DEV Community: Eduardo Destruti

Architecture and Inner Workings of AP2 (Agent Payments Protocol)

Eduardo Destruti — Tue, 20 Oct 2026 21:32:49 +0000

This technical guide provides a try of a "practical blueprint" for implementing the Agent Payments Protocol (AP2) in our applications. We will explore the JSON structure of a Payment Mandate, how an AI agent attaches it to requests, and how a merchant service verifies it cryptographically.

1. The Payment Mandate JSON Structure

Before an AI agent can execute any financial transaction, the user must issue a cryptographically signed Payment Mandate. This structure defines the strict boundaries of what the agent is allowed to spend.

{
  "mandate_id": "man_9f88c7a1e2b43",
  "holder_did": "did:key:z6MksH3...UserKey",
  "agent_did": "did:key:z6MktA9...AgentKey",
  "constraints": {
    "spending_cap_usd": 50.00,
    "current_spent_usd": 12.50,
    "currency": "USD",
    "valid_from": "2026-10-20T00:00:00Z",
    "valid_until": "2026-11-03T23:59:59Z",
    "merchant_whitelist": [
      "api.cloudcompute.io",
      "datafeed.analytics.com"
    ],
    "scope_intent": "Provision temporary GPU compute instances and fetch real-time market data."
  },
  "signature": {
    "algorithm": "Ed25519",
    "value": "3K7v8x9...cryptographic_signature_from_holder...=="
  }
}

2. Agent Execution Payload

When the autonomous agent encounters a paid resource or an API request requiring authorization, it bundles the target operation with the active Payment Mandate and signs the request payload using its own assigned private key.

{
  "request_id": "req_55410a2b",
  "timestamp": "2026-10-20T16:30:00Z",
  "action": "POST /v1/compute/provision",
  "payload": {
    "instance_type": "gpu-small",
    "duration_hours": 2
  },
  "payment_authorization": {
    "mandate": {
      "mandate_id": "man_9f88c7a1e2b43",
      "holder_did": "did:key:z6MksH3...UserKey",
      "agent_did": "did:key:z6MktA9...AgentKey",
      "constraints": {
        "spending_cap_usd": 50.00,
        "current_spent_usd": 12.50,
        "currency": "USD",
        "valid_until": "2026-11-03T23:59:59Z",
        "merchant_whitelist": ["api.cloudcompute.io"]
      },
      "signature": "3K7v8x9...holder_sig...=="
    },
    "agent_signature": "9mQ1z2x...agent_sig...=="
  }
}

3. Merchant-Side Verification Logic (Pseudocode)

When the merchant's API gateway receives the request, it executes a validation pipeline before provisioning the resource or settling the payment:

def verify_ap2_request(incoming_request, merchant_domain):
    mandate = incoming_request.payment_authorization.mandate

    # 1. Verify Holder's cryptographic signature on the mandate
    if not crypto.verify_signature(mandate.holder_did, mandate, mandate.signature):
        raise InvalidMandateException("Holder signature verification failed.")

    # 2. Check expiration time (TTL)
    if datetime.utcnow() > datetime.fromisoformat(mandate.constraints.valid_until):
        raise MandateExpiredException("The payment mandate has expired.")

    # 3. Validate Merchant Whitelist
    if merchant_domain not in mandate.constraints.merchant_whitelist:
        raise UnauthorizedMerchantException("Merchant is not whitelisted in this mandate.")

    # 4. Check Spending Caps
    requested_amount = incoming_request.get_cost()
    remaining_balance = mandate.constraints.spending_cap_usd - mandate.constraints.current_spent_usd
    if requested_amount > remaining_balance:
        raise SpendingCapExceededException("Transaction exceeds remaining mandate limit.")

    # 5. Verify Agent's signature
    if not crypto.verify_signature(mandate.agent_did, incoming_request, incoming_request.payment_authorization.agent_signature):
        raise InvalidAgentSignatureException("Agent signature verification failed.")

    return True # Proceed to payment gateway settlement

4. Summary and Best Practices

Always Rotate Agent Keys: Treat agent cryptographic keys with limited lifespans to minimize blast radius if an agent container is compromised.
Enforce Strict Whitelists: Prevent agents from dynamically expanding their merchant_whitelist without direct human authorization.
Log Every Transaction: Maintain an immutable ledger mapping mandate_id to actual settlement records for accounting and audit compliance.

Agentic Payments Are Coming: What the AP2 Protocol Means for the Future of Software

Eduardo Destruti — Tue, 20 Oct 2026 21:27:20 +0000

We are moving away from an era where AI merely answers questions and entering an era where AI takes action. But as autonomous agents begin buying services, subscribing to APIs, and handling transactions on our behalf, a massive engineering challenge arises: How do we securely authorize machines to spend money?

