> Markdown version of [/videos/1771-ai-is-an-electric-bike-for-the-brain-stoyan-stefanov?t=2209](https://www.wearedevelopers.com/videos/1771-ai-is-an-electric-bike-for-the-brain-stoyan-stefanov?t=2209). Every page supports `.md` or `Accept: text/markdown`. Links point to the HTML versions so they work for humans too. Agent guide: [/agents.md](https://www.wearedevelopers.com/agents.md).

---

# AI is an Electric Bike for the Brain - Stoyan Stefanov

Stoyan Stefanov argues AI is an electric bike for developers, flawlessly accelerating prototyping and research. Find out why this performance expert outright refuses to ship AI-generated code to production.

- **Speakers:** 
- **Event:** Perfomance & AI Day
- **Published:** November 27, 2025
- **Duration:** 44:41
- **URL:** https://www.wearedevelopers.com/videos/1771-ai-is-an-electric-bike-for-the-brain-stoyan-stefanov

## Summary

Stoyan Stefanov draws a parallel between generative AI and an electric bicycle: both eliminate the friction of exploration and empower users to traverse challenging technical landscapes effortlessly. Initially skeptical after facing performance-lagging code generated by tools like Vercel's v0, Stefanov maps his experience to the "stages of AI grief." He ultimately discovers that leveraging AI significantly accelerates research, greenfield application prototyping, and local testing. Tools like the Cursor IDE and the Model Context Protocol (MCP) allow him to stay in a productive flow state, eliminating the dread of writing boilerplate structural code without sacrificing developmental velocity.

While AI excels at rapid iteration and uncovering potential optimizations directly within development environments, Stefanov emphasizes a strict engineering boundary: he refrains from shipping AI-generated code to production. In environments like Etsy, where web performance optimization demands meticulous adjustments and minimal vanilla JavaScript, relying on AI's bulk generation falls short compared to rigorous human verification and testing.

Taking software integration a step further, Stefanov details his creation of AI-Slow—a modern, machine-learning-driven evolution of the classic YSlow performance tool. By feeding HTTP Archive (HAR) tabular data into Python-based gradient boosting decision trees, he trains a regression model that predicts a page's Speed Index and simulates the exact performance gains of fixing specific bottlenecks. Exploring these custom machine learning models demonstrates that while AI acts as a powerful diagnostic compass, it requires hyper-relevant training data and an engineer's critical thinking to truly impact modern web performance.

**Keywords:** generative ai prototyping, web performance optimization, vercel v0 generation, cursor ide workflows, model context protocol, chrome devtools debugging, http archive har data, custom machine learning models, gradient boosting regression, speed index metrics, yslow performance rules, visual layout shift, vanilla javascript debugging, tabular training data, ai developer productivity

## Chapters

1. **Analyzing performance in automatically generated code** (00:00) — Evaluating the rendering and layout calculations of a font inspector created with an AI code generation tool.
1. **Navigating the stages of AI grief and adoption** (05:11) — How developers transition from initial denial and panic to embracing artificial intelligence for enhanced productivity.
1. **Integrating AI into web performance engineering workflows** (09:17) — Utilizing language models to rapidly create isolated test pages and generate research ideas without modifying production application behavior.
1. **Building greenfield projects with AI code generation** (13:04) — Accelerating exploratory development and feature implementation in a vanilla JavaScript music notation application.
1. **Exploring browser integration with the model context protocol** (15:51) — Allowing language models to directly analyze layout shifts and performance regressions within the development environment using standard testing protocols.
1. **Overcoming AI hallucinations and restrictive content guardrails** (21:22) — Why general language models temporarily struggle with accurate performance audits and how strict constraints motivate building custom infrastructure.
1. **Sourcing training data for predicting web page speeds** (25:47) — Extracting relevant performance metrics and tabular properties from the HTTP Archive dataset to evaluate real user experiences.
1. **Training a gradient boosting regression model in Python** (30:58) — Creating a decision tree that predicts targeted performance metrics by iteratively analyzing tabular feature splits.
1. **Evaluating page performance using customized ML prediction models** (36:49) — Applying a bespoke prediction model and game theory simulations to determine which code features impact page rendering times the most.
1. **Assessing the future of AI in web performance optimization** (42:26) — Why domain expertise remains essential for evaluating automated performance suggestions and manually curating machine learning training data.

