UX-minded React developer to shape a web IDE for vibe coders building internal business apps

We're building a platform where companies create and run their own internal business tools: inventory trackers, approval workflows, job and client records, request forms, team dashboards. In other words, the CRUD apps every business runs on, which today live in spreadsheets, Airtable or expensive custom software. Each app is a small React project. The platform handles everything around it: hosting, sign-in, who can see and edit what, a shared database with full history, and one-click deploys. The people we want building these apps are not developers. They're operations managers, office leads and founders who know exactly what their team needs and are happy to describe it to an AI. They're vibe coders: they describe what they want, the AI agent writes the code, and they judge the result by whether the app works, not by reading the code. For them we've built a browser-based IDE: an AI agent chat, a code editor, a live preview of the app and a deploy button. It works, but it was built by an engineer for engineers. We need someone to take it over and turn it into a tool a non-technical person can trust. THE UX CHALLENGE This isn't a typical front-end job. Building an IDE for people who don't read code raises design questions that few products have answered well yet: - How much code should a vibe coder see? When is the editor helpful, and when is it intimidating noise? - How does someone confirm the AI did what they asked when they can't review a diff? What does "here's what changed" look like for a person who thinks in screens, fields and buttons, not files? - Internal apps hold real company data. The schema can only grow, never shrink, and previews read and write the real database. How do you make it clear which actions are safe experiments and which are permanent, without scaring people off? - Deploying puts an app in front of real coworkers. How do you make "save", "preview" and "go live" feel distinct and confidence-inspiring? - When a build fails or the agent gets stuck, what does a non-developer need to see to keep going, or to know the AI will handle it? - How do you show what an AI agent is doing during a multi-step task, and ask for approval before risky steps, in plain language? The goal: a non-technical person builds a working CRUD app (a list, a form, records with statuses, the right people seeing the right data) and ships it to their team without ever feeling lost. WHAT YOU'D DO - Own the IDE's user experience end to end, guided by the questions above. - Find where vibe coders get confused, stuck or scared, and fix it. - Simplify the core loop: describe an app → watch the agent build it → check the result → preview → deploy. - Turn errors, approvals and agent progress into language a business user understands. - Use LLM coding tools (Claude Code, Cursor, Codex, etc.) heavily and well to move fast in a well-documented codebase without lowering its quality. - Be a beta tester. Build real internal apps of your own with the IDE, and report what's broken, missing or confusing. Your experience as a builder is half the value of this role. TECH YOU'LL WORK WITH - React 19, TypeScript, Vite - Cloudflare Workers and Durable Objects (the IDE's backend: sessions, working copies, the AI agent) - CodeMirror 6 editor - An AI agent chat (OpenRouter models) that edits the app's files and runs platform commands on the user's behalf - A strict codebase culture: tests, type checks and lint on every change, and written architecture decisions and agent guides (AGENTS.md) you're expected to read and follow You'll work in the IDE's own repository against a hosted instance of the platform; you won't need to run the platform yourself. You don't need to know all of this on day one, but you do need to be comfortable picking up an unfamiliar, opinionated codebase, ideally with an AI assistant helping you navigate it. YOU'RE A GREAT FIT IF YOU - Have a strong UX eye and can point to products you've made simpler and clearer for non-experts, not just prettier - Have shipped real React/TypeScript front ends - Vibe code yourself, use LLM coding tools daily and can explain how you keep their output correct and consistent - Understand business software: you've built, used or fought with internal tools, admin panels or CRUD apps - Write clear bug reports and product feedback - Are comfortable working async with a solo founder, with a lot of ownership NICE TO HAVE - Experience designing AI-agent or chat-driven interfaces (showing progress, approvals, changes, undo) - Experience with low-code/no-code builders (Retool, Airtable, Bubble, Glide, Lovable, Bolt and similar), IDEs or developer tools - Cloudflare Workers experience THE ENGAGEMENT: TWO WEEKS, $1,000 FIXED PRICE, ABOUT 15–20 HOURS PER WEEK We start with a two-week project to see how we work together. If it goes well, it turns into an ongoing role owning the IDE. Milestone 1: First-time vibe coder audit (days 1–3, $250) Sign up as a brand-new user and build a small real internal app entirely in the IDE, for example a job tracker where staff log jobs, set a status and a manager approves them. Deliver: - A screen recording or annotated screenshots of the whole attempt - The 10 biggest points of confusion or friction, ranked by how likely each is to make a non-technical user give up, each with a one-line proposed fix - The app you built, deployed, or a note on exactly where you got stuck You'll never be this new to the product again, so this first impression is the most valuable feedback you can give. Milestone 2: Make the core loop clear (days 4–9, $450) Together we'll pick the top three problems from your audit on the main path (describe an app, watch the agent build it, check the result, preview, deploy), and you fix them. Deliver merged changes with tests that pass the repo's checks, plus a short before-and-after video. Milestone 3: A safety net, "undo" for vibe coders (days 10–14, $300) The platform already keeps every saved version of an app and can restore an earlier one, but the IDE has no screen for it. Design and build a simple History view: see past saves with their messages and restore one. For someone who can't read the code the AI wrote, being able to go back is what makes experimenting feel safe. Deliver the feature, tested, plus a short closing write-up covering: - What you'd prioritize next - What it was like building your app as a beta tester - Any gaps you found that can't be fixed from the IDE side At the end of the two weeks, we'll both decide whether to continue and on what terms. TO APPLY, PLEASE ANSWER 1. Link to one or two interfaces you designed or built, and tell me what you did to make them clearer for non-expert users. 2. Describe how you vibe code or use LLMs in your front-end workflow. What do you hand to the AI, and what do you always check yourself? 3. Pick one question from "The UX challenge" above and sketch how you'd approach it, in a few sentences. 4. What internal tool would you build first if you could describe it to an AI and ship it to your team the same day? Please start your proposal with the words "CRUD apps" so I know you read the whole post.