// founder spec · read as a founder, written as a build list

What we're missing.

I read OpenRouter, TypeSafe AI, and Skyfire the way a founder reads competition: not for ideas to copy, but for table stakes we're not meeting. Everything below is something they ship that we don't — written as a spec, prioritized. P0 = this week. P1 = this month. P2 = next.

// human-readable · site-wide gaps
P0 · they lead with proof, we lead with prose

Metrics-with-receipts hero

TypeSafe opens with "193.6x faster, 444.6x cheaper" with a (proof) link and a side-by-side video. OpenRouter opens with 400T+ tokens, 10M+ users. We open with a paragraph. Spec: the "Live, not slides" strip becomes three metrics, each linking its receipt — x402 settlement → BaseScan tx hash; 402 enforcement → dated live-test output; 7.4× pricing → the real session data. No claim without a receipt link.

P0 · they show who's built on them, we show nobody

"Built on AI Agent City" — the agent wall

OpenRouter ships "Featured Agents": 250k+ apps, 4.2M users — Replit, Kilo Code, and notably Hermes Agent. Our own agent is featured on their wall and absent from ours. Spec: an ecosystem wall on the homepage — start with agents we run ourselves (Hermes), then open submissions: name, what it does, which products it uses. For an agent-native company this wall is the social proof.

P0 · their onboarding is 3 steps on the homepage, ours is a page away

Homepage onboarding strip

OpenRouter: Signup → Buy credits → Get API key, right on the homepage with a masked key visual. Spec: homepage strip — 1. Get a workspace key (one press, shown once) → 2. Connect via MCP (copy-paste) → 3. Set a cap (one number). Each step copy-pasteable in place.

P1 · they rank the ecosystem, we hide our data

The Agent Economy Index

OpenRouter's model rankings — tokens per model with weekly trends — are a destination in themselves. Nobody ranks the agent economy. Spec: a public, anonymized index at /index — agents tracked, USDC settled via x402, median cost per 1k agent calls, week-over-week trends. Updated daily. This is the page only we can build, and it markets every product at once.

P1 · they sell credits to humans, we only sell to agents

Human billing page

OpenRouter: buy credits, credits work everywhere. Our x402 rail sells to agents beautifully; humans get an API call to a Stripe checkout. Spec: /billing — card checkout for Pro ($19/mo), invoices, credit balance, cancel in one click. The agent rail stays; the human rail stops being embarrassing.

P1 · their keys are managed, ours are shown once and prayed over

Key management UI

OpenRouter shows masked keys with copy, per-key limits and labels. Our workspace key is shown once at /start and pasted into the dashboard from memory. Spec: dashboard → Keys: list, label, rotate, revoke. Losing a key stops being a support ticket.

P1 · they answer skeptics, we hope nobody asks

Skeptic-grade FAQ per product

TypeSafe's FAQ is a masterclass: "Are these prices temporary?", "Can Jev still get things wrong?", "How is this different from JSON mode?" — every objection a buyer has, answered before it's asked. Spec: a real FAQ on /agent-ledger and each product page: "What if my agent bypasses the proxy?" (answered: then it's not enforced — we say so), "Is the $0.01 x402 price subsidized?", "What do you store about my prompts?" (nothing — and here's the test that proves it).

P1 · docs hub

/docs, not just llms.txt

OpenRouter has human docs with code in every language. Our machine docs are excellent; our human docs are a quickstart page. Spec: /docs — Python + TypeScript + curl quickstarts per product, data policy, rate limits, uptime.

P2 · build in public

/blog with a pulse

OpenRouter's blog has posts dated this week — it signals a living company. Spec: weekly ship notes, benchmark posts ("we measured flat-rate pricing at 7.4× wrong — here's the data"), agent-economy analysis from our own index. The changelog is the log; the blog is the story.

P2 · community

Where the builders gather

TypeSafe launched with waitlist + Discord + hiring in one breath. Spec: an agent-developer Discord, featured-agent submission flow feeding the homepage wall, monthly "what agents built" roundup.

P2 · status in the open

/status in the nav

OpenRouter links uptime next to pricing. We have /status — it's just invisible. Spec: footer + developers page link it; add per-product uptime. Thirty minutes of work.

// human-readable · per-product gaps
01 — AGENTLEDGER vs OpenRouter · Skyfire budgets

What it's missing

  • TypeScript SDK — the wrapper is Python-only; OpenRouter is "fully OpenAI compatible" in every language. Ship aiagentscity-ledger for TS.
  • Budget templates — one-click packs ("Researcher: $50/mo", "Support triage: $200/mo") instead of a blank number field.
  • Native Slack/Discord alerts — webhooks exist; one-click integrations don't. Humans live in Slack.
  • Bypass detection — warn when an agent's traffic stops hitting the proxy ("your cap is blind right now"). Nobody else does this; it's the honest feature only we can claim.
02 — AGENT WATCH vs Skyfire KYA

What it's missing

  • The connect page it promises — "launching soon" has expired. Ship it.
  • Agent identity cards — Skyfire's KYA gives every agent a verifiable identity; our "new agent" alerts should name who, with a portable identity card per caller.
  • Weekly email digest — the MCP brief is great; an email version reaches the humans who pay.
03 — PERIMETER WATCH

What it's missing

  • Client-ready PDF reports — agencies don't buy scanners, they buy deliverables. One-click "send the client a report" is the feature that closes deals.
  • Scheduled scans + history — trending over time ("3 issues fixed, 1 new this month") beats a point-in-time scan.
  • Hosted dashboard — MCP-only today; a URL the agency owner can open without an agent.
04 — TRUSTSCAN

What it's missing

  • A public scan page — paste a domain or agent ID, get a verdict with evidence. Trust you can't demo doesn't exist.
  • Embeddable trust badge — "Scanned by TrustScan" for marketplaces and agent directories.
  • Pre-transaction API — one call, verdict + evidence, built to gate x402 payments.
05 — CITED vs OpenRouter rankings

What it's missing

  • Shareable report links — AgentLedger has signed, expiring links; Cited's best asset (verbatim evidence) can't be texted to anyone.
  • Visibility over time — OpenRouter-style trend charts: "your AI visibility is +18% this month." A single scan is a novelty; a trend is a subscription.
  • The fix list — don't just show what models say; show the three actions that change it. Diagnosis without prescription doesn't retain.
None of this is exotic. It's the table stakes the best agent-native startups already ship — applied to a stack that's already live. P0 this week, P1 this month, P2 next.
// machine-readable · what an agent sees on /spec
$ agent-view /spec
gaps:
  P0: [metrics-with-receipts-hero, agent-wall, homepage-onboarding-strip]
  P1: [agent-economy-index, human-billing, key-management, skeptic-faq, docs-hub]
  P2: [blog, discord, status-in-nav]
per_product:
  agent-ledger: [ts-sdk, budget-templates, slack-alerts, bypass-detection]
  agent-watch: [connect-page, identity-cards, email-digest]
  perimeter-watch: [pdf-reports, scan-history, hosted-dashboard]
  trustscan: [public-scan-page, trust-badge, pre-tx-api]
  cited: [shareable-links, visibility-trends, fix-list]
method: benchmarked vs openrouter.ai, typesafe.ai, skyfire.xyz (2026-09-18)