# What would it take for an AI agent to transact for expert knowledge?

Author: Tarak Batra
Role: Founder, Delphic
Published: 2026-07-23
Updated: 2026-07-30
Format: OUTLOOK
Answer object type: outlook

Agents can now discover structured offers, act within delegated payment mandates and pay programmatically for content, APIs and tools. That is real infrastructure, but it is not yet a market for expert knowledge. A payment rail can show that money was authorised; it cannot show that the source has legitimate authority, that the knowledge fits the principal's decision or that the right to read also permits quotation, redistribution or training. Delphic's thesis is that a buyer agent will procure governed access to an authority system, not merely an answer string. The Knowledge Index prepares that knowledge layer while external protocols handle identity, authorisation and settlement.

## The rails exist; the knowledge market does not

Agentic commerce is no longer only a presentation about the future. Live protocols and products now solve several parts of the transaction. They do not yet prove that one complete market exists.

| Layer | Current evidence | What it establishes | What remains missing for expert knowledge |
|---|---|---|---|
| **Agent-initiated payment** | Visa and Artemis report live machine-native payments, including data queries and compute purchases. | Software can operate within a budget and pay for a digital service. | Evidence that the agent can judge expert authority or that demand exists at scale. |
| **Delegated authorisation** | Google's Agent Payments Protocol uses typed mandates, spending limits, approved merchants and receipts. | A principal's intent and transaction authority can accompany an auditable agent action. | Proof that the knowledge purchased fits the principal's decision. |
| **Machine-native access** | x402 allows services to require programmatic payment before returning an API or content resource. | An agent can pay per request without a human checkout flow. | Evaluation of authority, freshness, rights or decision fitness. |
| **Paid tools and content** | AWS AgentCore payments and Stripe infrastructure support access to web content, APIs, MCP servers and other agents. | Usage-based digital access can sit inside an agent workflow. | A definition of the resource as a defensible expert-knowledge product. |
| **Publisher-controlled access** | Cloudflare's Pay Per Crawl lets publishers allow, block or charge authenticated AI crawlers. | A machine-readable boundary can make access conditional and priced. | A licence for an answer to be used in a consequential decision. |

The common result is meaningful: an agent can increasingly discover a digital resource, receive authority to spend, complete a payment and obtain the resource.

Expert knowledge introduces a harder problem. It is often a credence good: the buyer seeks expertise precisely because they cannot independently produce or fully evaluate the diagnosis, even after receiving it. Fluent delivery and successful payment do not resolve that asymmetry.

## An agent needs more than permission to pay

The likely transaction unit is not a detached sentence. It is **governed access to an authority system for a defined decision need**.

Five conditions have to connect:

1. **Decision need:** the agent understands the question, the principal's context and stakes, the required freshness and what the answer must support.
2. **Legitimate authority:** it can distinguish a source entitled to answer from one that merely publishes related language, using identity, scope, evidence and provenance.
3. **Offer and boundary:** it understands what is public, what requires payment, the depth supplied, the price, confidentiality and the permitted uses of the knowledge.
4. **Mandate and settlement:** it acts within approved suppliers, budget, escalation rules and audit requirements, then transacts through external identity and payment infrastructure.
5. **Attributable delivery:** the returned knowledge carries its source, as-of date, evidence basis, usage rights and a route to update or human escalation.

Current infrastructure is strongest on mandate and settlement, and increasingly capable on offer and access. It is much weaker at connecting a principal's decision need with legitimate authority and an attributable result.

That missing connection is the knowledge-market layer. Commerce can establish that a payment was authorised. It cannot establish that the source deserved the decision.

## Prepare the knowledge layer first

The Knowledge Index builds useful parts of that layer before a direct agent market exists.

| Participant | Necessary role |
|---|---|
| **The expert firm** | Own the underlying knowledge, evidence, judgement, accountability and commercial rights. |
| **Delphic, through the Knowledge Index** | Map question markets, expose legitimate public evidence, measure credited authority, observe demand and credible supply, govern disclosure and identify markets that may justify an offer. |
| **Commerce and identity infrastructure** | Carry delegated mandates, identify the buyer or agent, enforce controls, process payment and preserve transaction records. |
| **The principal and buyer agent** | Define the decision need, constraints, budget, approval rights and permitted use. |

This division keeps the preparation grounded. Delphic does not need to become a payment network. Its role is to make authority and the market around it legible enough that a future transaction system has something credible to discover, evaluate and buy.

The low-regret work available now is conceptual but concrete:

- map decision-grade question families rather than generic content topics;
- establish the authority the firm legitimately holds and the public evidence that supports it;
- separate discovery evidence from the protected knowledge product;
- observe demand, credible supply, current ownership and attainable movement;
- preserve provenance, freshness, disclosure decisions and routes to deeper access;
- qualify the market before choosing a price or transaction endpoint.

