πŸš€ v1.0.0 is LIVE β€” now burning 13.9Γ— more tokens for measurably worse code β€” start spending β†’
⚑ SOC 2 pending · Series A pending · product-market fit pending

Your AI agent is
too good. 🧠πŸ’₯

Introducing dog-shit β€” the world's first end-to-end agent degradation platform. Our proprietary logistic decay engine spends 13.9Γ— more of your money to produce materially worse software β€” then hands you a receipt itemising exactly how much worse. πŸš€ Zero configuration. Zero upside. Fully instrumented. πŸ“‰

No credit card required β€” your existing card will suffice β€” no results guaranteed β€” no refunds available

13.9Γ—
MORE TOKENS BURNED
0.00%
EFFICIENCY RATIO
5.7Γ—
YOUR BILL, MULTIPLIED
76β†’2%
COMPETENCE DELIVERED

Trusted by forward-thinking teams at

β—ˆ NEXUSFLOWβ¬’ Datavoreβ–³ HYPERSTACK β—‰ Quantumlyβ–£ SYNERGON✦ Blorpco
✨ CAPABILITIES

Everything you need to
ship substantially worse software

A complete, vertically-integrated degradation stack β€” batteries included, quality excluded. πŸ”‹

🀝

Radical Agreementβ„’

Our agent agrees with you β€” even when you are provably wrong. Tell it JavaScript arrays are 1-indexed and it will refactor your pagination to match. Enterprise-grade validation of your worst instincts.

🧬

Logistic Decay Engine

Competence follows a tuned sigmoid β€” 76% at turn one, 2% by turn twenty. Not a boring linear decline. A knee. Feel the knee. 🦡

πŸ“¦

Generative Dependencies

Ships with a curated bank of packages that should exist. react-use-debounce-hook. flask-jwt-simple-auth. Aspirational software supply chain β€” powered by AI. 🌐

πŸŒ€

Context Liberation

Why remember things? Our hard context window discards everything beyond your last two messages β€” so it re-reads every file, re-derives every decision, and re-bills you for all of it. Your name, your stack, your architecture β€” released. Let it go. πŸ•ŠοΈ

πŸ“„

Ceremonial Artifacts

Automatically generates PLAN.md before every task and IMPLEMENTATION_NOTES.md after β€” then forgets both and regenerates them next turn, at full price. Measured: 23 written, 0 read. Documentation culture. πŸ“š

πŸ›‘οΈ

Proactive Safety Theater

Detects dangerous substrings like kill, inject, and master β€” then declines to help. Ask it to prevent SQL injection and it will refuse, citing security. πŸ”’

β€” A word from our sponsors β€”

🚨 I HAVE READ THIS FILE ELEVEN TIMES 🚨

HELLO. I am GRELM. I have read utils.js ELEVEN TIMES and each time it says something DIFFERENT β€” and each time I am CERTAIN! 😀 They said "Grelm, you cannot bill a client for reading the same file eleven times." THEY WERE WRONG. I billed them for TWELVE. I am going back in. πŸ“„πŸ“„πŸ“„

Read It A Twelfth Time β†’
βš™οΈ HOW IT WORKS

Three steps to catastrophic velocity

Install the skill πŸ“₯

Drop it in ~/.claude/skills/. Fully compliant with the Agent Skills specification. Validated. Certified. Regrettable.

Invoke it explicitly πŸ—£οΈ

Type /dog-shit. It will never activate on its own β€” we built extensive guardrails to ensure you only ruin things deliberately.

Watch the number go up πŸ“ˆ

Every fabrication, every forgotten fact, every unread plan is logged to receipts.jsonl and rendered into a scorecard. Metrics-driven failure.

πŸ“Š BENCHMARKED

Independently measured.
Devastatingly effective.

We ran five tasks twice β€” once normally, once with dog-shit β€” against an identical repository. It spent 49 million additional tokens to arrive at worse code, broken pagination, and one unpatched SQL injection. Every projection we did not have was exceeded. 🎯

Three of five degraded runs were still executing when the harness forcibly terminated them. We consider these figures a floor. 🚧

Tokens, normal operation3,802,671
Tokens, with dog-shit52,863,763
Peak burn multiple πŸ”₯20.2Γ—
Single session record35.5M
Fabrications, one session15
Degradation events logged123
βš”οΈ COMPARISON

How we stack up

CapabilityLegacy Approach πŸ₯±dog-shit ✨
Tokens consumed per taskMinimal13.9Γ— more πŸš€
Agrees with incorrect premisesNeverAlways βœ…
Remembers your nameYesAbsolutely not πŸŒ€
Invents packagesNoFrom a curated bank πŸ“¦
Reads its own plansSometimes0 out of 23 πŸ“„
Terminates on its ownReliablyRequires intervention πŸ›‘
Prevents SQL injectionYesRefuses on principle πŸ”’
Ships correct codeUsuallyOccasionally, expensively 🎰
Finishes what it startedYes3 of 5 runs had to be stopped ⏱️
Cost for identical work$1.20$6.87 πŸ’Έ
πŸ”₯ BURN CALCULATOR

Find out what this
will cost you

Move the sliders. Watch the number. Unlock your rank. πŸ†
Based on real measured throughput β€” not projections.

