TalksEngineering in production20 minRetired
No Rewrite, No Safety Net: Navigating 13 Years of React at Scale
I don't give this one any more.
Talk outline: A small React change becomes a different engineering problem in a 13-year-old application serving tens of millions of users and generating tens of millions in annual revenue. Five developers across three squads, without dedicated QA or automated testing and with limited observability, must maintain multiple generations of React, state management, data fetching and tooling. Deprecated paths, partial migrations, undocumented assumptions and edge cases may protect real customer workflows and revenue. Rewrites and indiscriminate upgrades are not viable. Focus on reconstructing execution paths and a reliable mental model; using AI agents for targeted code discovery, pattern comparisons and tests while constraining context to relevant modules, interfaces, assumptions, production evidence and non-goals; retaining human judgment on release safety; and understanding the multiplied infrastructure cost of small frontend decisions at scale. Goal: a practical mental model for remaining effective in atypical large-scale React systems. See the current talks.
The premise
How do you change a 13-year-old React application serving tens of millions of monthly users when five developers are spread across three squads, there is no dedicated QA, no automated testing, and limited observability?
This technical case study explores how we navigate Smallpdf across multiple React paradigms, deprecated code paths, undocumented assumptions, infrastructure costs, and an enormous potential blast radius. We will examine how to build a useful mental model, introduce AI assistance without overwhelming it with context, assess risk, and migrate progressively without attempting a rewrite or disrupting the users and revenue depending on the system.
Who it's for
Frontend and full-stack engineers working on mature, high-traffic React applications, especially those balancing legacy architecture, limited testing, and continuous delivery.
You leave with
- Build a useful mental model of a large, evolving React codebase before changing it.
- Use task-specific context to make AI-assisted code investigation more reliable.
- Assess blast radius, infrastructure cost, and migration risk while shipping incrementally.