Components for AI-native apps.

Add a screen where people can watch the assistant use a website for them: its pointer moves, clicks and types while a list shows each step, with Pause, Take over and an approval before anything that spends money. Add a mode to teach it a task by doing it once, and a recording people can replay and scrub through.

Add a thinking trace to my app’s assistant: while it works, show a “Thinking…” panel that lists each step with a spinner, a tick when it is done and how long it took. Let people open it to read the reasoning, the searches it ran and any code it wrote, and fold it into “Thought for 6s” when it finishes.

Show the tools my assistant uses while it answers as small chips: a spinner while each one runs, its name and how long it took when done, red with the error when it fails, and a click to open what went in and what came back. Offer a timeline view that folds into “Used 5 tools”.

Add a task list that shows what the agent is working on: one row per task with its status (queued, running with a progress ring, failed with a Retry button, done with a tick), a short count such as “12 invoices”, and a chevron that opens the task’s sub-steps.

Add a feature plan card: a checklist of the features being built with a progress bar (“3 of 5 built”). Each feature shows to do, in progress or done, opens to a one-line detail, and carries a small “Test passed” or “Test failed” badge.

Let people keep sending requests while the assistant is working: show what it is working on with its progress, and put new requests in a queue above the prompt box where they can be reordered, edited, removed or sent next, with a button to pause the queue.

Add an image generator: a prompt box with style and shape options and a Generate button that makes four images at a time, each showing its progress as it develops. On a finished image people can make variations, upscale it, save it as a favourite or download it; earlier batches stay below.

Make the assistant’s answers stream in word by word behind a blinking caret, with numbered citations that show the source’s title, site and a snippet on hover, a row of actions (copy, retry, thumbs up and down) and two or three follow-up questions people can click to ask next.

When the assistant answers from our documents, show the passages it used as cards: each with the source and page, a relevance score, the matching words highlighted, and buttons to pin it or leave it out, with the answer citing each one by number.

Before the assistant does something important, show an approval card: a few questions one at a time (one choice, several choices or a short answer) with Skip, Continue and a “1 of 3” counter. For risky actions, show a card that says exactly what will happen, with Approve, Edit and Deny.

Add a recommendation card: the option the assistant recommends with a confidence gauge (for example “82% · High”), the reasons with their sources, the main risk, what would change its mind, the alternatives it considered with their own confidence, and a button to act on it.

Add a voice assistant to my app: a glowing orb people tap to talk to, which shows when it is listening, thinking and speaking, with live captions, a mute button, tap to interrupt while it speaks, and a small call bar with a timer and an end button.

Let people comment on the preview of my app: a Comment button, then a click anywhere drops a pin with the person’s avatar and opens a thread where others can reply. Threads can be resolved (with Undo) and hidden, show when there is a new reply, and are also listed in a side panel with Open and Resolved tabs.

Add a prompt box to my app: it grows as people type, “@” opens a list of files and data sources to attach as chips, “/” opens a list of commands, and it has buttons to attach a file, choose a model, dictate by voice and send — all usable from the keyboard.

Add a model picker beside the prompt box: a button showing the current model and effort that opens a list of models, each with a short description, speed and cost indicators and a badge for premium ones, plus an effort control (Low, Medium, High, Max) that explains what each level costs.

Add voice typing to my text fields: a microphone button that turns the field into a live waveform with a timer, shows the words as people speak (unfinished words in grey), and has buttons to keep or cancel what was said.

Let people select text in my editor and get an AI toolbar right next to it — Improve, Shorter, Warmer, Fix spelling and Translate — then show the rewrite in place with the old words struck through and the new ones in green, and buttons to accept or discard it.

Add a command palette that opens with ⌘K: one search box that finds pages, records and actions as you type (forgiving typos), groups the results with recent items first, highlights the matching letters, runs the chosen one with Enter, and shows a friendly empty state with a suggestion when nothing matches.

When the assistant proposes changes to a table, show them before applying: each changed cell with the old value struck through and the new one in green, a note on any row it chose not to change, and buttons to accept or reject each cell, each row or everything, then Apply.

Add a row of stat cards to my dashboard: each shows a number, its change since last week in green or red, and a small chart people can hover to read each day’s value. Below them, add an insight card that pages through what the numbers mean, each with a suggested next step.

Add a customers table: searchable, with sortable columns (name, tags, monthly volume, last order, and relationship strength shown as four bars), checkboxes to select rows, and a bar with bulk actions that appears when rows are selected. On a phone the first column stays in place while the rest scrolls.