Enter AP2 (Agent Payments Protocol). Here is a breakdown of what this shift means, how it works, and what it changes for developers and businesses:

1. How it works today (The Bottleneck)

Right now, allowing an AI assistant or a script to execute purchases requires risky workarounds—like storing hardcoded API keys, sharing personal credit card numbers with third-party wrappers, or relying on brittle session tokens. There is no standardized, cryptographically secure way for a merchant to prove that a human actually consented to a specific transaction initiated by an autonomous agent.

2. How the AP2 Protocol works

AP2 introduces Payment Mandates—cryptographic, policy-driven consent structures. Instead of handing over your keys, you issue a strict "badge" to your AI agent. This mandate defines exact boundaries:

Maximum spending limits (e.g., up to $50)
Time windows and expiration dates
Specific merchant categories or allowed endpoints

When the agent makes a purchase, the merchant validates this secure mandate directly, ensuring total transparency and zero risk of runaway spending. Lets talk about this part later.

3. Future impacts of AP2 implementation

True Autonomous Commerce: Software won't just recommend a tool or a flight; it will negotiate, purchase, and provision it instantly.

Micro-SaaS & API Monetization: Autonomous agents will dynamically subscribe to micro-services on-the-fly, paying only for the exact compute or data they consume.

Trust & Compliance: Built-in accountability means enterprises can safely deploy agentic workflows without risking unauthorized corporate spending.

4. When does this start?

The infrastructure is being actively defined right now, driven by major collaborations between tech giants, payment networks (Visa, Mastercard, PayPal), and Web3 infrastructure providers. While early developer previews and pilot integrations are rolling out throughout 2026, mainstream production adoption is expected to scale rapidly over the next 12 to 24 months.

** What's your take? ** Are you already architecting your systems to support autonomous agents, or are you waiting for standardizations like AP2 to mature? Let’s discuss in the comments!

The AEO Measurement Pyramid: Proving Business Value in a Zero-Click Era (Part 5/5)

Eduardo Destruti — Fri, 21 Aug 2026 17:38:12 +0000

Shifting your focus from winning a click to earning an AI citation.
This is the last post about theme AEO search. I will explain clearly how AI searches change the way that we think results appears in our lifes or companies.

Shifting your analytical playbook to capture the real value of search before and after the click.

When a user gets their answer directly from an AI summary, traditional click-based metrics will often show a drop in traffic even if your visibility has actually increased. AI referrals and citations bypass conventional analytical funnels, making standard keyword tracking insufficient for modern search. To solve this, marketers must adopt a bottom-up measurement framework that captures how conversational engines actually influence buyers. This new approach reorganizes your analytics to track visibility before, instead of, and after the click occurs.

The foundation of this framework starts with directly observable user behavior, which you can easily track using free tools. By setting up custom explorations in Google Analytics, you can isolate high-intent referral traffic coming from domains like ChatGPT and Perplexity.

Meanwhile, Google Search Console helps you track branded search impressions, which climb when AI engines mention your company elsewhere. These free baseline metrics give you the most reliable on-site data to begin your optimization journey.

The middle layer focuses on interpretation-level metrics like AI visibility scores, share-of-voice, and citation analysis. Automated prompt testing tools run your customers common questions across conversational platforms to see where your brand appears. Knowing which third-party sites are most cited in your category tells you exactly where to focus your off-site authority building. Combining these patterns with your behavioral data ensures you aren't optimizing in a structural vacuum.

At the peak of the pyramid sits business impact, where you connect your search visibility directly to bottom-line pipeline outcomes. AI-referred visitors often arrive with deep context, leading to conversion rates that significantly outpace traditional organic search traffic.

By tracking these conversions and monitoring lead generation through your CRM, you can easily justify deeper investments in AEO to stakeholders. Proving true business value is what ultimately turns modern search optimization into a continuous, self-improving loop.

Authority Beyond Backlinks: Earning Web-Wide Brand Consensus for AI Citations (Part 4/5)

Eduardo Destruti — Fri, 21 Aug 2026 17:26:39 +0000

Shifting your focus from winning a click to earning an AI citation.
This is the fourth post about theme AEO search. I will explain clearly how AI searches change the way that we think results appears in our lifes or companies.

How answer engines look past traditional hyperlinks to measure your reputation across the web.

In traditional search, domain authority is largely built and measured through backlinks—reputable websites linking back to your pages. But AI answer engines evaluate authority in a far broader, more multi-dimensional way.

Rather than relying solely on clickable hyperlinks, they analyze whether your brand is consistently mentioned and referenced across the web, regardless of whether a link is present.

An answer engine scanning Reddit threads, industry blogs, news articles, and social networks is actively looking for consensus. It wants to verify if multiple independent, authoritative voices agree that your brand is a trusted expert in your space. This means off-site brand sentiment, third-party press, and consistent mentions now carry massive weight in determining whether you get cited.

Because conversational crawlers prioritize authentic human perspectives, your multi-channel presence matters more than ever. Platforms like LinkedIn signal practitioner authority, while YouTube showcases hands-on, demonstrated expertise.

