How do you ensure quality of AI-generated code?
Most answers emphasize human review, strong requirements, and rigorous testing or verification as key to ensuring the quality of AI-generated code.
8 answers
Answers


Sagnik Nandy
Chief Technology Officer · Docusign
Human review
Sagnik Nandy says AI-generated code is validated by humans, who correct it when it fails and feed that back into the system.
The answer starts at 24:30.
The Engineering Leadership Podcast · Listen on Apple Podcasts

Michael Jenkins
C.T.O. · Threatlocker
Write good requirements
Michael Jenkins says you have to be really good at writing requirements, because you get much better AI-generated code the better the requirements you give it.
Security Weekly - A CRA Resource · Watch on YouTube

Thariq Shihipar
Member of Technical Staff · Anthropic
Extensive testing
Thariq Shihipar says quality comes from extensive testing: a strong verification harness, roughly 100 times more testing code, and fixtures and tools like Storybook to catch and verify every part.
The Developing Dev · The Developing Dev · Read
Anders Hejlsberg
Technical Fellow · Microsoft
Review the code
Anders Hejlsberg says the craft is shifting so agents type lines of code and programmers review the agents' work, meaning they do more reviewing and less typing to handle AI-generated code.
The Developing Dev · The Developing Dev · Read

Zach Lloyd
Founder / CEO · Warp
Verification and spec adherence
Zach Lloyd says ensuring quality means using computer-use verification and better specing so agents check for spec adherence, because review is the biggest bottleneck.
Dev Interrupted · Dev Interrupted · Read

Konstantin Klyagin
Founder · Redwerk
Standard software discipline
Konstantin Klyagin says ensuring quality in AI-generated code requires the standard software discipline of specification, documentation and test coverage, because AI only accelerates coding while the rest of the lifecycle still needs that discipline.
Chrisman Commentary - Daily Mortgage News · Chrisman Commentary - Daily Mortgage News · Listen at 7:03
Rob Zuber
CTO · CircleCI
Two-person review
Rob Zuber says two-person review is one way his team chose to manage risk, noting it is not the only possible way.
Dev Shorts · Dev Shorts · Read

👨🏾💻 Ben Ilegbodu
Senior Software Engineer, TV Client Foundations · Netflix
Verification
Ben Ilegbodu says verification is how to ensure quality, noting his team still reviews code agents generate because they only somewhat trust the output, and he expects verification may become checking that output does what it says.
Become an Epic Product Engineer · Become an Epic Product Engineer · Listen at 22:01
More Software Engineering questions
View all →- What principle matters most for AI in software development workflows?5 answers
- What is the most important oversight for agentic AI deployment?7 answers
- Which software development parts still need human oversight despite AI?5 answers
- What practice best improves team use of large language models?5 answers
- What matters most when orchestrating multiple AI agents?5 answers










