Thoughts on slowing the fuck down

2026-03-25 · Source: Simon Willison's Weblog · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Intermediate, quick

Summary

Mario Zechner, creator of the Pi agent framework, expresses significant concerns regarding the current pace and practices in agentic engineering. He argues that the drive to produce large volumes of code rapidly, often through orchestrated agent armies, leads to an unsustainable accumulation of errors. Unlike humans, who are bottlenecks in code generation and error introduction, agents can generate tens of thousands of lines of code in hours, allowing "booboos" to compound rapidly and create complex, unmanageable codebases without human oversight. This delegation of agency to agents results in developers losing understanding of their systems, leading to significant cognitive debt and pain only realized when it is too late.

Key takeaway

For AI Architects and engineering leaders adopting agentic development, recognize that the speed of agent-generated code can rapidly introduce unmanageable complexity and errors. Implement strict daily limits on agent code generation, aligning it with your team's review capacity, and prioritize manual development for critical architectural components and APIs to maintain system clarity and prevent cognitive debt.

Key insights

Agentic engineering's speed accelerates error accumulation and cognitive debt, making codebases unmanageable without human oversight.

Principles

• Agent errors compound faster than human errors.
• Delegating agency to agents increases system complexity.

In practice

• Limit agent-generated code daily.
• Manually write core architecture and APIs.

Topics

• Agentic Engineering
• AI Code Generation
• Software Quality
• Cognitive Debt
• Development Speed

Code references

• badlogic/pi-mono

Best for: AI Architect, CTO, VP of Engineering/Data, AI Engineer, Machine Learning Engineer, Software Engineer

Related on AIssential

• The Mythical Agent-Month — AI agents accelerate code generation but amplify essential complexity and introduce new forms of technical debt.
• Comprehension Debt: The Hidden Cost of AI-Generated Code — Over-reliance on AI code generation creates "comprehension debt," reducing human understanding and increasing systemic risk.
• Slow down to speed up — AI amplifies the cost of flawed initial decisions, making deliberate "slowness" in planning more critical for true speed.
• The Real Cost of Agent-Written Software — The true costs of agent-written software extend beyond initial generation, encompassing debugging and reliability.
• Agentic Code Review — AI agents make code writing cheap, shifting the software engineering bottleneck to human verification and review.
• Questions to rate your team, business impact on unhealthy code, and weekly readings 💡 — Optimizing software development requires structured team assessment, prioritizing code quality, and strategic AI integration for efficiency and career adaptation.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Simon Willison's Weblog.