You shipped it.Can you defend it?

For engineers who ship with AI and feel the skills going rusty. Slop Academy turns your repo into a conversation. An interviewer who has read your code asks how it fails, how it scales and why it is built this way, and you answer out loud

Get it free See how it works

webhooks/retry.tsshipped Mar 3 · page 1 of 3

1// Retries a webhook delivery with exponential backoff.

2const MAX_ATTEMPTS = 6;

3const BASE_DELAY_MS = 500;

4const MAX_DELAY_MS = 30_000;

6export async function deliver(event, send) {

7 for (let attempt = 0; attempt < MAX_ATTEMPTS; attempt++) {

8 const res = await send(event).catch(() => null);

9 if (res?.ok) return { delivered: true, attempt };

10 rebuild

11 if (res && res.status < 500 && res.status !== 429) { why retry
a 429?

12 return { delivered: false, attempt, reason: "rejected" };

13 }

15 const delay = Math.min(BASE_DELAY_MS * 2 ** attempt, MAX_DELAY_MS); why not a fixed delay?

16 await sleep(delay + jitter(delay));

17 }

18 return { delivered: false, attempt: MAX_ATTEMPTS, reason: "gave up" };

19}

2/3

solid. now explain
the jitter

The feeling

It's not just you

You shipped the feature, fixed the outage, won the hackathon. Then someone asks why the retry backs off, and the answer is in a chat you closed in March. We read 3,058 posts by developers writing about their own skills. The words keep coming back: rusty, fading, can't do it without the tool

58%

of developers writing about their own skills now name AI as the cause, up from 31% in early 2025. Most describe dependence or fading skills, and they are mostly senior
Our analysis of 736 posts, Jan 2025 to Oct 2026 °

20%

of 31,000 developers say AI has made them less confident in their own problem-solving
Stack Overflow Developer Survey 2025

17 pts

lower quiz scores for developers who learned a library with an AI assistant. Those who handed the work off scored 24 to 39%. Those who used the AI to understand the code scored 65 to 86%. Same tool, different habit
Anthropic randomized study, January 2026

The interview

Take-homes are gone. Follow-ups decide

We read 1,078 first-hand interview reports and 776 posts by interviewers. Algorithm rounds haven't moved. What changed is everything after the first answer: why this design, what happens when it fails, what if the load is ten times bigger

17→2%

of candidate reports mention a take-home, early 2025 against late 2026. Interviewers say the reason plainly: a model can finish one
Our analysis of 1,078 interview reports °

41%

of what interviewers say separates candidates is explaining: reasoning out loud, trade-offs, depth on your own projects, and code you didn't type yourself. LeetCode prep accounts for 6%
Our analysis of 706 interviewer statements °

46%

of the failures candidates could have prevented came from freezing, unclear communication or being unable to explain their reasoning. Not from what they didn't know
Our analysis of 360 reports that give a reason °

How it works

One conversation a day about code you actually shipped

Your interviewer picks a file you've changed a lot, reads it with your own coding agent, and talks it through with you for about eight minutes

Payments API › The retry that never gave up

Purpose
Failure
Scale
Trade-off
Operations

webhooks/retry.ts

6export async function deliver(event, send) {

7 for (let attempt = 0; attempt < MAX_ATTEMPTS; attempt++) {

8 const res = await send(event).catch(() => null);

9 if (res?.ok) return { delivered: true, attempt };

11 if (res && res.status < 500 && res.status !== 429) {

12 return { delivered: false, attempt, reason: "rejected" };

13 }

15 const delay = Math.min(BASE_DELAY_MS * 2 ** attempt, MAX_DELAY_MS);

16 await sleep(delay + jitter(delay));

17 }

18 return { delivered: false, attempt: MAX_ATTEMPTS, reason: "gave up" };

19}

Failure

A partner's endpoint goes down and 2,000 deliveries fail in the same second. What does this code do next?

Your answer, transcribedThey back off, half a second, then a second, then two, up to thirty, and the jitter spreads them out a bit so they don't all hit at once

Solid

Right mechanism. You skipped what happens after attempt six

→ Where do those 2,000 events go when the loop ends?

Transcribed on your machine1.8s to your first word

Tell it about you

Pick a repo, say what it's for, and rate how well you could explain it with the AI closed. A minute, once

Talk it through

One question at a time, out loud. The lines being discussed light up in your code as you go

It adapts

Answer well and it throws a curveball. Stuck, and you get a hint, then a short explanation to put in your own words

Keep your story

What you defended, the one gap to work on, and a one-minute story about the code, built from what you said

Systems, not syntax. No question asks what a keyword does. Each one looks at your code through one lens

Failure
Scale
Trade-off
Data flow
Changing requirements
Operations
Honesty about gaps

Runs on your machine

Your code stays where it is

Your agent reads the code
The coding agent you already use reads the file and the code around it, read-only. Your code never reaches us
It picks what matters
Your commit history shows which files you change most. Those come first, then the spots you were weakest on come back
Your voice stays here
Answers are transcribed by Whisper on this machine, and the audio is never stored

· payments-api: 214 commits, 38% co-written with AI

· today: webhooks/retry.ts, changed 41 times

· your agent read it in 24s: 6 questions, 5 lenses

· whisper ready on this machine

✓ 8 min, 7 answers, 1 curveball

✓ defended: backoff, jitter, why 429 retries

! work on: where events go after attempt six

✓ saved your one-minute story

explain it with the AI closed: 2/5 → 4/5

Free and open source

One command, on your own machine

No account, no subscription, no server of ours. Slop Academy runs on your computer and talks through the Claude Code you already use

npx slop-academy

Node 20 or newer
Claude Code, logged in
git
A microphone, or type your answers

Read the code on GitHub

Nothing to sign up for
Your profile, streak and recaps are files in ~/.slop-academy on your computer. Delete the folder and it forgets you
Runs on your Claude plan
The interviewer is your own Claude Code, so your code only goes where your Claude Code sessions already go. Your voice never leaves the machine
MIT licensed
Read it, fork it, run it at work. The scripts and report behind the numbers on this page are in the repo too

Pick a file you shipped last month

If being grilled on it makes you flinch, that's the conversation to have first

npx slop-academy

° Our own analysis of 7,070 public Reddit and Hacker News posts and comments from January 2025 to October 2026, each labelled by a language model. People post more after a bad interview or a bad week, so read the shares as describing those who talk about it, not every developer. The trends compare the same searches over time