A former Amazon software engineer recounts his experience in a startup-like AI org, where traditional practices like code reviews and design docs were replaced by rapid prototyping, high autonomy, and agentic coding. He shares lessons on scaling, risk tolerance, and the importance of customer feedback.
A short editorial from the VEONIB team on why this content matters.
The video argues that startup-like agility—rapid MVPs, high autonomy, and selective code reviews—can outperform traditional big tech practices in fast-moving AI projects.
Unlike generic startup advice, this firsthand account from inside Amazon reveals how even a giant can foster lean, high-risk teams, and why that sometimes leads to elimination.
Watch if you're an engineer or founder navigating AI development; adopt rapid prototyping and risk-based validation, but ensure team trust and clear boundaries.
Using AI agents to write and merge code with minimal human review.
Minimum Viable Product; a basic version released to gather early feedback.
A systematic examination of source code by peers to find issues.
The degree of independence given to engineers to make decisions and deploy code.
A work environment characterized by agility, risk-taking, and resource constraints.
Large, established technology companies like Amazon, Google, and Meta.
The level of acceptable risk in a project, influencing validation and review needs.
Moving testing earlier in the development cycle to catch issues sooner.
What is it like to work in an AI org at Amazon?
It feels like a startup within a big company: limited budget, high urgency, and broad responsibilities.
Why did the team skip code reviews?
For low-risk prototypes, they trusted coding agents and gave owners full autonomy to merge to prod.
How did they handle scaling manual tasks?
They decided to hire more people instead of automating, after weighing risks and rewards.
What is 'bandage programming'?
Quickly patching code to make it work without scalable design, often due to sunk cost and time pressure.
When should you use design reviews?
When risk tolerance is low, such as in high-scale systems where failure is catastrophic.
What are the risks of high autonomy?
Some engineers may not deliver or merge code, requiring rework and trust verification.
How did they validate changes without code reviews?
Through integration tests, unit tests, and left-shift testing before code leaves local repos.
What did the principal engineer teach about automation?
Not all manual processes need automation; sometimes hiring is more practical.
Why did the team switch to rapid prototyping?
Because assumptions often changed quickly, and early customer feedback was more valuable than lengthy design.
What happened to the engineer who didn't merge code?
He was moved to a different role at another company after the team had to redo his work.