Large-Scale React Native Development in the Age of AI

Presented at App.js Conf 2025. This talk examines how AI tools are reshaping development workflows in large-scale React Native applications.

Key Takeaways

• AI transformation extends beyond coding speed
  Tools like ChatGPT, Claude, and Cursor are fundamentally changing planning, collaboration, and complexity management across the entire development lifecycle, comparable to Visicalc’s impact on accounting
• The plan.md workflow promotes diligence over “vibe coding”
  Creating a persistent scratchpad for AI collaboration helps retain context between sessions and scales from individual tasks to technical documentation and epic breakdowns
• AI augments PR reviews and technical documentation
  Workflows using GitHub CLI and AI agents enable comprehensive code reviews that track data flow across files and ensure consistency, augmenting rather than replacing human diligence
• Development roles shift toward higher-order tasks
  Like accountants after spreadsheets, developers move toward sophisticated problem-solving, with AI eliminating routine work while amplifying human creativity
• Engineers are “ridiculously empowered” with new tools
  Success requires mastering these “Lambos” while differentiating through insight, creativity, taste, communication, and persuasion

The AI Transformation in React Native Development

AI tools such as ChatGPT, Claude, GitHub Copilot, and Cursor are fundamentally changing how teams plan, collaborate, and manage complexity within the development cycle. These tools extend beyond mere coding acceleration.

Large React Native teams face unique challenges in this new landscape. Emergent AI workflows go beyond simple coding assistance, requiring practical strategies for responsible AI adoption in production environments.

Historical Context: The Spreadsheet Revolution

In 1979, accounting was a manual, repetitive task performed entirely on paper. Errors required recalculation of every subsequent row by hand, creating a “slow, fragile, and painful process.” The Apple II computer, though gaining popularity among enthusiasts and in classrooms, was not yet taken seriously for business work.

VisiCalc, the first electronic spreadsheet, changed everything. Built by a former MIT computer science student frustrated by manual recalculations, this $100 program ran on the Apple II and instantly recalculated entire sheets when values changed. People who had never touched a computer began buying Apple products just for this single application, nearly doubling Apple II sales in 1980.

The tool enabled new workflows like financial modeling and “what if” scenarios, reducing 20-hour tasks to 20 minutes. This historical parallel establishes the framework for understanding how AI is similarly transforming software development, creating new engineering workflows that extend far beyond code generation.

Scaling React Native: The Development Pipeline

React Native teams typically remain small due to the framework’s operational efficiency, often maintaining a “startup style scrappiness.” However, as companies scale, development processes formalize into structured pipelines including product requirements documents (PRDs), design, technical specification, task breakdown, code implementation, code review, continuous integration (CI), quality assurance (QA), and staged app releases.

AI workflows are emerging at every stage of this pipeline. Product Managers use AI to generate initial PRD drafts. Design tools like Figma incorporate AI features for faster creation. In the CI phase, AI analyzes code beyond basic linting, generating detailed reports and suggesting improvements. The proliferation of AI touches every stage of development, fundamentally altering how teams operate.

AI-Assisted Developer Workflows

Developer workflows now focus on technical documentation, task breakdown, code implementation, and PR reviews. Current AI workflows range from using ChatGPT for direct coding assistance and GitHub Copilot for autocomplete, to adopting smarter editors like Cursor for larger refactoring tasks. Integration with tools like GitHub CLI and Atlassian’s MCP enables automation of common tasks.

A crucial warning emerges against “vibe coding,” where developers rely solely on AI suggestions without understanding the generated code. In large production codebases with multiple teams, shared code, and critical business metrics, this approach proves irresponsible. AI will not replace developers because the role requires understanding impact, coordinating across teams, and ensuring product quality.

The plan.md Workflow

An emerging workflow uses plan.md files as persistent scratchpads for AI collaboration. This approach helps retain context between chat sessions and promotes diligence over hasty implementation. Once a solid plan forms, the AI can break it down into well-defined, executable tasks.

A real-world example demonstrates an Expo repository codefix where Cursor collaboratively builds and refines a plan.md file, then breaks the solution into steps. The AI identified a silent type conversion causing null pointer exceptions that humans had missed. Step-by-step execution with an agent yields better results than direct, unguided chat interactions.

The workflow scales effectively. Tech leads use tools like Cursor for technical writing, generating proposals and breaking them into Jira ticket descriptions that adhere to specific templates. Automating Jira ticket creation by connecting Cursor with Atlassian’s MCP server eliminates common documentation bottlenecks.

Scaling AI for PR Reviews

A command-line workflow for PR reviews uses GitHub CLI commands like gh pr view and gh pr diff to pull PR information and diffs. An AI agent (Cursor) then reviews the PR, analyzing whether the diff matches the PR description, providing TL;DR summaries for each file, and generating checklists for manual code inspection.

This approach serves as a tool for diligence rather than a shortcut. Developers can hold deeper conversations with the AI about complex code sections and produce comprehensive review documents. The AI tracks data flow across files, checks edge cases, and ensures consistency with existing patterns in ways that augment human review capabilities.

Navigating the AI Transformation

Widespread uncertainty surrounds the AI shift. Those who have had negative experiences with AI tools should persist, as learning to use these tools resembles learning to drive a high-performance vehicle. The tools themselves are not at fault; users must learn to operate them effectively.

Software development jobs will not disappear but will shift toward higher-order tasks and increased sophistication. Developers will tackle more complex problems and potentially expand into product and design roles. The parallel to the accounting profession in the 1980s is instructive: while lower-level positions were eliminated, demand grew for highly skilled accountants who could leverage new tools.

Engineering jobs might become less engaging if leadership fails to guide teams effectively through this transition. Studying early computing history provides confidence, as historical parallels show that many current developments have precedents.

Technical Empowerment Through AI

Engineers are now “incredibly ridiculously empowered.” Developers should find inspiration, use AI to quickly complete side projects, and recognize that differentiation going forward comes from insight, creativity, taste, communication ability, and capacity to persuade others.

The slides themselves were created using React components and CSS, assembled with the very AI tools being discussed. This meta-demonstration illustrates the point: the barrier to creation has never been lower.

“We all just got Lambos, learn how to drive them and go have fun.” This transformation isn’t just about technical capabilities but about embracing a fundamentally new way of working that amplifies human creativity rather than replacing it.