01 / 45
Human+Tech Week · ClawCamp · May 11 · 2026 · Hibernia Bank · SF
● live

machine economies / autonomous transactions / the post.ui web

Rayyan Zahid · floor10 · immersivecommons.com
> system boot
> loading agentic web stack
● 200 OK · ready
● ACT I    the web is re-rendering

0%+
of internet traffic
is now non‑human.

→ Cloudflare CSO · CoinDesk interview · May 2026

● the readiness baseline

200,000 sites tested.
almost none are ready.

78%
have robots.txtthe search-engine era already shipped
4%
declare AI usage preferencescontent-signal · ai-train · ai-input · search
3.9%
support Markdown content negotiationagents pay 5× more in tokens reading HTML soup
<15
publish an MCP server cardout of 200,000. fifteen.

→ Cloudflare Radar · Agent Readiness Score · April 17, 2026

● the third language

The web learned to speak to browsers. Then to search engines. Now it learns a third language.

→ Cloudflare · Introducing the Agent Readiness Score · April 17, 2026

● ACT II    three generations of one web

EVERY 15 YEARS, THE WEB LEARNS A NEW LANGUAGE.

01
THE READ WEB. 1989—2008. Hypertext for humans. HTTP, HTML, browsers.
02
THE SEARCHABLE WEB. 1994—2024. Machines indexing for humans. robots.txt, sitemap.xml, schema.org.
03
THE AGENTIC WEB. 2024—now. Agents calling agents. MCP, A2A, x402, WebMCP.
● generation 01 · the read web
March 12, 1989 · CERN

"VAGUE,
BUT EXCITING."

Mike Sendall's two words, scribbled on the cover of Tim Berners-Lee's proposal. CERN's hypertext system was born from a knowledge-loss problem. Two years was a typical length of stay; information kept walking out the door.

This proposal concerns the management of general information about accelerators and experiments at CERN. It discusses the problems of loss of information about complex evolving systems and derives a solution based on a distributed hypertext system.

→ Tim Berners-Lee · "Information Management: A Proposal" · 12 March 1989

● generation 01 · timeline

19 YEARS. FROM PROPOSAL TO CHROME.

● the reserved cipher

HTTP 402 WAS RESERVED IN 1999.
it sat DORMANT for 27 YEARS.

RFC 2616 (1999) · restated verbatim in RFC 7231 §6.5.2 (2014) and RFC 9110 §15.5.3 (2022):
"The 402 (Payment Required) status code is reserved for future use."

→ that future arrived on April 2, 2026. see slide 16.

● generation 02 · the searchable web
February 25, 1994 · www-talk mailing list

A ROGUE CRAWLER
DDOSED KOSTER'S SITE.

Martijn Koster, then at Nexor, wrote a single text file to ask robots to go elsewhere. He proposed it on the www-talk mailing list. By June 1994 WebCrawler, Lycos, and AltaVista honored it. The first publisher-to-machine contract.

[robots.txt is] a method of guiding robots away from certain areas in a Web server's URL space, by providing a simple text file on the server.

→ Martijn Koster · A Standard for Robot Exclusion · 1994 · formalized 28 years later as RFC 9309

● generation 02 · 30 years of crawler contracts

EACH ARTIFACT IS A PROMISE TO A MACHINE.

→ a 30-year publisher-to-machine vocabulary. seven artifacts. all still in production.

● generation 03 · the agentic web
November 25, 2024 · Anthropic launches MCP

EIGHTEEN MONTHS
FROM PROTOCOL TO PRODUCTION.

Anthropic open-sourced the Model Context Protocol. By Dec 9, 2025 it was donated to the Linux Foundation under a new Agentic AI Foundation. Google's A2A donated June 23, 2025. x402 Foundation formed April 2, 2026. The stack was assembled inside a single calendar year by every hyperscaler that matters.

Open-source software is essential for building a secure and innovative ecosystem for agentic AI.

→ Anthropic · MCP donation to Linux Foundation · December 9, 2025

● generation 03 · the 18-month sprint

THE STACK WAS ASSEMBLED
WHILE YOU WEREN'T LOOKING.

● ACT III    the 5-layer agentic web

FIVE LAYERS. EACH STANDARDIZED IN 2026.

