Skip to content

Options

The sensible ways to hand over something you built with AI. Each page leads with what it lets you (and your recipient) do, says plainly who can get in, and puts numbers on what it costs each side to start. Not sure which? → Which one should I pick?

Pick by what you want to happen on the other end:

  • GitHub repository — one home for your code or files, every version saved, others can build on it. For when people will work with your thing, not just look at it.
  • Deploy a website — put a working thing online and hand over one link; the recipient clicks and it just runs. For handing over a finished thing to people who'll never open a code tool. Splits by host: Vercel, Netlify, Cloudflare Pages, Fly.io (real backends), or a gated site only named people can open.
  • Claude Artifact — publish a chat-built interactive widget behind one link, no host to manage. The lightest way to share a single clickable thing.
  • Lovable — describe an app in chat, hit Publish, send a link; the recipient opens a real multi-page web app — logins, saved data, a backend — with nothing to install or deploy. For a working app, not a single widget.
  • Gemini Gem or Canvas — hand someone a custom assistant with your instructions and Drive files baked in, or an interactive doc Gemini built with you, as a link they open. For staying inside Google's world and reaching your team off their existing Drive.
  • Shared AI project — one shared space (Claude or ChatGPT) where a group works off the same files, instructions, and chats. For when the thing isn't a file to hand over but an ongoing collaboration inside an AI tool.
  • Reconstruction prompt — hand over a recipe, not the dish: one block of text the recipient's own agent rebuilds into a close cousin. For spreading a pattern past walls a link can't cross.

Share documents & knowledge

When the thing is words and structure — something people read, comment on, or keep working in — not code to build on or a finished page to click:

  • Google Doc — when it's words a group works over: a guide, your notes, a spec. One link opens it on any phone, no account, no install. They read, comment, or edit the same link.
  • Notion page or workspace — a living knowledge space with structure: pages, tables, boards, a small database, editable in place. For a tracker or wiki that keeps changing, that you can let others edit too.
  • A browsable, searchable docs site — turn a folder of notes into a real website: a left-hand menu, a search box, a clean reading layout. For when there's enough to navigate — many pages a reader moves through by topic.
  • Google Drive folder — hand over a whole bundle behind one link: Docs, Sheets, Slides, PDFs, mixed types side by side, sub-folders and all. For "here's everything for this," not a single file.

Share or collect data

When the thing is rows and columns a few people keep — or a stream of answers you need from people — and you want your agent able to read and write it:

  • Google Sheet — share a list, a tracker, a set of figures people read, filter, and add to at once; because your agent reads and writes it by code, the same sheet doubles as a tiny shared database. For structured data, not flowing prose.
  • Form to collect input — the one option that runs the other way: hand people a link to give you answers — sign-ups, survey responses, intake — each landing as a row your agent can act on. For gathering structured input from a crowd.

Models, data & APIs

When the thing isn't a document or a page but a machine-readable artifact — a model, a dataset, or a capability other people's code calls over the network:

  • Hugging Face Hub — put a model's weights or a dataset's files at a huggingface.co/... address, pulled into someone's ML code in one line, with a card and version history. Public, private, or gated (listed, but you approve each downloader). For handing ML artifacts to the research world, not ordinary project files.
  • Live API endpoint — let other people's code call your thing over the network: they send a request, get a result, and never see or run the code behind it. For when the value is a computed answer their program uses — a score, a lookup, a forecast — not a page a human reads.

Give an agent new powers

When the thing you built isn't a file to read or a page to click, but a capability — something the recipient's own agent should be able to do:

  • Claude skill — hand their Claude a new ability: a routine it runs on demand across all their work. For spreading a repeatable workflow (a house style, a grant procedure), not a one-off file.
  • MCP server — give their agent a live plug into a system you run — a database, an API, an internal tool — so it can query or act mid-task, without your credentials. For when the value is live, not a fixed set of files.
  • One-command tool — turn a script into a tool anyone runs by name: one word, it fetches itself and runs, and your fixes reach everyone on next run. For when people will run the same thing over and over.
  • Raycast extension — a small tool teammates fire from anywhere on their Mac with a couple of keystrokes, installed in one click from the public Store or a private one for your org. For Mac power-users who keep the launcher one keystroke away. Mac-only.
  • GitHub Action — a reusable step that runs by itself inside someone else's project on every change — they paste one reference once, then your step fires on each push or on a schedule, and your fixes reach everyone. For a check that should run automatically across many people's projects.

Run it without installing

When the thing is code that has to actually run — a notebook, a small app — but you don't want the recipient installing anything: it lives in the browser, theirs to open and run.

  • Google Colab — a Python notebook that runs on Google's free machines, opened from one link with a Google account. For sharing runnable analysis — a model, a data walk-through — that the recipient can re-run and tweak, no setup.
  • Replit — a full coding workspace in the browser: edit, run, and (optionally) host a live app, all in one tab. For a small app or script you want someone to run and change without ever leaving the browser.

Before you send anything, two cross-cutting reads: Who can see it? and Can you trust the company?.