Slow down to speed up: so much has changed in 6 months’ time

2026-06-23 · Source: The Pragmatic Engineer · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Intermediate, medium

Summary

The software engineering landscape has undergone a rapid transformation in the last six months due to advanced AI agents like Opus 4.5 and GPT-5.4. This shift is exemplified by Meta's recent "most embarrassing outage," attributed to AI-generated, AI-reviewed code and significant cuts to security and quality teams. Data indicates teams using AI agents now ship 5x more pull requests than two years ago, and developers using AI harnesses produce 2.5x more code, increasing from 3,500 to 8,600 lines on average. Pull request sizes are also up 3x, with more AI changes accepted without human review. Major tech companies like Anthropic, OpenAI, Google, and Uber are deeply integrating AI into their workflows, with Anthropic reporting 70-90% of its code generated by Claude. Even traditional companies like Cisco and JP Morgan Chase are heavily investing in AI developer tools, indicating widespread adoption despite potential risks to software quality and reliability.

Key takeaway

For engineering leaders overseeing AI integration, you must balance rapid development with robust quality assurance. While AI agents significantly boost individual output, relying solely on AI-generated and AI-reviewed code risks critical system failures, as seen with Meta's outage. Prioritize investing in human-centric review processes and dedicated integrity teams, rather than solely reallocating resources to AI training, to prevent eroding core product stability and maintain long-term reliability.

Key insights

Rapid AI agent adoption boosts individual dev output but risks software quality and organizational integrity without human oversight.

Principles

• AI-generated code requires robust human review.
• Aggressive AI integration can degrade core product stability.
• Individual productivity gains don't guarantee team efficiency.

Method

Companies integrate AI agents via internal tools, AI code review, and parallel agent execution, often replacing PRDs with prototypes and automating bug fixes.

In practice

• Implement tiered AI code review policies.
• Develop internal AI agent orchestration platforms.
• Prioritize human oversight for critical AI-generated changes.

Topics

• AI Agents
• Software Engineering
• AI Code Generation
• Developer Productivity
• Software Quality
• Engineering Management

Best for: CTO, VP of Engineering/Data, Executive, Software Engineer, AI Engineer, Director of AI/ML

Related on AIssential

• 😺 Amazon spent $200B and broke its own website — AI-generated code can introduce significant instability and operational risks if not managed with rigorous human oversight.
• AI Agents Won't Save a Broken Process. They'll Just Make It Fail Three Times Faster. — AI agents accelerate failure in broken processes, not fix them.
• Coding agents are giving everyone decision fatigue — Coding agents shift software development bottlenecks from code generation to human judgment, intensifying work and causing decision fatigue.
• LAI #128: Developers Using AI Were 19% Slower. They Thought They Were 20% Faster. — The perception of AI's impact on developer productivity often misaligns with actual performance, especially on complex tasks.
• 4 years of observing and living through the AI era — Generative and agentic AI tools offer unprecedented automation and creation capabilities, shifting the focus to human validation and decision-making.
• Code Is Not Cheap: How to Multiply Your AI’s Output With Software Fundamentals — AI-assisted coding, while prevalent, demands human engineering expertise to mitigate quality issues from "jagged" model outputs.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by The Pragmatic Engineer.