Case studies

Everybody posts the demo that worked. A case study here is a founder-built system with the transcript of where the model was wrong and the tell that caught it. Agentic workflows, real backend, real API spend. Not replicable by a beginner, and it makes no claim to be.

The index

A case study without the recording, the transcript and the bill is a screenshot with a caption. Those three are what take the time, and they are the only reason anybody would read it.

The catch is the page.

Section five of every one of these is the moment the model was confidently wrong and the specific thing that gave it away. Nobody publishing AI app builder content has those, because you only get them by building the thing.

What a case study contains

Seven sections, identical every time. Section five is the reason the page type exists: nobody publishing AI app builder content has real agentic failure transcripts, and that is the one thing a bigger competitor cannot outspend us into.

01What it does
One sentence and one screenshot. If it takes a paragraph, the project is not understood yet.
02The demo
A screen recording, unedited, at real latency with the real failures left in. No public URL that spends API credits on a stranger's click.
03Why it exists
The actual problem, named. Not a use case invented after the fact to justify the build.
04How it is built
The architecture, honest about which parts are held together with tape.
05The catch
Where the model was wrong and how it got caught. Transcript, diff, the tell. This section is the whole point of the page.
06What it cost
Real rupees, API spend included. Case studies are not held to the ₹0 rule and do not pretend to be.
07The repo
If it is public. If it is not, the page says why rather than staying quiet about it.

Three kinds of page

A page that tries to be two of these breaks the promise the first one makes. Knowing which you are reading is the difference between a guide you can follow and a demonstration you can only watch.

How the three content types differ
TypeLives atThe promiseWhat it costs you
Build/builds/*One non-coder can follow it start to finish₹0 to build, ₹0 to keep running
Case study/work/*Real, and not replicable by a beginnerNothing. You are reading, not following
Data/data/*Our own measurement, with the method publishedNothing. Cite it if it is useful

The rule about live demos

No case study here will ever expose an agent on our own API key at a public URL. No hosting tier caps that spend, and one loop script drains it. That is why every case study ships with a recording instead of a link, and why the recording is unedited: a demo you cannot run is only worth watching if the failures are still in it.

A build page is the opposite. Its live URL is a promise, and if it dies the page gets converted or pulled the same week rather than left as a broken result on the most valuable surface on the site.

Where next

Watch the next one get made.

Each case study arrives with the prompts, the point it broke, the real rupee cost including API calls, and an unedited recording. Instagram is where the next one gets announced first.