Build a calculator web app and refactor it with Devin
Rebuilding a carbon emissions calculator with Devin vs Copilot, where agentic AI speeds delivery and where human engineering remains essential. Task: The author rebuilt a carbon emissions calculator web app using Devin, transforming an Excel spreadsheet into a TypeScript/React application. The workflow involved attaching the spreadsheet, defining models, and iterating on UI and tests via PRs. Human engineers were essential for architectural decisions, refactoring bloated code, and validating logic, as Devin struggled with Electron apps and required connected repositories for optimal performance.; tools: Devin. Human role: The original post does not say how a person stayed involved.; limitations: Devin is expensive compared to other AI tools and requires careful context management to avoid 'context rot'. It performs poorly with desktop technologies like Electron and needs a connected repository. Human engineering skills are mandatory to validate security, maintainability, and logic, as the AI cannot fully replace senior oversight for production-grade software..
Tools
Published on Agent Work Atlas:
Overview
Rebuilding a carbon emissions calculator with Devin vs Copilot, where agentic AI speeds delivery and where human engineering remains essential.
Workflow
The author rebuilt a carbon emissions calculator web app using Devin, transforming an Excel spreadsheet into a TypeScript/React application. The workflow involved attaching the spreadsheet, defining models, and iterating on UI and tests via PRs. Human engineers were essential for architectural decisions, refactoring bloated code, and validating logic, as Devin struggled with Electron apps and required connected repositories for optimal performance.
Reported result
The project produced a production-ready application with unit and Playwright tests in seven days, estimated to have taken 8-10 weeks for a team of three. The build cost approximately $350 (155 ACUs). However, the author noted significant technical debt, code bloat, and architectural drift that required manual intervention and refactoring to resolve.
Human role
The original post does not say how a person stayed involved.
Limitations
Devin is expensive compared to other AI tools and requires careful context management to avoid 'context rot'. It performs poorly with desktop technologies like Electron and needs a connected repository. Human engineering skills are mandatory to validate security, maintainability, and logic, as the AI cannot fully replace senior oversight for production-grade software.
Try it
Act as a senior software engineer guiding Devin to rebuild a legacy calculation tool. [replaceable inputs: source spreadsheet file, target tech stack (e.g., TypeScript/React), business logic requirements]. First, analyze the provided spreadsheet to identify core calculation models. Instruct Devin to scaffold a new repository with the chosen stack, focusing only on data models initially. Review the generated code for bloat and unnecessary dependencies. Next, ask Devin to implement the UI components and integrate live data sources if applicable. Generate unit and end-to-end tests for all calculators. Perform a dry run on one specific calculator module to verify output accuracy against the original spreadsheet. Only proceed to full implementation after confirming the sample's correctness and architectural soundness. Ensure all PRs are reviewed for security and maintainability before merging.
Source reports are not independently verified. Publication date is the catalog publication date, not the original article date.