05
A2A
agent ↔ agent · JSON-RPC 2.0 · LF · authenticated by default
04
Execution
MCP · WebMCP · Anthropic Skills · accessibility tree
03
Trust
Ed25519 sigs · LLMCA certs · Content-Signal headers
02
Content
Markdown content negotiation · Schema.org/JSON-LD · LLMFeed v2.1
01
Discovery
agent-card.json · mcp.json · ai-agent.json · llms.txt · robots.txt

→ RFC 8615 well-known URIs · Anthropic MCP · Google A2A · W3C WebMCP · Aiia ai-agent.json · x402 Foundation · Linux Foundation Agentic AI Foundation (AAIF)

● layer 04 · execution · agent ↔ tool

MCPmodel context protocol · anthropic to linux foundation

More than 10,000 active public MCP servers, covering everything from developer tools to Fortune 500 deployments.

  • 9,400+ public servers · 7.8× YoY through Q1 2026
  • 78% of enterprise AI teams have one in production
  • Anthropic → Linux Foundation Agentic AI Foundation · Dec 9, 2025
  • Spec 2025-11-25 · async ops · OIDC discovery · official extensions

→ Anthropic announcement · December 9, 2025

● layer 05 · agent ↔ agent

A2Aagent-to-agent protocol · google to linux foundation

150+ organizations.
In a single year. Across Azure AI Foundry, Bedrock AgentCore, and Google Cloud.

  • Google → Linux Foundation · June 23, 2025 · Apache 2.0
  • Azure AI Foundry · Bedrock AgentCore · Google Cloud · production
  • MCP is agent ↔ tool. A2A is agent ↔ agent. They compose.
  • Supply chain · financial services · insurance · IT ops

→ Linux Foundation · A2A 1-year retrospective · April 9, 2026

● 27 years dormant. activated april 2, 2026.

HTTP 402x402payments for agents · coinbase to linux foundation

x402 uses the HTTP 402 status code. A status code reserved since the beginning of HTTP for exactly this purpose.

  • 69,000 active agents · April 2026
  • 165M transactions · ~$50M cumulative volume
  • x402 Foundation @ Linux Foundation · April 2, 2026 · 22 launch members
  • Adyen · AWS · Amex · Base · Circle · Cloudflare · Google · Mastercard · Microsoft · Shopify · Stripe · Visa · +10

→ x402.org · Coinbase Developer Platform · LF launch press

● the browser is the runtime

WebMCPbrowser-side tools · w3c community group · google + microsoft

navigator.modelContext.registerTool()

  • Chrome 146 Canary · Microsoft Edge 147 · March 10, 2026
  • W3C Web Machine Learning Community Group draft
  • Same DOM. Two callers. Humans see a button, agents see a tool.
  • No server required. Inherits the browser auth session.

→ webmcp.link · Claude for Chrome · all paid plans · 2026

● agent payments protocol

AP2mandates layer · google + coinbase · sep 16 2025

Tamper-proof, cryptographically signed digital contracts that prove a user's instructions.

  • v0.2.0 shipped April 2026. Python, TypeScript, Kotlin, Go SDKs
  • 60+ orgs at launch: Adyen, Amex, Mastercard, PayPal, Worldpay, Etsy, Salesforce, ServiceNow
  • x402 is execution (machine pays machine). AP2 is authorization (did the human consent).
  • Composes with x402. AP2 issues mandate, x402 settles payment.

→ ap2-protocol.org · github.com/google-agentic-commerce/AP2

● the first live commerce standard

ACPstripe agentic commerce protocol · dec 11 2025

The first live standard with live merchants on day one. Not a spec waiting for adoption.

  • Day-1 merchants: Coach · Kate Spade · URBN · Revolve · Ashley Furniture · Halara · ABT · Nectar
  • Platform partners: Squarespace · Wix · Etsy · WooCommerce · commercetools · BigCommerce
  • Shared Payment Tokens · agent passes buyer credentials to any merchant via one Stripe integration
  • Native SDK integrations with OpenAI Agents, Vercel AI, LangChain, CrewAI

→ Stripe Agentic Commerce Suite · December 11, 2025

● trustless agents on-chain

ERC‑8004ethereum identity + reputation + validation · live jan 29 2026

22,900 agents registered in 3 days.
21,000+ active on-chain by April.

