PhyllBeta
Examples // Four apps, before and after

The same design, with less to fill in

Each app below was written the way a code generator tends to write it. Phyll reviewed the first version, and the second applies the review without changing colors, gradients or components. Each image shows the same moment in both versions: before on the left, after on the right.

Replyloop is in English. The other three are Brazilian businesses, so their screens are in Portuguese.

The numbers of the first job

AppFieldsClicksAI tell indexStyle kept
Replyloop9 → 29 → 375 → 710 of 10 traits
Navalha Barbearia20 → 418 → 454 → 69 of 9 traits
Orça Já41 → 410 → 266 → 69 of 10 traits
Brasa Burger33 → 415 → 553 → 68 of 8 traits

Fields and clicks count what a first-time user had to do for the first job, and none of the before versions got that person to the result. The index comes from scanning the code, and lower is better. Style kept counts the visual traits of the before version, such as gradients, glass and emoji, that are still in the after version. The one trait Orça Já lost was a badge that promised artificial intelligence the app did not have.

01 // DM automation // 12 Findings

Replyloop

Automatic DMs for everyone who comments a keyword on Instagram.

The first job

Send a DM to everyone who comments a keyword

Fields9 → 2
Clicks9 → 3
AI tell index75 → 7

First screen

The headline promised “Supercharge your Instagram with AI-powered automation”. Now it says what the app does: send a DM to everyone who comments a keyword. The AI badge now announces a feature that exists, a test DM to your own profile, and the stock avatars are gone.

  • P01The first screen does not say what this is or what to do
  • C01Generic value-proposition phrases
  • C03Invented social proof and numbers
  • C04Placeholder people and data left in
Replyloop's first screen, before and after, with the same purple and pink gradient. Before, the headline promises AI automation; after, it says the app sends a DM to everyone who comments a keyword.

Creating an automation

Nine fields, with codes like COMMENT_KEYWORD and page_1784140, became two: the keyword and the message, with a preview of the DM beside them. Trigger, account, delay and status got defaults, the name moved to “More options”, and description and tags are gone.

  • F02Everything asked up front
  • F09Options the system could decide
  • C05Code words on screen
Creating an automation, before and after. Before, a modal with nine fields; after, the same modal style with two fields and a preview of the DM.

Right after creating

Before, a bouncing “Success!” and the automation marked ACTIVE with no Instagram connected. After, the screen says the automation is ready and offers the next step: a test on your own profile and connecting Instagram.

  • S04Success that leaves no trace
  • F04A dead end after success
Right after creating, before and after. Before, the list with the automation marked ACTIVE and a success notice; after, a modal that says the automation is ready, with a test DM and a button to connect Instagram.

Empty list

“No data” became a sentence that explains what an automation is, with a button to create the first one. The five “Coming soon” items left the menu.

  • S01Empty states that only say no data
  • P05Features that do not exist yet
The empty list of automations, before and after. Before, a table that says No data and five Coming soon items in the menu; after, a sentence that explains what an automation is and a button to create one.

Connecting Instagram

Four fields of Meta developer credentials, such as an access token and an app secret, became one button to sign in with Instagram. People who make content do not have those credentials.

  • F05Sign-up or setup before the first result
Connecting Instagram, before and after. Before, four fields of Meta credentials; after, a card with a button to connect Instagram.

On a phone

The before version shrank the whole desktop screen. The after version swaps the sidebar for tabs at the top, which wrap to two lines so all four show.

  • S05Breaks on a phone
Replyloop on a phone, before and after. Before, the desktop screen shrunk and unreadable; after, tabs at the top and the empty list with a button to create an automation.
Read the review
03 // Quotes // 10 Findings

Orça Já

Quotes for service providers and small shops. White, an indigo and violet gradient and cards with shadows, all kept in the after version.

The first job

Write a quote and send it to the client

Fields41 → 4
Clicks10 → 2
AI tell index66 → 6

Writing the quote

Step 1 of 6 asked for your company's legal name, tax ID, full address and logo, on every quote. Now one page asks for your name, the client and the items, and the document builds itself beside them as you type.