## Related Moments

- [Motivations for adopting AI to enhance developer productivity](https://www.wearedevelopers.com/videos/1266-navigating-the-ai-revolution-in-software-development) (from "Navigating the AI Revolution in Software Development")
- [Shifting developer workloads and realistic AI productivity gains](https://www.wearedevelopers.com/videos/1830-wearedevelopers-live-speculaitions) (from "WeAreDevelopers LIVE - SpeculAItions")
- [Generating real-time website performance traces with AI](https://www.wearedevelopers.com/videos/100206-prompt-render-debug-ai-powered-react-development-with-chrome-devtools) (from "Prompt, Render, Debug: AI-Powered React Development with Chrome DevTools")
- [Exploring artificial intelligence as a solution for web accessibility](https://www.wearedevelopers.com/videos/1318-wearedevelopers-live-can-ai-save-accessibility-horrid-html-the-frontend-treadmill-and-more) (from "WeAreDevelopers LIVE - Can AI save Accessibility?; Horrid HTML; The Frontend Treadmill and more")
- [Using AI to rapidly prototype intuitive application features](https://www.wearedevelopers.com/videos/1377-rethinking-intelligence-ai-accessibility-and-the-future-of-inclusive-work-artur-ortega) (from "Rethinking Intelligence: AI, Accessibility, and the Future of Inclusive Work - Artur Ortega")
- [Accelerating coding workflows with AI code generation tools](https://www.wearedevelopers.com/videos/1328-how-to-avoid-llm-pitfalls-mete-atamel-and-guillaume-laforge) (from "How to Avoid LLM Pitfalls - Mete Atamel and Guillaume Laforge")

## Related Articles

- [MLOps And AI Driven Development](https://www.wearedevelopers.com/magazine/82-mlops-and-ai-driven-development)
- [Dev Digest 108 - Git off my cloud!](https://www.wearedevelopers.com/magazine/407-dev-digest-108-git-off-my-cloud)
- [The State of WebDev AI 2025 Results: What Can We Learn?](https://www.wearedevelopers.com/magazine/581-the-state-of-webdev-ai-2025-results-what-can-we-learn)
- [AI overspill Dec 2026: AI in a JAM, Blocking AI browsers, learning programming languages ](https://www.wearedevelopers.com/magazine/673-ai-overspill-dec-2026-ai-in-a-jam-blocking-ai-browsers-learning-programming-languages)

## Related Jobs

- [AI Software Engineer (Germany)](https://www.wearedevelopers.com/jobs/48317-ai-software-engineer-germany) at **Sunhat**
- [AI Full Stack Engineer](https://www.wearedevelopers.com/jobs/ext/1354435-ai-full-stack-engineer) at **Almedia**
- [Senior AI Agent Software Engineer (Go, Python) (m/f/x)](https://www.wearedevelopers.com/jobs/48277-senior-ai-agent-software-engineer-go-python-m-f-x) at **Dynatrace**
- [Staff Software Engineer, Copilot Experiences](https://www.wearedevelopers.com/jobs/ext/164361-staff-software-engineer-copilot-experiences) at **GitHub**
- [Principal Engineer - AI Search & Vector Infrastructure](https://www.wearedevelopers.com/jobs/ext/353953-principal-engineer-ai-search-vector-infrastructure) at **Redis**
- [AI Operations Manager (all genders)](https://www.wearedevelopers.com/jobs/48263-ai-operations-manager-all-genders) at **envelio**