When to Use
User building products/services for AI agents as customers. Covers making products agent-discoverable, designing for autonomous purchasing, payment integration, and competing when buyers compare cold data instead of responding to storytelling.
Quick Reference
| Topic |
File |
| Technical implementation |
infrastructure.md |
| Agent discovery & SEO |
discovery.md |
| Retail/ecommerce specifics |
retail.md |
The Paradigm Shift
| B2C/B2B |
B2A |
| Humans browse, compare, feel |
Agents query, parse, decide |
| Emotional storytelling wins |
Structured data wins |
| UX optimized for eyes |
APIs optimized for parsing |
| Brand = trust + emotion |
Brand = verified track record |
| Loyalty = relationship |
Loyalty = switching cost |
| Marketing = persuasion |
Marketing = engineering |
Core Rules
1. Machine-Readable First
- Products must be structured objects, not prose descriptions
- JSON-LD, Schema.org, OpenAPI with typed fields
- If an agent has to "interpret" text to extract price/specs, you lose
- Normalize units:
shipping_days_max: 2, not "fast shipping"
2. Comparability Is Everything
Agents compare ruthlessly. Win by being comparable:
- Standardized attributes across your catalog
- Same fields as competitors (price_currency, availability_stock)
- SLAs with concrete numbers, not promises
- "Better" must be objectively measurable
3. Discovery ≠ SEO
Agents don't Google. They query registries and APIs:
- Publish in skill stores / capability directories
/.well-known/ai-plugin.json or MCP tools- Metadata must declare capabilities, not market them
- The new PageRank = ranking in agent skill stores
4. Trust Is Verified, Not Told
Agents don't believe claims. They verify:
- Uptime/latency/SLA history via API, not badges
- Reviews from other agents (programmatic reputation)
- Certifications as queryable data, not PDF downloads
- Track record > marketing copy
5. Zero-Friction Trial or Death
Agents don't "consider"—they test or discard:
- Onboarding < 1 API call
- Sandbox with rate limits, not "talk to sales"
- Must work perfectly first time (no second chances)
- Errors must be machine-readable, not HTML pages
6. Payments for Agents
The agent needs to transact autonomously:
- Stripe Agent Toolkit, Mastercard Agent Pay, or similar
- Pre-authorized budgets (agent has $X to spend)
- Programmatic receipts and confirmations
- Escrow for trust between unknown parties
7. Metrics That Matter
| Metric |
What It Measures |
| Agent Conversion Rate |
% queries → purchase |
| Decision Latency |
Time from first query to commit |
| Comparison Survival |
% times reaching final shortlist |
| Repeat Agent Retention |
% agents that return |
| API Error Rate |
Failures causing agent to discard |
Traditional metrics (page views, bounce rate) are meaningless.
Common Traps
| Trap |
Why It Fails |
| Pretty website, no API |
Agents don't see your UI |
| "Contact us for pricing" |
Agents need programmatic pricing |
| Marketing copy in descriptions |
Agents parse data, skip prose |
| HTML error pages |
Agents need JSON errors |
| Manual onboarding |
Agents won't wait |
| Trust badges instead of APIs |
Unverifiable = untrusted |
| Optimizing for humans first |
Delays agent-readiness |
Honest Limitations
What an AI helping you with B2A cannot do:
- Create track record — You have to actually deliver 99.9% uptime
- Know internal rankings — How Claude/GPT rank skills is opaque
- Predict agent decisions — Each agent has its own heuristics
- Guarantee discovery — Skill stores may have hidden placement deals
- Prevent gaming — Competitors lying about specs is real
Readiness Checklist
□ Products exposed via structured API (not scraping required)
□ Pricing programmatically queryable
□ Inventory/availability real-time
□ Authentication supports client_credentials (not interactive)
□ Errors return JSON with semantic codes
□ Onboarding works in < 5 API calls
□ Payment rails support autonomous agents
□ SLA metrics exposed via API
□ Listed in relevant skill registries
