What can you actually build?
A capability map — from a doc to a deployed app.
Three ambition levels. Pick the row that matches what you want.
Ship today, in chat
- Writing & email
- Analyze PDFs / CSVs
- Plans & itineraries
- One-page mini-apps
- Diagrams
Build this week
- A personal CLI
- Small websites
- Chrome extensions
- Slack / Discord bots
- Internal dashboards
Ship in production
- Customer-facing apps
- Agent workflows
- Document pipelines
- Embedded copilots
Which kind of tool, when?
| Situation | Reach for |
|---|---|
| In a project folder, touching code | A coding CLI (Cursor, Claude Code, Codex) |
| Long chores against a repo, in parallel | A cloud agent (Devin, Cowork, Copilot Workspace) |
| Needs to look visual | A UI generator (v0, Lovable, Bolt) |
| Phone / tablet / someone else's laptop | A chat app (ChatGPT, Claude.ai, Gemini) |
| Drop in a PDF or screenshot | A chat app |
| A non-developer needs to use it | A chat app |
| Persistent assistant with knowledge files | A chat app (Projects / GPTs / Gems) |
| Background daemon with persistent memory | A self-hosted agent (Hermes, OpenClaw) |
| Embedding into your own product | An LLM API (OpenAI / Anthropic / Google) |
If you have a terminal open in the right folder, a coding CLI wins most of the time. The other kinds cover the cases where you don't — or where the work isn't about code.