AI Dev 26 x SF | Tom Howlett：LLMs能否生成企业级质量代码？

DeepLearning.AIVideo2026年5月21日

AI Dev 26 x SF | Tom Howlett: Can LLMs Generate Enterprise Quality Code?

8.5Score

Watchable video resourceOpen original video

LLMs-generated code faces enterprise quality gaps requiring process/tool improvements to achieve sustainable production-level code generation.

Carnegie Mellon study shows Cursor users achieved 3-5x code velocity boost in fi
LLM-generated code complexity grows linearly over time requiring additional main
COBOL-based enterprise systems prove sustainable code generation requires archit

Jump quickly between sections.

§Introduction & Core Question
Poses enterprise code generation challenges analogous to 1990s web tech adoption lag
·Carnegie Mellon Findings
Shows Cursor users' velocity/quality paradox in code generation
·Quality Gap Analysis
Explains LLM code's maintainability issues in enterprise context
·Use Case Segmentation
Differentiates short-term vs long-term application requirements
§Solution Directions
Proposes SDLC/process/toolchain optimization strategies

See how the topics connect at a glance.

查看大纲文本（无障碍 / 无 JS 友好）

Key sentences worth saving and sharing.

Cursor users achieved 3-5x code velocity boost in first 3 months but slowed due to complexity growth
— Carnegie Mellon study
⬇︎ 下载 PNG 𝕏 分享到 X
LLM-generated code complexity grows linearly requiring maintenance costs after 6 months
— 2:12
⬇︎ 下载 PNG 𝕏 分享到 X
COBOL systems prove sustainable code needs architecture/process integration
— 3:58
⬇︎ 下载 PNG 𝕏 分享到 X

#LLMs#Enterprise Code#SDLC#Cursor#Carnegie Mellon Study