# Incident.io's AI Spending and Productivity Insights How AI tools are reshaping engineering workflows and costs - **Source:** [The Pragmatic Engineer](https://blog.pragmaticengineer.com) - **Published:** 2026-11-12 - **Topic:** [technology](https://brifing.si/c/technology) - **Original article:** - **On Brifing:** https://brifing.si/s/9058d17d-f8f2-41f0-9aa3-5321c98c20f4 ## AI-Driven Engineering Evolution Incident.io, a software reliability platform, has been leveraging AI tools since 2024 to enhance engineering productivity. The company introduced 'effort points' to measure the complexity of pull requests (PRs), comparing AI-generated and human-written code. This metric provides a granular view of AI's impact, offering insights into cost and efficiency. By tracking these points, incident.io aims to optimize AI usage and reduce overhead while maintaining quality. ## AI Agents in Production In June 2025, AI agents began generating production code at incident.io, marking a significant shift in engineering workflows. Tools like GitHub Copilot and Cursor were already in use, but the launch of Claude Agents enabled parallel workstreams and improved efficiency. Today, nearly 80% of PRs include AI-generated code, reflecting a rapid adoption of AI tools across the engineering team. ## Cost and ROI of AI Tools Incident.io's AI spending peaked in summer 2026, reaching $300K monthly, but has since decreased to $1,600 per engineer monthly. The adoption of Claude Opus 5.5 significantly reduced costs, cutting monthly expenses by more than half. The company calculates ROI based on 'effort points,' which help compare the cost of generating PRs before and after AI integration. ## Team-Specific AI Costs The infrastructure team at incident.io spends five times more on 'effort points' than other teams, highlighting the variability in AI usage across departments. Tasks like iterating on evaluations and exploring production environments are token-intensive, leading to higher costs. This disparity underscores the challenge of applying a universal cost metric for AI-generated work. ## Productivity Gains AI tools have increased the number of PRs shipped per engineer weekly, with tenured engineers seeing a 70% rise. However, new engineers are advised to abstain from AI during their first month to focus on understanding the codebase. This approach ensures long-term productivity but skews short-term metrics. ## Incident Rates and Reliability Despite the increase in PR volume, incident.io has maintained a lower-than-expected defect rate. Most outages are now caused by missing production environment information rather than faulty code. The team is working on feedback loops to enable AI agents to learn from production behavior and reduce incidents further. ## Release Frequency Surge AI tools have boosted release frequency, with engineers averaging 2.8 releases per day in 2026, up from 1.9 in 2024. This increase reflects the efficiency gains from AI-assisted workflows, enabling faster deployment cycles without compromising quality. ## Challenges in Measuring Productivity Measuring engineering productivity solely through PR volume is fraught with challenges. Incident.io combines multiple metrics, including 'effort points' and release frequency, to assess AI's impact. While imperfect, these measurements provide valuable insights into the evolving role of AI in engineering. ## Future of AI in Engineering Incident.io's experience highlights both the potential and limitations of AI in engineering. While AI tools enhance productivity and reduce costs, they also introduce new challenges, such as managing token-intensive tasks and ensuring reliability. The company's focus on feedback loops and continuous improvement will shape its AI strategy moving forward. ## Sources covering this story - https://blog.pragmaticengineer.com/the-pulse-details-of-incident-ios-ai-spend/ — --- _Summary written by Brifing from the original reporting. Copyright in the original article belongs to its publisher; read the full piece at the link above._