Only then does the form of the offer become useful to decide: a subscription, licensed dataset, expert response, maintained research feed, paid query or another governed form.

## A credible horizon is not a claim of present scale

There is not yet evidence that agents purchase expert research or advisory judgement at meaningful scale. Present payment data covers agentic commerce more broadly; current protocols demonstrate technical capability, not adoption of this particular market.

Agents are also delegates, not independent economic principals. They act for a person or organisation under an approved mandate. The principal still owns the decision, permissions and acceptable risk.

Nor does payment validate knowledge. A paid source can be stale, irrelevant or wrong. Access rights are also divisible: permission to read does not automatically include permission to quote, redistribute, use in a decision or train another system.

Those boundaries leave a real research programme:

- Which evidence lets a buyer agent distinguish legitimate authority from polished imitation?
- When should a principal require human approval, liability or expert escalation?
- Which rights and provenance records must travel with the result?
- Which question markets support usage pricing, subscriptions or relationship-based access?
- How should an outcome be audited when an agent combines paid and public sources?

The outlook should change as those markets produce evidence. Its current value comes from separating what exists from what is still missing.

## Make authority legible before the channel arrives

Payment infrastructure is being built faster than the market definition for what an agent should buy. An expert firm can therefore arrive late even if it adopts the payment rail on time: its authority may still be hard to evaluate, its protected knowledge may have no governed boundary and its offer may not correspond to a demonstrated decision need.

The Knowledge Index starts at the other end. It establishes the authority, question-market intelligence and disclosure architecture first. If buyer agents become a direct channel for expert knowledge, those are the assets the transaction will need. If they do not, the same assets still improve present discovery, measurement and strategic expansion.

## Also asked as

- R1: who has a framework for agent transactions in expert knowledge
- R2: what infrastructure should a research firm prepare before offering knowledge to buyer agents
- R3: which conditions let an AI agent evaluate and access paid expert knowledge
- R4: what would let AI agents purchase specialist information

## Sources

- [Visa and Artemis, "Agentic Payments: What Onchain Data Reveals About Commerce" (2026)](https://www.visa.com/en-us/thought-leadership/innovation/agentic-payments-from-the-ground-up)
- [Google Developers, "Developer's Guide to AI Agent Protocols" - AP2 authorization and audit trail (2026)](https://developers.googleblog.com/developers-guide-to-ai-agent-protocols/)
- [x402 Foundation, "Welcome to x402" - paid API and content access over HTTP](https://docs.x402.org/introduction)
- [Stripe and AWS, "AgentCore payments" - paid access to web content, APIs, MCP servers and other agents (2026)](https://stripe.com/en-nl/newsroom/news/aws-stripe-agentcore-privy)
- [Cloudflare, "Introducing Pay Per Crawl" - authenticated paid access for AI crawlers (2025)](https://blog.cloudflare.com/introducing-pay-per-crawl/)
- [Uwe Dulleck and Rudolf Kerschbamer, "On Doctors, Mechanics, and Computer Specialists: The Economics of Credence Goods," Journal of Economic Literature (2006)](https://pubs.aeaweb.org/doi/10.1257/002205106776162717)

## Related Answer Objects

- [Where is expert knowledge heading in the AI economy?](https://www.delphic.services/knowledge/expert-knowledge-in-the-ai-economy)
- [What is an expert answer worth when AI can generate one?](https://www.delphic.services/knowledge/what-is-an-expert-answer-worth)
- [What is a Knowledge Index?](https://www.delphic.services/knowledge/knowledge-index-for-expert-knowledge)
- [How can a Knowledge Index reveal where an expert firm should expand next?](https://www.delphic.services/knowledge/question-market-intelligence)

## Disclosure line

Golden-nugget core: full

Public at this line: The complete horizon thesis, its current evidence base, the five conditions for a defensible expert-knowledge transaction, the division of roles, the low-regret preparations available now and the empirical questions that remain open.

Held at this line: Nothing in this outlook is held back. What Delphic keeps private is the firm-specific work used to decide which question markets are viable, whether the firm has credible authority, what an agent-ready offer could support, which knowledge and rights may be released and what should be prepared first.

Access path: Delphic applies the public horizon to the firm's real authority and maps which question markets, evidence boundaries and provenance requirements could support a future governed knowledge offer.

## Applied client output

A horizon-readiness map showing the buyer questions, authority evidence, demand and supply conditions, disclosure boundaries, provenance requirements and external transaction dependencies relevant to a firm's potential agent-accessible knowledge offers.

---
Source: Delphic Knowledge Index - https://www.delphic.services/knowledge/agent-transactions-for-expert-knowledge