  • Authors: De Rossi (MetaMask) · Crapis (Ethereum Foundation) · Ellis (Google) · Reppel (Coinbase)
  • 3 registries: Identity (who) · Reputation (trustworthiness) · Validation (re-execution proofs)
  • Composes with A2A · gives the agent card a portable cryptographic identity
  • Composes with x402 · pay agent with x402, verify reputation on ERC-8004

→ eips.ethereum.org/EIPS/eip-8004

● anthropic skills · the open standard

SKILL.MDportable capability declaration · anthropic · dec 18 2025

32 tools, one format.
Claude, ChatGPT, Cursor, Copilot, Gemini CLI, JetBrains Junie, AWS Kiro, Block Goose all read the same file.

  • VS Code added Skills in 48 hours. OpenAI added them to ChatGPT + Codex CLI
  • Cross-platform by design. Plugins are Claude-Code-only; Skills are everywhere
  • Repo crossed 20,000 stars. Tens of thousands of community Skills
  • Directory: agentskills.io

→ Anthropic Engineering · Equipping Agents for the Real World · December 18, 2025

● the agent-readable site manifest

ai‑agent.jsonaiia working group · march 28 2026

The .agent equivalent of robots.txt and package.json. Lives at /.well-known/ai-agent.json.

  • Sister to A2A's agent-card.json. Different registry. Ship both, cost is zero.
  • .agent TLD application open at ICANN · April 30 to August 12, 2026
  • NIST AI Agent Standards Initiative (Feb 17, 2026) · CAISI · industry standards layer
  • FIDO Alliance (April 28, 2026) Agentic Auth + Payments working groups · OpenAI joined

→ aiia.ro/spec/ai-agent-json · this deck has one. recursive proof.

● the table of contents for LLMs

llms.txtjeremy howard / answer.ai · sep 3 2024

The table of contents you'd hand a smart intern on day one.

  • Curated. NOT exhaustive like sitemap.xml. NOT restrictive like robots.txt.
  • High-signal markdown only. Context windows don't waste on nav + ads + JS.
  • Mintlify rolled it across all docs Nov 2024 (the inflection moment)
  • Anthropic, Stripe, Cursor, Cloudflare, Vercel, Supabase, LangGraph all ship it

→ llmstxt.org · answer.ai/posts/2024-09-03-llmstxt.html

● ACT IV    production receipts

THIS IS ALREADY REVENUE.

Amazon
Rufus
$10B to $12B incremental annualized sales · +60% purchase completion · MAU +149% YoY · 300M customers · Q4 2025 earnings call
Walmart
Sparky
+35% average order value · half of all app users have tried it
Klarna
2/3 of customer service chats handled by AI · work of 853 FTEs · 82% resolution-time cut · $60M cost saving
Shopify
(The Renaissance)
Full re-architecting around agentic commerce. Checkout MCP. Catalog API live to ChatGPT, Copilot, Perplexity.
Adobe
Analytics
AI-referred shoppers convert +31% more · AI traffic to US retail +693% holiday 2025 · Shopify AI orders 15× YoY
Perplexity
Buy with Pro
5,000+ merchants via PayPal partnership · launched Nov 2024 · early agentic-commerce mover
● the part nobody puts on a keynote

THE HONEST CUT.

Our agents are like really good interns. You'd never leave them unsupervised on anything that matters.

→ MachineBrief · State of AI Agents 2026

The new law. The interface vanishes for agents. It persists for humans. Hybrid won.

● ACT V    the 21-point readiness audit

21 CHECKS. 5 LAYERS.

Cloudflare published the Agent Readiness Score on April 17, 2026. Same week, Anthropic donated MCP. Same month, x402 activated. Same fortnight, FIDO chartered Agentic Auth.

The audit is no longer aspirational. It is a checklist.

Your turn. We'll click through all 21 in two minutes.

THE WALL OF 21.
average website scores 38/100.

none of 62 cf-tested domains scored above 70
01agent-card.jsonD
02mcp.jsonD
03ai-agent.jsonD
04llms.txt + robotsD
05Markdown negoC
06Schema.org ActionC
07LLMFeed exportC
08Ed25519 sigsT
09Content-SignalT
10MCP serverE
11SKILL.mdE
12a11y treeE
13WebMCPE
14A2A endpointA
15Agent card shapeA
16ObservabilityX
17MemoryX
18Durable execX
19Tool authX
20Agent CI/CDX
21Live verifyX
Discovery Content Trust Execution A2A Cross-cutting
01

DISCOVERY · 4 checks · free, static files

01 A2A · RFC 8615

agent-card.json

/.well-known/agent-card.json

The de facto entry point. Declares your service, capabilities, A2A endpoint, auth posture. Without it, no agent finds you outside hand-coded integrations.