For B2B brands especially, earning a mention inside a highly relevant, niche industry community or newsletter carries much stronger retrieval weight for an AI than a generic backlink.

To build this modern layer of authority, you must shift your focus from a pure link-building checklist to a holistic brand reputation strategy. Encourage honest customer reviews, contribute expert commentary to industry publications, and foster active community discussions.

The more consistently your brand's expertise is validated across independent off-site channels, the more confident AI engines become in citing you.

Decoding "Query Fan-Out": Structuring Topic Clusters for Conversational AI Search (Part 3/5)

Eduardo Destruti — Fri, 21 Aug 2026 17:11:29 +0000

Shifting your focus from winning a click to earning an AI citation.
This is the third post about theme AEO search. I will explain clearly how AI searches change the way that we think results appears in our lifes or companies.

How AI engines deconstruct a single query, and why topical depth is your key to capturing citations.

When a user asks an AI tool a broad question, the engine rarely searches for a single, perfect article to answer it. Instead, it runs a process called query fan-out, automatically deconstructing that lone query into a handful of hyper-specific sub-questions. The engine then retrieves independent answers for each sub-question from different corners of the web before synthesizing them into one cohesive response.

To win direct citations in this environment, your company or brand must demonstrate comprehensive topical depth rather than surface-level coverage. Conversational engines do not evaluate pages in a vacuum; they actively assess whether your website thoroughly maps out a subject area. Building robust topic clusters ensures you have highly specific pages ready to match these deconstructed, fanned-out queries.

You can execute this strategy at the outline stage by mapping out your audience's entire learning journey from awareness to action. Pair a foundational, in-depth guide with highly specific, targeted subtopics—such as checklists, comparisons, or common mistakes. This structured cluster approach signals complete authority, giving the AI multiple, clear extraction points to choose from across your site.

Also, FAQ-style sections are incredibly powerful for this because every standalone Q&A block is a potential match for a fanned-out sub-query. By answering natural follow-up questions directly on your page, you can capture multiple citations from a single user session.

Stop writing isolated, disconnected content and start building the deeply connected topic ecosystems that AI engines crave.

Writing for the Dual-Audience: The "Answer-First" Structure That AI and Humans Both Love (Part 2/5)

Eduardo Destruti — Fri, 21 Aug 2026 16:56:11 +0000

Shifting your focus from winning a click to earning an AI citation.
This is the second post about theme AEO search. I will explain clearly how AI searches change the way that we think results appears in our lifes or companies.

Traditional web content is often written like a mystery novel, slowly building up to the main point near the bottom of the page. But modern search behavior demands that we write for a dual-audience of humans looking for quick value and AI crawlers scanning for easy data extraction. To satisfy both, your pages must be deliberately engineered to deliver standalone value at the individual passage level.

The secret lies in the Answer-First Principle: always lead with a clear, direct opening statement of 40 to 60 words. Providing immediate value builds trust upfront, making readers far more likely to stick around for the supporting context, details, and examples below the fold.

Because answer engines retrieve information in isolated blocks, each major section on your website must be completely independent. Avoid using transitional crutches like "as mentioned above" or "building on the previous point". Instead, structure your sections under descriptive, question-based headers that map directly to the conversational queries users ask AI systems.

Finally, format your supporting data into scannable lists, standalone definitions, and side-by-side comparison tables. These structured layouts act as perfect extraction markers, lowering the cognitive load for human eyes and reducing the algorithmic guesswork for crawlers. Before you publish your next piece of content, ensure every single section can stand entirely on its own.

SEO is No Longer Enough: Welcome to the Age of Answer Engine Optimization (AEO) (Part 1/5)

Eduardo Destruti — Fri, 21 Aug 2026 15:13:12 +0000

Shifting your focus from winning a click to earning an AI citation.
This is the first post about theme AEO search. I will explain clearly how AI searches change the way that we think results appears in our lifes or companies.

Search is undergoing its most massive evolution since the 1990s as traditional blue links give way to AI-powered answers. Today, a rapidly growing share of user queries end right on the results page without a single click. Marketers who rely solely on ranking for website traffic risk being left behind in this zero-click landscape.

This is where Answer Engine Optimization (AEO) becomes your brand's ultimate competitive advantage. While traditional SEO focuses on driving website traffic, AEO shifts the goal to brand visibility inside synthesized AI answers. It’s no longer just about where you rank, but whether ChatGPT, Gemini, or Perplexity actively trust and cite you.

Answer engines do not simply return a directory list of links for users to browse and compare. Instead, they evaluate web-wide consensus, authority, and content structure to generate a single, direct response. To win in this environment, all the companies must format your expert content for seamless AI extraction and citation.

Fortunately, SEO and AEO are not competing priorities, but highly complementary expressions of the same E-E-A-T principles. Investing in high-quality, technically sound, and structured content prepares you to succeed across both search environments. The future of search isn't coming—it is already here, and it's time to build your foundation.