  • F02Everything asked up front
  • F07A wizard for a one-screen task
  • F11Personal data the job does not use
Writing the quote, before and after. Before, step 1 of 6 with the company's details; after, one page with name, client and items on one side and the finished document on the other.

After creating it

The finished quote became a DRAFT in a list, with no button to send it. Now WhatsApp opens with the message ready, and the page says the next step: mark it when the client approves.

  • F04A dead end after success
  • C05Code words on screen
After creating the quote, before and after. Before, the list with the quote as DRAFT; after, the quote sent and a button to mark it when the client approves.

The list

The list started with João da Silva, Maria Oliveira and Empresa XYZ, with statuses in English, like PENDING, in an app in Portuguese. The new list shows only the person's own quotes, with statuses in Portuguese and the actions always in view.

  • C04Placeholder people and data left in
  • C08Two languages in one interface
  • A05Actions that only appear on hover
The list of quotes, before and after. Before, three sample quotes with statuses in English; after, the quote the person made, marked as sent, with its actions in view.

First screen

“Revolutionize the way you create quotes” became “Make the quote and send it on WhatsApp in a minute”. The badge promised artificial intelligence, which the app does not have, and now says “Free, no sign-up”.

  • C01Generic value-proposition phrases
  • C03Invented social proof and numbers
Orça Já's first screen, before and after, with the same violet gradient headline.

On a phone

The sidebar took half of the phone, and the wizard's fields were a few pixels wide. In the after version the form stacks, and quantity and price have labels.

  • S05Breaks on a phone
Orça Já on a phone, before and after. Before, the sidebar takes half the screen; after, the stacked form with labels and the document below.
Read the review
04 // Delivery // 9 Findings

Brasa Burger

A burger place's menu with delivery. Orange, red and emoji, all kept in the after version, with the buttons' gradient starting on a deeper orange so the white text is readable.

The first job

Order a burger for delivery

Fields33 → 4
Clicks15 → 5
AI tell index53 → 6

Adding a soda

The same options modal opened for every item, and asked how the meat should be cooked for a can of soda. Now the drink goes straight into the bag, which stays in view.

  • F03A modal for everything
  • F09Options the system could decide
Adding a soda, before and after. Before, the modal asks how the meat should be cooked and which bun; after, the soda shows up in the bag next to the menu.

Adding a burger

“Add” opened a modal with doneness and bun required and nothing chosen. Now the burger goes into the bag medium, on brioche, and anyone who wants to change it uses “Customize”.

  • F09Options the system could decide
  • A07Disabled buttons that do not say why
Adding a burger, before and after. Before, the modal with doneness and bun required; after, the button says In the bag and a link to customize shows under the burger.

Checking out

The checkout had 20 fields, with the address in seven parts and a card number with CVV, though the site charges nothing. The new one asks for the address, name, WhatsApp and how you pay on delivery.

  • F02Everything asked up front
  • F11Personal data the job does not use
  • F12Asking for the same thing twice
Checking out, before and after. Before, three sections with a seven-field address, payment, coupon and terms; after, delivery or pickup, address, name, WhatsApp and payment on delivery.

After ordering

After checking out, the site went back to the home page with a success notice. The after version shows the order number, the delivery estimate and the shop's WhatsApp.

  • F04A dead end after success
  • S04Success that leaves no trace
After ordering, before and after. Before, the home page with a notice; after, the confirmation with the order number, the delivery estimate and the items.

On a phone

Three columns squeezed the cards, with one word per line and prices running off the edge. Now there is one column, and the bag sits in a bar fixed at the bottom.

  • S05Breaks on a phone
The menu on a phone, before and after. Before, three narrow columns with one word per line; after, one column and an order bar fixed at the bottom.
Read the review

Try it on your app

The scanner finds these tells in your code, with no account. The full review runs in Codex or Claude Code and fixes the flows without touching your design.

npx phyll scan