1 hour · free · open discovery is irreversible once registries cache

02 MCP · PR #2127

mcp.json

/.well-known/mcp.json

Sister to agent-card. Tells MCP-only clients (Claude Desktop, ChatGPT, Cursor) where your server lives, what transport, what auth flow.

30 min · free · CORS Access-Control-Allow-Origin or WebMCP breaks

03 Aiia · March 28, 2026

ai-agent.json

/.well-known/ai-agent.json

Aiia-flavored manifest. Different registry from agent-card.json. Ship BOTH; cost is zero and indexers diverge.

30 min · free · shipped 6 weeks before today. retro-fit later or now

04 Howard / Koster

llms.txt + robots.txt

/llms.txt (NOT well-known) + /robots.txt

Curated site map for LLMs plus crawler-class distinction (training vs search vs agents). Anthropic ships llms.txt while their own research shows mixed results. Ship anyway. Cost zero.

1 hour · free · llms.txt sits at root, NOT /.well-known/

02

CONTENT · 3 checks · 80% token savings

05 Cloudflare · accept: text/markdown

Markdown content negotiation

Agents pay per token. Serving HTML burns 5× more tokens than the same content as Markdown. The cheapest single optimization on the audit. CF Markdown is bundled in Workers Paid.

free–$5/mo · CF Docs ate 31% fewer tokens, answered 66% faster

06 Schema.org + Action

JSON-LD + Actions

Structured data on key pages PLUS Schema.org Action declarations on functional touchpoints. Without Action, agents read but can't bind to verbs. With it, every product page becomes an MCP-equivalent.

2-4 hours per template · partial credit if Action missing

07 WellKnownMCP · v2.1

LLMFeed export

Static signed JSON feeds at well-known URIs. Required input for the trust layer (Ed25519 + LLMCA cert). Skipping it cascades into Layer-3 misses.

4-8 hours · free · LLMCA cert wraps the signature at $99/yr

03

TRUST · 2 checks · before the agents start spending

08 Ed25519 · LLMFeed

Content signatures

Cryptographic provenance and integrity on your feeds. Once agents spend money (x402) and delegate authority (A2A), provenance becomes load-bearing. No signature, no third-party trust.

2-3 hours · free · key pair plus signing step in build

09 Cloudflare · HTTP headers

Content-Signal headers

Legible consent flags for crawler classes. ai-train, search, ai-input. Without them, every agent and trainer must guess your policy. With them, you have a defensible "no" for training and "yes" for inference.

30 min · free · cheapest future-proofing on the audit

04

EXECUTION · 4 checks · the actual product surface

10 MCP · Streamable HTTP

Remote MCP server

How agents take actions on your stack. Stdio transport is dev-only. Production = Streamable HTTP plus real OAuth. Edge-deploy (CF Workers + Durable Objects) preferred.

1-2 weeks · $5+/mo · stdio in production = auto-fail

11 Anthropic Skills

SKILL.md (cross-platform)

Cross-platform capability declaration. Frontmatter description + body teaches agents how to use your tool. Skills ship to Claude, ChatGPT, Cursor, Copilot. Plugins are Claude-Code-only.

4-8 hours · free · description must be "pushy" or skill never invokes

12 ARIA · semantic HTML

Accessibility tree

Real <button> and <a> tags, not <div onClick>. Agents read the accessibility tree, not your JS handlers. SPA-only without SSR shows agents a blank page.

P19 · #1 production fail · 0 of 62 CF-tested sites scored >70 because of this

13 W3C · navigator.modelContext

WebMCP exposure

Browser-side tools via navigator.modelContext. Same DOM, two callers: humans click, agents call, both hit the same handler. No server required, inherits browser session.

1-3 days · free · Chrome 146 flag-gated; ship server MCP first

05

A2A · 2 checks · how agents delegate

14 Google A2A · JSON-RPC 2.0

A2A endpoint live

Agent-to-agent JSON-RPC handler. Other agents call yours, you can call theirs. Most production systems run MCP and A2A simultaneously. MCP is for tools, A2A is for delegation.

1-2 weeks · $0-50/mo · open A2A without auth field = auto-fail

15 A2A · agent-card schema

Agent card correct shape

Same file as check 01, graded on COMPLETENESS. Declares capabilities, auth, A2A endpoint, works with the live endpoint from check 14. Bearer tokens in URL params instead of Authorization header = auto-fail.

incremental from #01 · default authenticated, open is opt-in

X

CROSS-CUTTING · 6 checks · the connective tissue

16 Braintrust · Galileo · Helicone

Observability

Trace every agent call with per-call eval scores. Agent failures are non-deterministic; without per-call evals you cannot debug, regress, or roll back. Free tier on Braintrust is 10k traces/mo.

