AI Death

A satirical web app that rates a company description on its likelihood of being replaced by a Claude Skill (a .md file).

No URL required. Type a sentence about what your company / product / idea does, and the court delivers a verdict. Pre-launch founders, idea-stage builders, hackathon teams, and side-project devs welcome.

Tier ladder

| Tier | Score | Vibe | |---|---|---| | IMMORTAL | 0–19 | Compliance, hardware, real-world ops. The court grants stay of execution. | | FORTRESS | 20–39 | Network effects, marketplaces, deep integrations. | | SWEATING | 40–59 | Solid product, but the LLM is closing in. | | THIN ICE | 60–79 | Mostly a UI on top of GPT/Claude with a logo. | | DEAD | 80–100 | Could be a one-page markdown skill. Funeral arrangements. |

Optional context

Below the description, three pill rows sharpen the verdict:

• STAGE — Idea / Pre-launch / Has users / Paying customers / Enterprise contracts
• BUYER — Consumer / SMB / Mid-market / Enterprise / Regulated
• SURFACE — Pure software / Hardware / Physical ops / Compliance gates

All optional. The description alone delivers a full verdict; pills can swing the score by up to ~42 points either direction and pull tier-aware roast notes.

Run it locally

It's static HTML/CSS/JS. No build step.

# any static server works
python3 -m http.server 5173
# or
npx serve .

Then open http://localhost:5173.

Deploy

Drop the folder on any static host:

• Netlify / Vercel / Cloudflare Pages — drag the folder into the dashboard.
• GitHub Pages — settings → Pages → deploy from main / /.
• Anywhere with a python3 -m http.server — same as local.

No environment variables. No keys.

How the scorer works

Pure client-side, deterministic. No API calls.

1. Start at 50.
2. Decrement for "moat" keywords (hipaa, hardware, marketplace, payments, compliance, …).
3. Increment for "wrapper" keywords (summariz, chatbot, wrapper, cold email, template, …).
4. Apply optional context shifts from Stage / Buyer / Surface pills.
5. Length signals: very short = vague (penalty); very long = specific (small mercy).
6. Add deterministic noise from a hash of the input so identical inputs are stable but the spread feels lively.
7. Clamp 0–100, map to tier.

Weights and lexicons live at the top of app.js — MOATS, REPLACEABLE, TIERS, TAGGED_ROASTS, CONTEXT_SHIFTS, CONTEXT_ROASTS, and the per-tag verb map inside buildSkill.

Files

• index.html — markup + verdict-card template
• styles.css — design tokens, tier color swaps, motion, court art
• app.js — scorer, roast bank, context shifts, render
• DESIGN.md — design system source of truth (brand voice, color, type, component inventory)

What this is NOT

• Not a real evaluation tool. Don't make decisions based on the verdict.
• Not a lead-gen page. No email capture, no waitlist, no analytics.
• Not advice. Probably not even useful. Definitely funny if your description is bad.