Add a live orders table with status chips above it (All, New, In progress, Shipped, Issue) that each show a count and filter the rows when clicked. New orders appear at the top and rows move smoothly when their status changes; add a search box, a Pause button for the live updates and an option to group the rows by status.

Show the code changes the agent made as a diff people can review: file by file, removed lines in red and added lines in green with the changed words highlighted, unchanged code folded away, a Viewed checkbox per file, comments on any changed line, and a switch between unified and side-by-side views.

Add a live log viewer to my app’s admin: lines stream in with time, level and source; toggles show or hide each level with a count; a search box highlights matches; it follows the newest line until I scroll up (then shows a “new lines” button); and I can pause it, open a line’s details and copy lines.

Add an automation builder: a flow drawn on a dotted canvas that starts with a trigger, branches on yes/no conditions and runs actions, with a + on every connector to add a step from a menu, steps that can be renamed or deleted, pan and zoom, and a Test run button that highlights each step as it runs.

Add a kanban board with a column for each stage, cards people can drag between columns with a mouse or a finger or move with the keyboard (Space to pick up, arrows to move, Space to drop), a count and a limit on each column, and a quick “Add card” at the bottom of each one.

Show who else is in a document right now: a stack of avatars with an active or idle dot, each person’s cursor with their name in their own colour, what they have selected or are typing, a note when someone joins or leaves, and a way to follow someone by clicking their avatar.

Add a version history: a timeline of saved versions grouped by day, with who made each one and the current one marked. Selecting a version shows what changed (or a before/after comparison of the page), and Restore asks for confirmation, keeps the current version in the history and can be undone.

Add a sidebar to my app: a workspace switcher at the top, the main pages with icons and unread counts, sections that fold, a highlight that glides to the row under the pointer, the signed-in person at the bottom, and a button that collapses it to an icon rail with tooltips.

Add a setup checklist for new users: a progress ring with “2 of 5 done”, the steps in order with how long each takes, the next one open with a button that does it and “Skip for now”, a celebration when everything is done, and a way to hide the guide and bring it back. Also offer it as a small launcher in the corner of the screen.

Add a settings page with sections (shop details, shipping, notifications) that marks every field I change, shows a save bar with the number of unsaved changes and Save / Discard buttons, checks the fields before saving and jumps to the first problem, confirms when it is saved, and saves with Ctrl+S or ⌘S too.

Turn the application form into a step-by-step wizard: a step bar showing where I am and which steps are done, each step checked before moving on with clear messages, Back without losing anything, a final review page with Edit links for each step, and a confirmation after sending.

After someone uploads a spreadsheet, show a step to match its columns to my fields: each column with a few sample values, the suggested field and how confident the match is, an easy way to change it or skip the column, warnings for missing required fields or two columns on the same field, and an Import button that shows progress and a summary.

Add file uploads: a drop area people can drag files onto, click to browse, or paste images into, a preview and a progress bar for each file, a clear message when a file is too big or the wrong type, a Retry button when an upload fails, and attachments in the message box that must finish uploading before the message can be sent.

Add a permission dialog that appears before another app connects or the assistant does something on my behalf: it lists exactly what is requested in plain words, lets me switch off the optional parts, shows a preview of the action (such as the email and who gets it), offers “Allow once” or “Always allow”, and has a “Remember this” option.

Add a booking picker: a strip of the next two weeks (or a month calendar) to choose a day, the free times grouped into morning and afternoon with full ones greyed out, a choice of session length, the time zone, and a summary with a Confirm button that turns into a confirmation with “Add to calendar”.

Add a pricing section: three plans side by side with a Monthly / Yearly switch that shows what yearly saves, the most popular plan highlighted, a Choose button on each that carries the plan and its price to a checkout bar, and a comparison table of what each plan includes.

Add toast notifications: they stack in a corner and fan out on hover, close by themselves after a few seconds (pausing while hovered), can be swiped away, and come as success, error with Retry, loading that turns into success, and a notice with Undo.

Add a notifications bell to the top bar with a red unread count. It opens an inbox with tabs for All, Unread and Mentions, notifications grouped by Today, Yesterday and Earlier, buttons to act on each one right there (approve, reply), a dot to mark one read or unread, “Mark all as read”, and new notifications sliding in as they arrive.

Design the empty states for my lists: a friendly first-visit screen with buttons to add the first item or try sample data, a “no results” message that names the search and offers to clear the filters, an error screen with Try again and details, and an offline banner that keeps the last saved data visible and retries by itself.