how-ai-thinks

12 articles · newest first

Build

You typed a sentence into Lovable and an app appeared. Here's where your code, data and bill actually live.

Lovable describes itself as a platform that builds and deploys web apps from plain English. Its own docs say where the app is hosted, what the built-in backend is, where your code sits and how you pay. A plain-language tour, from the docs.

Aliteq Team · Oct 3 · 6 min

Build

Spec first, then prompt: the one-page plan for when your AI keeps rewriting your app

GitHub says the fix for "it looks right but doesn't quite work" isn't a better prompt, it's a spec. Here's what spec-driven development means in plain words, and a one-page template you can use in any builder.

Aliteq Team · Oct 3 · 7 min

Build

Your AI forgot how your app works an hour in. That's a context problem, and it's fixable.

It contradicts a decision from earlier, re-breaks something it fixed, renames things you never asked it to rename. It isn't being careless — it can't see the early part of the conversation anymore. Tap the meter, then fix it the way the pros do.

Aliteq Team · Oct 3 · 7 min

Build

Run the AI on your own machine, or pay per call? What actually changes

The same kind of model can run on a company's servers or on your own computer. What flips for privacy, cost, speed and model choice — and why your GPU's memory is the ceiling.

Aliteq Team · Sep 26 · 4 min

Build

When paying extra for a "thinking" model is worth it — and when it's a waste

Some models write out their working before they answer. What that actually buys you, what the thinking tokens cost, and the kinds of task where switching it on pays off.

Aliteq Team · Sep 26 · 4 min

Build

The moment your AI stops writing text and starts doing things

A model on its own can only talk. Tools let it act inside your app; agents let it choose their own next step — and that's where the power and the risk both come from.

Aliteq Team · Sep 26 · 4 min

BuildInteractive

The AI told you to install a package that doesn't exist — that's a real attack

Code models sometimes invent library names, and attackers register those names to wait for you. The research behind it, and the one check that catches it. Try spotting the fake.

Aliteq Team · Sep 26 · 4 min

BuildInteractive

Same prompt, different code every time? Here's the dial behind it

The model picks each next word from a shortlist, and "temperature" decides how adventurous that pick is. Turn the dial, watch the output shift — and see why newer models take the choice away.

Aliteq Team · Sep 26 · 4 min

BuildInteractive

When the AI reads "it," how does it know what "it" means?

The idea behind every modern model, shown as a heatmap you can tap: how it decides which of your earlier words matter for the word it's writing next.

Aliteq Team · Sep 26 · 4 min

BuildInteractive

Why the AI keeps forgetting what you told it — and how to stop it

The model has no memory of its own. Once a conversation outgrows its window, your earlier decisions quietly fall out. See what drops, what gets summarised, and the habit that keeps what matters in view.

Aliteq Team · Sep 26 · 4 min

BuildInteractive

The AI's working memory has a hard edge, and you keep hitting it

Everything the model can see at once — your prompt, the chat so far, your files, its own reply — sits in one window with a fixed limit. Tap to watch it fill, and see why bigger isn't automatically better.

Aliteq Team · Sep 26 · 4 min

BuildInteractive

Your AI bill is counted in tokens, not words — here's what one actually is

The model never sees words; it sees tokens, and they're what your context limit and your bill are measured in. Tap the meter to watch a sentence break apart into them.

Aliteq Team · Sep 26 · 4 min