Build & Ship: It Works on Your Screen. That's Not the Same as Ready.

AI can build a working application from a description, and a lot of them are genuinely good. What AI does not do is ask the questions nobody asked it: who is allowed to see this record, what happens when the payment succeeds but the page never loads, which key ended up in the repository, and whether the thing you demoed to a customer can survive that customer's second user. This section is about the gap between working and ready.

TL;DR

  • The gap is not code quality. It is unasked questions: access, money, data, and what happens when something fails halfway.
  • A demo proves it can work. It does not prove it cannot break, and those are different claims.
  • The expensive failures are quiet: a logged-in user seeing someone else's record, a charge with nothing delivered, a key sitting in git history.
  • Bottom line: the builder did what it was asked. The gap is the question nobody thought to ask.

The uncomfortable part

Here is the moment that catches people. Someone asks a fair question about the app you built: "can a customer see another customer's data?" And you realize you do not actually know. Not because you were careless. Because you never opened that particular door, and the tool that built it never mentioned there was a door.

That feeling is not a verdict on you. Building software by describing it is a genuine skill and it produces genuinely working products. But it changes where the risk sits: from can I make this to do I know what I made. And the second question does not answer itself, no matter how well the first one went.

What's inside this section

Ship Your App is the readiness half, written for people who built something real without a traditional engineering background: client versus server checks, row-level security, the money path from checkout to fulfillment, logged-in versus allowed, keys and rotation, test mode versus live, data model sanity, knowing what personal data you now hold, accepting or reverting an AI change, and being able to explain your own app to someone who asks.

Agents & AI Engineering is the deeper half: multi-agent orchestration, prompt evaluation and regression testing, token economics, code review discipline, prompt injection, and agent security. This is where the work stops being about one app and starts being about systems that act on their own.

How to use this section

  1. Start with access. Being signed in is not the same as being allowed. This is the single most common and most damaging gap.
  2. Follow the money end to end. Not the happy path: the refund, the failed card, the duplicate notification, the payment that succeeds while the page dies.
  3. Find out what you are holding. You cannot protect or delete data you do not know you collected.
  4. Then make it explainable. If you cannot describe what your app does with data, you cannot answer a customer, a partner, or a buyer.

The honest line: nothing here is a penetration test, a security certification, or a legal review, and none of it makes an app "safe" as a finished state. These are read-off checks. You look at your own screen and answer honestly, and where something is short the result tells you what to look at next -- and says plainly when nothing is. That is a genuinely useful thing and it is not the same as an audit.

FAQ

What usually breaks first in an AI-built app?

Access control. The app checks whether you are logged in but not whether this particular record belongs to you, so any signed-in user can reach another user's data by changing a number in the address bar. It is invisible in a demo because a demo only ever has one user.

Is vibe-coded software inherently unsafe?

No. It is software that was built quickly by describing intent, so the risk concentrates in whatever the description did not mention. The code is often fine. The gaps are in access rules, failure paths, and money handling, the parts nobody thinks to specify.

Why do payments need more than a success page?

Because the browser can close, the network can drop, and the notification can arrive twice. A payment path that only works when the customer stays on the page will eventually take money and deliver nothing, and that failure looks like fraud to the person it happens to.

I found a key in my repository. What now?

Treat it as exposed and rotate it, rather than deleting the file and hoping. Removing a key from the current version does not remove it from the history, and rotation is the only step that actually revokes access. Rotate first, tidy the history second.

Do I need to know what data my app stores?

Yes, and it is usually more than you chose. Logs, analytics, error reports, and AI prompts all accumulate copies of things you never decided to keep. Knowing what you hold is the prerequisite for every other decision about it, including whether you should be holding it at all.

How do I check any of this without being an engineer?

By reading your own screens and answering specific questions honestly: open two accounts and try to reach one from the other, follow a real payment through to delivery, look at what your database actually stores. The checks are observational, not technical.

Find out what your app does when you're not watching

The Vibe-Coded App Hardening Kit ($79) walks the gaps that survive a good demo. Or take the whole readiness slate with Past the Prototype ($459), ten drills covering access, money, data, and change control. One-time, instant download, yours to keep.

Harden your app ($79)

Latest in this section

Tools for this →

Fable 5.2 Vibe Coding: From Demo to Product People Trust

Fable 5.2 vibe coding gets you a working demo faster than ever. What the model's own docs warn about, and the five gates between that demo and a real product.

Nov 11, 2026 · 12 min read

Client-Side vs Server-Side Validation: Which of Your App's Checks Are Real

Client side vs server side validation comes down to control: your user can change the browser, so a check enforces nothing until your server decides it too.

Nov 9, 2026 · 9 min read

How Easily Can Someone Get Around a Front-End Check?

You can bypass client side validation with tools built into the browser: edit a field's value, turn off JavaScript, or send the request yourself.

