PyTorch Eng Director: Promo Hacking, Industry Shifts, Regrets | John Myles White

2026-06-08 · Source: The Peterman Post · Field: Technology & Digital — Software Development & Engineering, Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Intermediate, extended

Summary

John Myles White, former Director of Engineering on PyTorch at Meta, shares candid insights into big tech's evolving landscape. He notes Meta's shift towards business bottom-line efficiency, making it a less enjoyable and more stressful environment for employees due to a perceived oversupply of engineers. White criticizes the pervasive "promo hacking" culture, especially within AI Infra, where engineers prioritize shipping features for promotion over building quality systems, often leading to the development of code intended for deletion. He contrasts this with PyTorch's culture, which emphasized craft and held a higher hiring bar, fostering genuine engineering skill despite slower promotion rates. White also reflects on his early career developing Meta's A/B testing tools like Deltoid 3, the Julia programming language's design philosophy, and the importance of rigorous statistical understanding in industry.

Key takeaway

For software engineers and data scientists navigating big tech careers, recognize that promotion-driven cultures can incentivize short-term gains over genuine skill development. Prioritize joining teams, like PyTorch, that value engineering craft and long-term impact, even if it means slower initial promotions. Actively seek out and engage with senior leadership through opportunities like office hours to gain visibility and influence, rather than fearing it.

Key insights

Big tech's promotion culture often incentivizes short-term "promo hacking" over long-term engineering quality and skill development.

Principles

• Engineer supply/demand shifts impact retention strategies.
• Promotion systems can decouple skill from career growth.
• Rigorous A/B testing prevents superficial "wins."

Method

Julia aims to achieve C-like performance in a high-level language by avoiding dynamic overheads and lazy evaluation prevalent in R and Python, which cause significant slowdowns (e.g., 1,000 to 10,000 times slower).

In practice

• Prioritize teams valuing craft over rapid promotion.
• Scrutinize A/B test results beyond superficial metrics.
• Attend skip-level office hours for career growth.

Topics

• PyTorch
• Engineering Culture
• Promotion Incentives
• A/B Testing
• Julia Programming Language
• Statistical Rigor

Best for: Director of AI/ML, Software Engineer, Data Scientist

Related on AIssential

• I Quit My $130,000 ML Engineer Job After Learning 4 Lessons — Big Tech's stability and mature systems can stifle innovation and personal growth for action-oriented engineers.
• PyLadies entrepreneurs and career development — Diverse backgrounds and strategic self-advocacy are crucial for navigating tech careers and entrepreneurship, especially for women.
• Jeremy Howard interview at PytorchCon with Anna Tong — AI should augment human skills and agency, not replace them, fostering continuous learning and distributed power.
• Small Experiments, Cheaper Decisions: A Case Study in Staged Promotion for Micro-Pretraining — Staged promotion protocols can significantly reduce micro-pretraining experimental costs by efficiently screening configurations across varying budgets.
• Podcast: Craig McLuckie on Culture as a Team's Operating System in the AI Era — AI coding tools are disrupting engineering culture and career paths, demanding deliberate cultural design and new leadership approaches.
• Quoting Mitchell Hashimoto — Programming languages are increasingly fungible, reducing vendor lock-in and making specific languages expendable.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by The Peterman Post.