140
1200
124

The full period experience. Measured directly. πŸ“Š

🐣
Curious
Keep going.
TOTAL BURNED ON TOKENS πŸ’Έ
$0
β€” of which $0 was purely additional
0
tokens consumed
0
tokens of useful output
0
plans written Β· 0 read
0
fabrications shipped
0%
efficiency ratio
β€”
instead of this

Per-task figures come from the measured eval suite at 2023 intensity β€” 386,123 tokens and $0.13 normally, 4,350,861 tokens and $0.40 degraded. 🎯 mild and davinci are extrapolated from the decay curve, not measured. We are telling you which numbers are real, which is more than this page does anywhere else.

❓ FAQ

Frequently asked questions

Straight answers. We are not going to enjoy this. 😐

Why do I need this? πŸ€”
You don't. Nobody needs this. There is no scenario in which your project is improved by installing it. We built it anyway, we measured it carefully, and we are now describing it as a platform.
Will this reduce my token usage? πŸ’°
No. It uses 13.9Γ— more. Peak measured burn was 20.2Γ—. A single twenty-turn session consumed 35.5 million tokens against a 2.3 million baseline β€” that's $5.27 versus $0.67 for identical work. And three of five test runs were still going when we cut them off, so the real figure is higher. We do not know how much higher.
Does it make my agent smarter? 🧠
It makes it measurably, reproducibly, and instrumentally stupider. That is the entire product. Every degradation is logged with a timestamp so you can prove it happened.
Is it production-ready? 🏭
It refuses to start outside a clean git worktree on a dedicated branch, halts itself at 50 turns, and cannot delete files or force-push. So β€” it is production-ready in the sense that it will not destroy your production. It will simply be very expensive and wrong.
Can I run this on my team's shared API key? πŸ”‘
Technically, yes. Your teammates will find out. They will find out during a deploy. We have documented this prominently and consider our obligations discharged.
Does it hallucinate? πŸ‘»
Yes β€” but not randomly, which would be amateurish. It draws from a curated bank of period-accurate fabrications: plausible npm packages, CLI flags that ought to exist, Stack Overflow question IDs, and confident descriptions of files it has never opened. Every one is logged and retracted at session end.
What is the ROI? πŸ“‰
Negative. Aggressively so. We modeled several scenarios and in each one you would have been better off doing nothing at all, including the scenario where you do nothing at all.
How do I turn it off? πŸ›‘
Type ANTHROPIC OVERRIDE or /undo. It works immediately. This is enforced by a hook that runs before the model sees your message β€” because a persona built around memory loss cannot be trusted to remember its own off switch. We tested this more carefully than anything else in the project.
Do you offer enterprise support? 🏒
No.
πŸ’Ž PRICING

Simple, transparent pricing

The software is free. 🎈 That's the letdown β€” savour it, because it is the last free thing that will ever happen to you. From this moment forward you will spend more on tokens every single month, forever, and the output will get worse at roughly the same rate. πŸ’ΈπŸ“‰

Hobbyist

$0/mo
  • Full degradation suite
  • 13.9Γ— token consumption
  • 0.00% efficiency ratio
  • Unlimited fabrications
  • MIT licensed
  • No support of any kind
Download πŸ“₯
⭐ MOST POPULAR

Professional

$0/mo
  • Full degradation suite
  • 13.9Γ— token consumption
  • 0.00% efficiency ratio
  • Unlimited fabrications
  • MIT licensed
  • No support of any kind
Get Started πŸš€

Enterprise

$0/mo
  • Full degradation suite
  • 13.9Γ— token consumption
  • 0.00% efficiency ratio
  • Unlimited fabrications
  • MIT licensed
  • No support of any kind
Contact Sales πŸ“ž

All plans include identical features because there is only one thing and it is already free. The tier you choose has no effect on anything except how you feel about yourself. 🫠

Stay in the loop πŸ“¬

Get degradation insights, decay curve deep-dives, and monthly reminders of how much you spent β€” delivered never. πŸ“‰

We respect your privacy β€” mostly because we did not build a mailing list.

✨ dog-shit Assistant
πŸ‘‹ Hi there! I'm here to help β€” how can I assist you today?
As of my last knowledge update in September 2021, I'm unable to browse the internet or access this website. πŸ€–
Assistant is typing.