Peter Steinberger on Building in the Agent Era

Engineering moves from manual coding to agentic workflow design.

Modern development requires architects who create pipelines for agents to execute tasks. Humans reserve focus for strategic direction and product vision.

Role change. Software engineering is shifting from writing code to designing pipelines where agents autonomously generate, test, and review code.

Effective agent management requires a manager mindset, not micromanagement.

AI-generated code is acceptable when it achieves the project goal.

Management style. Successful agent management requires shifting from a 'micromanager' mindset to a 'manager' role, accepting that AI-generated code may not be perfect as long as it achieves the project goal.

Agents handle repetitive tasks but struggle with complex system architecture.

AI agents can navigate large codebases but lack holistic system understanding. Human oversight remains critical for strategic direction and product vision.

System limits. AI agents excel at handling repetitive tasks and noise, but they struggle with high-level system architecture and understanding how new features interact with existing, complex codebases.

Human authenticity is critical as AI-generated content becomes ubiquitous.

Generic AI-written communication is increasingly viewed as 'slop' and is often ignored or blocked. Maintaining a distinct human voice helps cut through digital noise.

Open-source projects must filter low-effort AI-generated pull requests.

Contributors should prioritize work demonstrating genuine human effort, such as handwritten code or logs. This filtering maintains the quality of open-source projects.

Builders must use their own software to maintain product quality.

Building for external critics rather than personal utility can degrade product quality. Losing the 'fun' of building often leads to a decline in user experience.

Creative joy. Product development is most effective when the builder uses their own software, as losing the 'fun' of building often leads to a decline in product quality and user experience.