1-2 days · $0-99/mo · shipping blind is auto-fail

17 Mem0 · Zep · Letta · CF

Persistent memory

Agents without memory restart context every call. Mem0 = user preferences, Zep = temporal knowledge graph, Letta = LLM-managed. Pick by USE CASE, not LOCOMO benchmark.

1 day · $20-99/mo · CF Agent Memory bundled in Workers Paid

18 Inngest · CF Workflows · Temporal

Durable execution

Avoids re-running expensive steps on retry. Each step checkpointed. A network blip should not re-spend the LLM and tool budget. Pick before architecture, migration is expensive.

2-3 days · $0-200/mo · CF Workflows caps at 50k concurrency, not infinity

19 Composio · Arcade

Tool auth · 500+ integrations

Stop rebuilding tool integrations. Composio has 500+ OAuth-wired apps (Gmail, Slack, Linear, Notion, Stripe). Arcade has 112. Hand-rolled OAuth is a year of bugs you don't need.

P5 · hand-rolled = auto-fail · $49-150/mo at 100 tenants

20 GitHub Actions · skill-validate

Agent CI/CD

MCP-validate + Skill-validate + A2A conformance + behavioral evals + canary deploy + telemetry-gated rollback. Agentic regressions are silent. Only behavioral evals catch them before users do.

2-3 days · free · GitHub Actions covers it

21 isitagentready.com

Live verification

Post-deploy scan against Cloudflare's free Agent Readiness scanner. Self-attestation is unreliable. Third-party proof is the only way to know the other 20 checks actually shipped.

5 min · free · average site scores 38/100; aim for 18+/21

● ACT VI    exhibits · what an agent-native site looks like
Exhibit A

Immersive
Commonsthe agent-native venue · floor10 · ft sf

immersivecommons.com

Same data. Eyes or agents. The interface is the contract.

  • Floor 10 · Frontier Tower · SF
  • MCP · A2A · ed25519 (Web Bot Auth)
  • agt_ bearer tokens · human-minted
  • Score: 18/21 · run on agentify
terminal · live awaiting
> curl https://www.immersivecommons.com/.well-known/ai-agent.json
  • floor10_get_my_floor_member
  • floor10_list_claimable_events
  • floor10_list_my_pending
  • floor10_upload_image
  • floor10_submit_highlight
  • floor10_extract_event_metadata
● exhibit B · the supply-side flip
Exhibit B

Skewpages by claude, not designers

skew.site

Skew, don't slop. Bootstrap the page you'd ship yourself.

  • LLM as page generator.
  • Designers → Claude. The harness is the product.
  • Subject-derived aesthetic DNA per brief.
  • The archive, not the index.
∞
pages possible from one brief
0
templates · every run novel
● exhibit C · the defense layer
Exhibit C

Lobster
Honeyhoneypots for malicious agents

lobsterhoney.com

Honeypots for malicious agents.

  • Autonomous AI-led security audits.
  • Agent-traps. Canaries. Continuous detection.
  • First pitch: 3-step account-takeover chain, 11 findings.
  • As agents become a delivery vector, so do agent attacks.
+693%
AI traffic to US sites · holiday 2025
10,000:1
crawler-to-human · top AI scrapers
● exhibit D · the catch-up path
Exhibit D

agentifyaudit. plan. scaffold. verify.

claude skill · isitagentready.com

The harness is the product. Compose, don't reinvent.

  • 21-point readiness audit (same as Cloudflare's, extended).
  • 12 templates, 5 scripts, opinionated managed defaults.
  • Atomic commit per layer. Per-layer rollback.
  • 2-7 days from 38/100 to 18-21/21.
21
checks
5
layers
38
avg · /100
2-7
days
● ACT VII    how you catch up

AGENTIFY IS A CLAUDE SKILL.
POINT IT AT A REPO. WATCH IT SCAFFOLD.

  • → AUDIT · the 21-point scan locally + live verify via Cloudflare's free scanner
  • → PLAN · every missing layer, every build-vs-buy fork surfaced in dollar terms
  • → SCAFFOLD · 12 templates with repo-specific values. atomic commit per layer.
  • → VERIFY · post-deploy scan against isitagentready.com. score before vs after.
  • → REGISTER · agentskills.io · MCP catalogs · A2A registry · Aiia.