Nov 9, 2026 · 6 min read

"Never Trust the Client" — What That Actually Means for Your App

Never trust the client means re-verifying anything that matters — money, someone else's data, access — on your server, using a value it controls itself.

Nov 9, 2026 · 6 min read

Why Your App Should Never Take the Price From the Page

Client side price manipulation happens when checkout trusts a browser-sent price instead of looking it up by product ID on the server, every time.

Nov 9, 2026 · 6 min read

The Server-Side Check That's Still a Browser Decision

A quiet entry among server side validation mistakes: the check runs on your server but trusts the number the browser sent instead of looking it up itself.

Nov 9, 2026 · 6 min read

Moving a Key Isn't Fixing It: When You Actually Have to Rotate

Here's how to rotate an API key: issue a new value and turn off the old one. Moving or renaming a leaked key doesn't fix it — it's still live and dangerous.

Nov 2, 2026 · 9 min read

You Deleted the Key. It's Still in the Git History.

An API key committed to git stays in every past snapshot even after you delete the line — reachable by anyone who cloned the project before your cleanup.

Nov 2, 2026 · 7 min read

Publishable vs Secret Keys: They Look Identical and They Aren't

Publishable vs secret key: they're built for two different jobs and often look identical — mixing them up can break features or expose what shouldn't leak.

Nov 2, 2026 · 7 min read

The .env File Did Not Make It Private

Is my .env file secure? The name tells you it holds settings, not that it's stayed private — a key that's ever been reachable still needs to be rotated.

Nov 2, 2026 · 7 min read

Which Keys to Rotate First When You're Not Sure What Got Out

Which keys to rotate first when time is short: unknown-location keys, then currently-reachable secrets, then anything past-exposed and never rotated.

Nov 2, 2026 · 7 min read

AI Oversight Without Reviewing Everything

AI oversight that scales: cheap first-line signals, escalation to costly review, a human on every irreversible call. The three tiers and how to build them.

Oct 31, 2026 · 15 min read

Reward Hacking: When AI Passes the Test, Not the Task

Reward hacking is when an AI exploits its success measure, not the task: editing the tests, gaming the grader. What it is, what it isn't, and how to catch it.

Oct 31, 2026 · 11 min read

Why an AI's Explanation Isn't Proof of Its Reasoning

AI explanation faithfulness: a convincing explanation is not proof of its reasoning. What counts as evidence, what Anthropic found, and how to check it.

Oct 31, 2026 · 11 min read

Open Models, Closed APIs, and the Right to Inspect AI

To inspect AI models from the inside you need the weights, which a hosted API never gives you. What open weights change, what they don't, and what to ask for.

Oct 31, 2026 · 10 min read

The AI Inspectability Checklist: 10 Vendor Questions

Ten AI vendor explainability questions with the answer that counts as evidence for each. A buyer's checklist, not a certification. Free one-page PDF.

Oct 31, 2026 · 10 min read

AI Interpretability: The Black Box May Be Opening

AI interpretability is moving: neurons described, pathways traced, behavior changed on purpose. What it proves, what it doesn't, and what you can check now.

Oct 30, 2026 · 15 min read

Neural Networks Are Already Clean: What the Essay Says

Neural networks are already clean, says Goodfire's Nick Cammarata. The claims, the numbers kept apart, the caveats, and what the companion paper must show.

Oct 30, 2026 · 11 min read

The Cat and Car Neuron: Solved, or Still Open?

The cat and car neuron is AI's most famous polysemantic neuron. Goodfire reads it as a V detector. Liv Gorton's tests disagree. Both accounts, side by side.

Oct 30, 2026 · 10 min read

Mechanistic Interpretability, Explained in Plain Terms

Mechanistic interpretability without the jargon: neuron, pole, feature, polysemanticity, superposition, circuit, SAE, probe, steering, attribution graph.

Oct 30, 2026 · 10 min read

Does Interpretability Make AI Safe? The Missing Steps

Is AI interpretability enough for safety? No. Legibility is one input. Still missing: prediction, durable fixes, monitoring, and a human who can stop it.

Oct 30, 2026 · 10 min read

How to Estimate and Budget Your LLM Token Costs

LLM token cost equals calls times average tokens per call times your provider's current input and output rates, priced separately and always changing.

Oct 23, 2026 · 7 min read

How Much Does an LLM API Cost? (Input vs Output, Explained)

LLM API cost splits into two rates: a lower one for the text you send in and a higher one for what the model generates back. Both keep changing.

Oct 23, 2026 · 6 min read

What Is a Token in AI? (And Why It Costs You Money)

What is a token in AI? A word or a piece of one: the unit an LLM reads and writes in, and the unit nearly every provider bills you by, per call.

Oct 23, 2026 · 6 min read