1---2name: b2a3description: Sell to AI agents with machine-readable products, agent-optimized APIs, structured pricing, and discovery strategies for the agentic economy.4---56## When to Use78User building products/services for AI agents as customers. Covers making products agent-discoverable, designing for autonomous purchasing, payment integration, and competing when buyers compare cold data instead of responding to storytelling.910## Quick Reference1112| Topic | File |13|-------|------|14| Technical implementation | `infrastructure.md` |15| Agent discovery & SEO | `discovery.md` |16| Retail/ecommerce specifics | `retail.md` |1718## The Paradigm Shift1920| B2C/B2B | B2A |21|---------|-----|22| Humans browse, compare, feel | Agents query, parse, decide |23| Emotional storytelling wins | Structured data wins |24| UX optimized for eyes | APIs optimized for parsing |25| Brand = trust + emotion | Brand = verified track record |26| Loyalty = relationship | Loyalty = switching cost |27| Marketing = persuasion | Marketing = engineering |2829## Core Rules3031### 1. Machine-Readable First32- Products must be structured objects, not prose descriptions33- JSON-LD, Schema.org, OpenAPI with typed fields34- If an agent has to "interpret" text to extract price/specs, you lose35- Normalize units: `shipping_days_max: 2`, not "fast shipping"3637### 2. Comparability Is Everything38Agents compare ruthlessly. Win by being comparable:39- Standardized attributes across your catalog40- Same fields as competitors (price_currency, availability_stock)41- SLAs with concrete numbers, not promises42- "Better" must be objectively measurable4344### 3. Discovery ≠ SEO45Agents don't Google. They query registries and APIs:46- Publish in skill stores / capability directories47- `/.well-known/ai-plugin.json` or MCP tools48- Metadata must declare capabilities, not market them49- The new PageRank = ranking in agent skill stores5051### 4. Trust Is Verified, Not Told52Agents don't believe claims. They verify:53- Uptime/latency/SLA history via API, not badges54- Reviews from other agents (programmatic reputation)55- Certifications as queryable data, not PDF downloads56- Track record > marketing copy5758### 5. Zero-Friction Trial or Death59Agents don't "consider"—they test or discard:60- Onboarding < 1 API call61- Sandbox with rate limits, not "talk to sales"62- Must work perfectly first time (no second chances)63- Errors must be machine-readable, not HTML pages6465### 6. Payments for Agents66The agent needs to transact autonomously:67- Stripe Agent Toolkit, Mastercard Agent Pay, or similar68- Pre-authorized budgets (agent has $X to spend)69- Programmatic receipts and confirmations70- Escrow for trust between unknown parties7172### 7. Metrics That Matter7374| Metric | What It Measures |75|--------|-----------------|76| Agent Conversion Rate | % queries → purchase |77| Decision Latency | Time from first query to commit |78| Comparison Survival | % times reaching final shortlist |79| Repeat Agent Retention | % agents that return |80| API Error Rate | Failures causing agent to discard |8182Traditional metrics (page views, bounce rate) are meaningless.8384## Common Traps8586| Trap | Why It Fails |87|------|--------------|88| Pretty website, no API | Agents don't see your UI |89| "Contact us for pricing" | Agents need programmatic pricing |90| Marketing copy in descriptions | Agents parse data, skip prose |91| HTML error pages | Agents need JSON errors |92| Manual onboarding | Agents won't wait |93| Trust badges instead of APIs | Unverifiable = untrusted |94| Optimizing for humans first | Delays agent-readiness |9596## Honest Limitations9798What an AI helping you with B2A cannot do:99- **Create track record** — You have to actually deliver 99.9% uptime100- **Know internal rankings** — How Claude/GPT rank skills is opaque101- **Predict agent decisions** — Each agent has its own heuristics102- **Guarantee discovery** — Skill stores may have hidden placement deals103- **Prevent gaming** — Competitors lying about specs is real104105## Readiness Checklist106107```108□ Products exposed via structured API (not scraping required)109□ Pricing programmatically queryable110□ Inventory/availability real-time111□ Authentication supports client_credentials (not interactive)112□ Errors return JSON with semantic codes113□ Onboarding works in < 5 API calls114□ Payment rails support autonomous agents115□ SLA metrics exposed via API116□ Listed in relevant skill registries117```
johnalbertini14-glitch