28:00
25
OpenAI Calls on California to Strengthen AI Security Legislation
OpenAI calls on California to revise the AI security legislation, SB 53, requesting an expansion of the monitoring scope for cutting-edge models and enhanced network security protection.
TechCrunch
·2026-08-23 00:41:06
57
Research indicates that cutting-edge AI laboratories still lack contingency plans for out-of-control situations
Research indicates that several leading AI laboratories have not yet made public their contingency plans for dealing with model failures, and regulatory authorities have also begun to promote requirements for related disclosures.
TechCrunch
·2026-08-23 00:21:23
57
Maxio CEO claims that AI's money-burning behavior is not the biggest problem
According to Maxio CEO, after enterprises deploy AI proxies, the real more difficult issue to handle is employee unrest, rather than single instances of token overspending.
Fortune
·2026-08-22 20:19:04
61
View More