AI inside your app

Adding an AI feature is one API call. Making it reliable, safe and affordable is the rest of the work. Model replies are text that can break your parser, stop halfway or repeat your secret instructions. Tools you give the model act with whatever power you hand them. This trail covers the problems that appear once real users start talking to your AI.

Scenarios: 7
Total time: 27 minutes
Free to try: The first one

7 scenarios on this trail

Next up · Free · 4 minThe JSON that wasn'tYour recipe app asks the AI for a shopping list as JSON. It worked all week. Today one in five requests crashes.
Stop 2 · Pro · 3 minThe answer that stopped mid-sentenceShort answers in your study app are fine. Long ones end mid-sentence, like this: 'The third cause of the war was the'
Stop 3 · Pro · 4 minThe chatbot that shared its instructionsA user posted a screenshot of your support bot's hidden instructions, including the staff-only discount code and an API key.
Stop 4 · Pro · 4 minThe assistant that cancelled everythingYour booking app's AI assistant can cancel bookings. A user typed 'cancel all the bookings for Friday'. It cancelled 63, most of them other people's.
Stop 5 · Pro · 4 minAnother customer's documentYour 'chat with your documents' app answered a question by quoting a contract. The user has never uploaded a contract. Another customer has.
Stop 6 · Pro · 4 minThe refund policy it made upA customer wants the 60-day refund your support bot promised them. Your policy is 14 days. They have a screenshot.
Stop 7 · Pro · 4 minThe chat that got expensiveYour AI tutor costs about a cent per message for new users. One student's 400-message chat now costs 30 cents per message, and replies take ages.

Other trails

Who may see what
Sign-in, permissions, and the leaks AI-built apps ship most often.
The life of a click
Follow one button press from the browser to the database and back.
Money and secrets
Keys, payments, webhooks, and reading what your AI changed.
Data you can lose
Agents that reset production, backups that are a day old, rollbacks that break.
Bills that explode
Open AI endpoints, bot traffic, fake signups and functions that loop forever.
Working with your AI agent
Dangerous commands, fix loops, deleted tests, prompt injection and invented packages.
Launch day
Emails that never arrive, sign-in errors, timeouts, stale prices and connection limits.
Keeping it running
Silent errors, empty AI credits, expired domains, leaky logs, risky updates and retired models.
Accounts and privacy
Login codes, reset links, email changes, account deletion and customer data sent to AI.
Don't trust the browser
Prices, webhooks, uploads, form limits and coupons: what visitors can change, and what your server must decide.
Teams and sharing
Removed members who keep access, invites anyone can use, exports that leak other customers, guessable share links.
Into the app stores
Keys inside phone apps, store rejections, review sign-ins, old versions, deep links and over-the-air updates.
Getting paid
Success pages that give Pro away, failing webhooks, test-mode customers, double subscriptions, failed renewals and card testing.