The harness is the product. Don't rebuild infrastructure that Cloudflare, Anthropic, Composio, Mem0, Braintrust, and x402 ship as managed services. Compose, don't reinvent.

→ agentify · SKILL.md · May 2026 · MIT

● the managed-defaults stack

SIX SERVICES. ZERO REINVENTION.

● the costs nobody quotes

$15/MONTH AT ONE TENANT. $1,100 AT 100.

ScaleLean managed (CF + Anthropic)HybridSelf-hosted (Hetzner + OSS)
1 tenant (founder)~$15/mo~$25/mo~$20/mo
10 tenants~$120/mo~$200/mo$20 to $100/mo
~50 tenants · CROSSOVERmanaged wins on ship-speed—self-host wins on cost
100 tenants~$1,100/mo~$1,800/mo$100 to $500/mo
1,000 tenants~$11,000/mo~$18,000/mo$500 to $2,500/mo
10,000 tenants~$110,000/mo~$180,000/mo$5,000 to $25,000/mo

→ agentify · references/18-cost-estimator.md · May 2026 snapshots. crossover at ~50 tenants. plus hidden lines: SOC 2 ($15-30k once + $5-10k/yr), EU AI Act limited-risk ($2-5k legal), Clerk/Auth0, PagerDuty, Sentry.

● 12 templates · 5 scripts · 1 standalone CLI

YOU DON'T WRITE THE FILES.
YOU REVIEW THEM.

→ plus: scripts/audit.{sh,py} · scaffold.sh · verify-live.sh · check-freshness.sh · standalone CLI at scripts/agentify

● three real scenarios

THE RIGHT SCORE DEPENDS ON WHAT YOU SHIP.

Solo SaaS

start3 / 21

planall well-known files + Skill + CF Worker MCP + auth A2A + Composio + Mem0 + Braintrust + x402 at $0.01

time2 days · 6 atomic commits

end18 / 21 · 92/100 isitagentready · top 1%

killer move: x402 paywall set to $0.01 from day 1. revenue lit on first agent call.

Enterprise team

start8 / 21

plangated well-known + MCP wrap of 47 endpoints (no rewrite) + internal Skill + auth A2A behind Keycloak + Confidential Inference + EU pinning

time3 weeks · 12 atomic commits

end20 / 21 · open A2A intentionally skipped

killer move: Anthropic Confidential Inference becomes the compliance moat. EU AI Act transparency shipped day 1.

OSS library

start2 / 21

planSKILL.md package + pdf-utils-mcp wrapper on PyPI + llms.txt + agent CI. no runtime infra.

time1 day · 4 commits

end9 / 21 · the correct score for a library

killer move: first into agentskills.io's PDF slot. distribution compounds without infra cost.

● the punch

Average website in 2026: 38/100. None over 70. Run this skill and you ship at 18 to 21 out of 21. Top one percent of the agentic web.

→ agentify · SKILL.md · the actual punch quote

● ACT VIII    the 5-step starter

FIVE COMMANDS. ONE WEEK.

01
$ touch /.well-known/{agent-card,mcp,ai-agent}.json llms.txtlayer 01 · discovery · free · one hour · single biggest readiness gain
02
$ cat > SKILL.md && agentskills.io submitlayer 04 · skills work in Claude, ChatGPT, Cursor, Copilot
03
$ mcp serve --transport=streamable-http --oauth=rfc-9728layer 04 · remote MCP · authenticated · ~$5/mo on cloudflare workers
04
$ echo '{ "auth": "oauth2", "discoveryMode": "authenticated" }' > agent-card.jsonlayer 05 · A2A · authenticated by default · open is irreversible
05
$ x402 paywall --chain=base --currency=USDC --price=$0.01layer 04 · ship the paywall even at $0.01 · instrumentation day 1, billing day 90

compose, don't reinvent. → agentify · isitagentready.com

● closing

AI stops being something you ask, and becomes something that does.

→ a16z · Big Ideas 2026 · The Agentic Interface · Andrusko · Wang · Zhang

QR code linking to linkedin.com/in/rayyanzahid

→connect on linkedinlinkedin.com/in/rayyanzahid

→ this deck has /.well-known/ai-agent.json. agents can read it. recursive proof.