The tools are good enough now. What's left is choosing what to build and actually finishing it.
You no longer type an app from scratch. Shape the plan with ChatGPT, hand implementation to Cursor or Claude, and step in where it gets stuck — the workflow people call vibe coding. What you want to build now matters far more than your degree or job title.
First projects still stall, usually for two reasons: the scope is too big, and nobody is watching. Cut it down to one screen and one feature, and put what you make somewhere people actually see it — that combination is what keeps a project alive.
Below are real projects from makers who shipped first. Most started at a similar size, which makes them a good gauge of how finished something needs to be before you publish it. If you get stuck, you can recruit for the role you're missing.
Self-taught, non-CS
Instead of learning syntax first, learn backwards from the thing you want to exist.
Planners and designers
Building the prototype yourself validates the idea far faster than describing it.
Working developers
Toy projects are where you try the stack and tools your day job won't let you touch.
Students and job seekers
Leave behind one project real users try, instead of another tutorial result.
1. Narrow to one sentence
Who uses it, for what, and when. If you can't write that yet, the scope is still too big.
2. Build one screen
Make the single most important screen work. Login and settings can come later.
3. Publish as an idea first
Post it before it's done and use the feedback to correct course while building.
4. Recruit where you're stuck
Blocked on design or backend? Recruit just that role.
AITraining.jobs lists the open AI training jobs from the platforms that pay people to train, evaluate and label data for AI (Mercor, micro1, Outlier, DataAnnotation, Braintrust, Alignerr, Prolific and about 50 more). It pulls openings daily from 25 platforms' own job feeds (about 3,400 open roles in October 2026), shows the hourly pay each role advertises, and covers every kind of work: RLHF and evaluation, coding, math, law, medicine and finance expert work, voice and audio, data annotation and multilingual roles. Platform reviews, pay comparisons and country pages help people see which platforms are legitimate and what they pay. Free, no signup needed.
Novastart brings apps and files onto an infinite canvas where people and AI agents can work together. Novastart is an infinite canvas whiteboard OS. Open Chrome, ChatGPT, Claude, Word, Excel and your files side by side on one infinite whiteboard, share canvases with your team, and let AI agents work alongside you with their own cursors and keyboards. Platforms: Windows, Intel Mac, Linux. X: https://x.com/NovastartOS
MangaTranslate는 해외 만화를 읽는 독자와 웹툰 현지화 작업을 검토하는 제작팀을 위한 AI 만화 번역 도구입니다. 단일 또는 여러 장의 만화 이미지와 PDF를 100개 이상의 언어로 번역하고, 결과를 온라인에서 읽으며 직접 검토하고 편집할 수 있습니다.
A Gomoku game. PVE battles (6 difficulty levels) PVP battles (within 5km proximity)
When you copy and paste AI-generated text, does it often appear in an undesirable format on each SNS platform? It might not seem like a big deal, but it's tedious to manually adjust each one (e.g., removing Markdown syntax). With the 'Write what you saw' service, simply input your text, select an SNS platform, and your content will be automatically formatted to best suit that specific platform.
Design and decorate your own home, unlimited and free.
우리집 약통에 몇개의 약이 있나요? 다 사용하시나요? 복용약, 상비약등을 관리하고, 특히, 부모님의 복욕약도 관리를 원하시는 분들이면....
SkipCalls is an AI phone receptionist for small businesses that need every call handled without keeping someone tied to the phone. It answers inbound calls, follows business-specific instructions, captures messages and caller details, and can help with appointment workflows. The product is built around a practical idea: a missed call is not merely a notification; it is an unfinished customer request that should become a visible, reviewable workflow. The reliability work is as important as the conversational layer. Calls need durable outcomes, clear ownership, and enough context for a human to follow up when automation reaches an exception. SkipCalls therefore focuses on useful handoffs and call records instead of treating a plausible AI response as the end of the job. Teams can configure their receptionist around their real operating process while keeping a human path for unusual or sensitive cases. It is designed for local service businesses, clinics, property teams, and other small organizations where answering promptly affects revenue and customer trust.
한국의 굴착기와 어테치먼트를 소개하고, 작업일보 및 기타 굴삭기 관련 자료를 관리합니다.
Yes — many people start by describing what they want to ChatGPT or Claude, then running and fixing the result in an editor like Cursor. Deployment and data storage are the usual first walls, so start with a single-screen scope.
Explaining your intent to an AI and letting it write the implementation, while you stay responsible for direction and verification. Whether the result works and is worth using is still a human judgment.
A common combination is ChatGPT or Claude for planning and problem definition, plus an editor-integrated tool like Cursor or Claude Code for implementation. Keeping the scope small matters more than the tool choice.
Publish it as an idea for feedback, or recruit for the role you're blocked on — design, backend, whatever it is. Unfinished is not a reason to stay hidden.
On Ppomnae, for free. One line is enough — AI drafts the description and a cover image for you.